A-Share · Pre-Market
A-Share Pre-Market Brief | 2026-07-31 Friday
Machine-translated from the Chinese original. In case of any discrepancy, the Chinese version prevails.
Single-name entries 30
Show 18 more
Scores are a subjective ordinal scale; gaps within a band carry no ranking meaning
⚠️ Risk warning (up front): This list is pre-market information triage and observation only. It does not constitute investment advice and contains no ratings, target prices, position sizing, or buy/sell recommendations. The labels "priority deep-dive / watch closely / watch only / Pass" used below are a ranking of research attention only and imply no judgment about future returns.
Data-scope disclosure first (please read this section before anything else)
- Time window: 2026-07-30 15:00 (A-share close) → 2026-07-31 07:00 (this report's cut-off). Today is Friday and the last trading day of July.
- The single most important scope fact this issue: overnight U.S. equities (the U.S. Eastern 7/30 session, 7/30 21:30–7/31 04:00 Beijing time) are this report's core evidence base, and their direction is the exact opposite of yesterday's A-share session. All U.S. equity data are stamped in Beijing time.
- Market-data sources: Sina full-market quotes
stock_zh_a_spot(5533 names, timestamp 15:30:01, i.e. the 7/30 close), East Money limit-up/limit-down/failed-limit pools, East Money block-trade disclosure detailstock_lhb_detail_em, Tonghuashun industry-sector summarystock_board_industry_summary_ths, Sina single-stock daily barsstock_zh_a_daily(forward-adjusted).- ⚠️ Interface gap (must be disclosed): East Money's historical quote interface
stock_zh_a_histreturnedConnectionErrorrepeatedly, so all single-stock historical interval returns in this report have been switched to the Sina daily-bar basis, which is not the same data source as the East Money basis cited in yesterday's recap; the Tonghuashun industry basis and East Money's industry classification are not directly comparable across days — this is the second consecutive day with this scope discontinuity.- ✅ The debt registered in yesterday's recap has been repaid: item 6 of section 6.4 of yesterday's recap registered "7/30 block-trade disclosure not obtained, must be back-filled tomorrow morning". This report has obtained all 119 block-trade disclosure records for 7/30 via
stock_lhb_detail_em, and the net buy/sell direction for 德明利 (Demingli, 001309), 江淮汽车 (JAC Motors, 600418), 紫光股份 (Unisplendour, 000938) and 长电科技 (JCET, 600584) directly rewrote this report's characterization of them (see item 20 of section 1 and section 5). Debt cleared.- Dimensions this report still does not cover: northbound net flows (the exchanges discontinued single-day net-buy disclosure from August 2024, with no substitute basis), FTSE China A50 futures night-session close (repeated searches failed to obtain reliable point-in-time data, so we write "no reliable data available" and do not speculate), and the complete 7/31 Shanghai/Shenzhen trading notices (a complete new-issue/lock-up-expiry list was not obtained).
- Two items for which only a "media-aggregated basis" was obtained, with no original company filing: 恒瑞医药 (Hengrui Pharma)'s approval of insulin degludec-type product, and 众合科技 (Unittec)'s RMB 243 million Shenyang metro contract win. Both have been flagged as down-weighted in section 1; the company filings govern.
- ⚠️ The most important scope issue found and corrected in this round of QC — overnight news must be split into two classes, "already priced by the 7/30 close" and "newly added after the 7/30 close":
- Class A (already public before the A-share 7/30 open, and already "vetoed" once by the 7/30 decline): Microsoft FY26Q4 results (U.S. Eastern 7/29 after-hours = about 05:00 Beijing time on 7/30), Meta results (same), Samsung Electronics' record Q2 results (pre-market in Korea on 7/30), the Fed's 7/29 rate decision, SK Hynix's Q2 miss (7/29).
- Class B (newly added after the U.S. Eastern 7/30 close, not yet priced by A-shares): Amazon's results and its capex raise to $220 billion, Apple's FQ3 results and soft guidance.
- The hard constraint that follows: SOXX +8.5%, Microsoft +15.5%, ASE +11.79% and the memory-stock surge during the 7/30 U.S. session all happened before Amazon's results, and cannot be explained by Amazon's capex. Sections 0, 1 and 2 of this report have been rewritten accordingly.
- ⚠️ Three pieces of major counter-evidence added (missing from the first draft, added after
risk-auditorQC): the 7/29 Fed decision and the U.S. equity slump, SK Hynix's Q2 miss and the Korean market plunge, and the fact that Samsung Electronics' record results were already public before the 7/30 A-share open while A-share semis still fell −7.08%. These three directly weaken the "180-degree overnight reversal" characterization; see section 0 and items 2/3/3b/26 of section 1.
0. Today in one sentence
Today is the day "a strong external repair" collides with "an internal trend that has already been falling for a month", and the tension between those two things is the core contradiction behind all of today's trading. Overnight U.S. equities (closed 04:00 Beijing time this morning): Nasdaq +2.78% at 25122.18, Dow +1.19%, S&P +1.66%; Microsoft +15.5% (about $450 billion of market cap added in a single day), the iShares Semiconductor ETF +8.5%, memory names surging as a group (SanDisk +22%, Micron +15%, Western Digital +15%, Seagate +12%, Silicon Motion +21.66%), TSMC +7.63%, ASE +11.79% — while yesterday's A-share session was precisely a freezing point: semiconductors −7.08%, 74 limit-downs > 52 limit-ups.
⚠️ But the word "reversal" must be qualified first, or the entire premise of this report is wrong: this is not a reversal from zero, it is the recovery of the previous session's hawkish shock. On U.S. Eastern 7/29 the Fed decision (held at 3.50%–3.75%, the fifth hold of the year, with 3 of 12 voters favoring a 25bp hike), Chair Warsh sharply reducing the use of forward guidance, and the CME watch tool showing a 78.8% probability of a September hike, combined with the U.S.–Iran situation pushing oil higher, produced a sharp U.S. selloff that day: Nasdaq −1.74% at 24442.94, S&P −1.52%, Dow −2.19%. Working backwards from this report's own closing numbers: 25122.18 − 24442.94 = 679.24 points, so 7/30's +2.78% merely recovered the 7/29 decline; the two-day net gain is about +1.0%, and the Dow at 52208.06 is still below 7/28's 52747.32 and has not recovered. More important is the direction: Warsh explicitly listed AI capex as a policy-assessment variable — meaning the stronger capex is, the greater the pressure to hike, which is the most direct macro counter-variable to this report's S-grade thesis.
There is only one genuinely "not yet priced" strongest catalyst: after the U.S. Eastern close on 7/30, Amazon raised its 2026 capital expenditure from $200 billion to $220 billion, with CNBC reporting the increase explicitly attributed to "rising memory costs"; AWS revenue +37% and undelivered backlog reaching $496 billion (versus $364 billion in the prior quarter); CEO Jassy said "even spending that much, we still won't be able to meet all demand in 2026, and the same goes for 2027". This is the first time "memory price increases" have been endorsed by the financial guidance of the largest downstream buyer rather than by a supplier's price-increase letter — it is the logically hardest item in this news cycle, and the only reason this report ranks "memory pricing" rather than "optical modules" as the number-one branch.
⚠️ Three things must be stated together on sequencing, or the causality is inverted: ① The 7/30 intraday surge in U.S. memory stocks (snapshot at 23:27 Beijing time) happened roughly 5 hours before Amazon's results; its drivers were Microsoft's results (U.S. Eastern 7/29 after-hours, Azure +43%, commercial RPO +84%) and Samsung Electronics' record Q2 — not Amazon; ② Microsoft's +15.5% is the price realization of "7/29 after-hours results", and the Azure +43% figure was already public before the A-share open on 7/30 — that is, A-shares fell yesterday with the Microsoft data already known; ③ Samsung Electronics disclosed Q2 before the Korean open on 7/30: revenue 171.50 trillion KRW +130%, operating profit 89.49 trillion KRW +1813.83%, net profit 71.27 trillion KRW +1344.46%, all record highs; the DS (memory and foundry) division's operating profit of 89.2 trillion KRW was a record high, with single-quarter earnings exceeding Nvidia's for the global top spot, and guidance pointing to strong second-half server-chip demand and a continuing supply shortage. This is harder evidence than Amazon's capex attribution (it is the manufacturer's own volume-and-price data), and it was already visible at the 09:30 A-share open on 7/30, on a day when A-share semiconductors still fell −7.08% with 74 limit-downs. Conclusion: the marginal impact of external supply-side evidence on this A-share chain was empirically measured as "weak" yesterday. The only genuinely new information today is Amazon's capex and Apple's guidance.
Strongest branches: memory price increases (S grade) > AI hardware oversold rebound driven by upward revisions to overseas cloud capex (A+ grade) > first full trading day after the Politburo meeting (A grade) > photovoltaic anti-involution (meeting held in Yancheng today, A grade). The likely direction of flows is "rotation out of yesterday's safe-haven assets and back into technology", i.e. yesterday's high-to-low rotation is likely to be partly reversed today. The driver type is overseas industry + earnings + price stacked together, with policy (the Politburo meeting) as the second layer.
But the pre-market state assessment must include five cold statements, or this list becomes a chase-the-high list:
- The A-share AI hardware chain did not "start falling yesterday" — it has been falling for a full month: 德明利 (Demingli) fell from its 6/26 intraday high of RMB 980.00 to its 7/29 intraday low of RMB 334.00 (−65.9%, with 7 limit-downs across 12 days), 兆易创新 (GigaDevice) from its 6/29 closing high of RMB 840.00 to RMB 371.10 on 7/30 (−55.8%), 太极实业 (Taiji Industry) from 32.10 on 7/1 to 15.66 (−51.2%), 永鼎股份 (Jiangsu Yongding) from 62.5 to 33.0 (−47.2%), 沪电股份 (WUS Printed Circuit) from its 7/1 intraday high of 158.20 to 96.55 (−39.0%). What is most likely to happen today is "a high-open rebound within a downtrend", not "a trend reversal", and the two call for completely different responses.
- The two strongest limit-up names yesterday both showed massive net selling on the block-trade disclosure: 德明利 (Demingli) hit the limit-up but showed net selling of RMB 706 million (buy RMB 1.249 billion / sell RMB 1.955 billion), 江淮汽车 (JAC Motors) hit the limit-up but showed net selling of RMB 393 million; while limit-down 紫光股份 (Unisplendour) showed net buying of +RMB 548 million and limit-down 长电科技 (JCET) net buying of +RMB 317 million. The direction is inverted — and this is precisely the item that item 4 of section 6.4 of yesterday's recap required be force-checked; today it delivered an answer opposite to price.
- The public attribution for yesterday's baijiu surge was "safe-haven money rotating back in as technology receded + valuation repair" (liquor analyst Cai Xuefei: it is essentially valuation repair rather than a fundamental turn), and today the external premise of a technology retreat has been thoroughly overturned. So day 3 for baijiu is not "continuation" — it must first prove it can hold up even while technology rebounds, and this is the item most easily misjudged today.
- ⚠️ The strongest logic — memory price increases — itself carries the strongest counter-evidence, and the two must be written side by side: TrendForce estimates 2026Q3 DRAM contract prices up only +13%
+18% QoQ and NAND +10%+15%, versus roughly +60% in 2026Q2 (conventional DRAM +58%~63%, NAND Flash +70%~75%) — the slope of the price increase is converging sharply. For memory manufacturers (Micron, SK Hynix) a high price level is profit; but for module makers (德明利/Demingli, 江波龙/Longsys, 佰维/Biwin) gross margin depends on the rate of change of price rather than its level: Demingli's historical single-quarter gross margin range is −1.89% (2023Q2) → 37.23% (2024Q1) → 1.32% (2024Q4) → 57.42% (2026Q1). Backing out Demingli's own 7/15 interim pre-announcement, 2026Q2 single-quarter net profit attributable to parent is already −5.7%~−29.7% QoQ, with net margin falling from 44.4% to 27.8%~30.1% — the profit peak may already have occurred in 2026Q1. Without this paragraph, this list would be dishonest. - July's memory-chain collapse has one frequently overlooked direct cause: 长鑫科技 (CXMT, 688825) listed on the STAR Market on 7/27, opening at RMB 49.50, up 471.59%, with a total market cap of about RMB 3.3 trillion, overtaking ICBC as the largest A-share by market cap, with total funds raised of RMB 57.919 billion (RMB 66.607 billion with the over-allotment fully exercised; the RMB 29.5 billion planned raise is the prospectus basis, and the two must be distinguished), the largest IPO in STAR Market history. The mapping premium of "CXMT shadow stocks" such as GigaDevice was dismantled in one stroke (GigaDevice limit-down on 7/28, another −6.81% on 7/29). This means today's rebound in the memory chain must simultaneously fight the structural change that "the shadow-stock logic has been falsified by the real thing" — and that will not disappear just because Micron rose 15% overnight.
1. News overview
Impact grades: S = changes an industry trend or directly drives order elasticity; A = clearly positive for one branch with multiple beneficiaries; B = positive but with a long chain requiring verification; C = sentiment stimulus with weak persistence. All times are Beijing time. Anything whose "publication time cannot be confirmed" or for which only a second-hand aggregated basis was obtained is flagged in the "Type" column and down-weighted.
| No. | Publication time | Source | Headline / core content | Type | Branch involved | Impact grade | Source link |
|---|---|---|---|---|---|---|---|
| 1 | 07-31 early morning (U.S. Eastern 7/30 after-hours) | CNBC | Amazon raises 2026 capital expenditure from $200 billion to $220 billion, explicitly attributed to "rising memory costs"; Q2 revenue $200.606 billion +20%, AWS revenue $42.232 billion +37% (fastest in 18 quarters), backlog $496 billion (prior quarter $364 billion); Jassy: "even so, we won't be able to meet all demand in 2026, and the same goes for 2027" | Overseas bellwether earnings | Memory pricing, AI compute, optical modules, PCB, packaging & testing | S | CNBC |
| 2 | 07-31 04:00 (U.S. close) | Sina Finance | All three U.S. indices surge together: Nasdaq +2.78% at 25122.18 (up 679.24 points), S&P +1.66% at 7437.63, Dow +1.19% at 52208.06; Microsoft +15.5%, its largest single-day gain in 18 years (Azure-driven); iShares Semiconductor ETF (SOXX) +8.5%; Meta −8% on weak guidance and a plunge in free cash flow | Overseas markets | Full AI compute chain, semiconductors | S | Sina Finance |
| 3 | 07-30 23:27 (U.S. intraday, about 5 hours before Amazon's results) | Sina Finance | The Philadelphia Semiconductor Index rose over 8% intraday, with memory majors surging across the board: SanDisk +22%, Micron +15%, Western Digital +15%, Seagate +12%; ⚠️ another frequently cited figure, "SK Hynix +15%", must be qualified: that quote is for its U.S. OTC-listed ADR, and its starting point was a post-listing low set on 7/29 after its Q2 miss (operating profit 60.54 trillion KRW < consensus of about 64 trillion, revenue 79.32 trillion < consensus of about 84 trillion); SK Hynix closed −0.79% on the Korean main board on 7/30, and the two venues cannot be mixed | Overseas industry | Memory pricing | A (downgraded from S, because the driver was Microsoft/Samsung earnings rather than Amazon, and SK Hynix's starting point was an earnings miss) | Sina Finance, Securities Times · SK Hynix earnings miss |
| 3b | 07-30 pre-market in Korea (already public before the A-share open) | Sina Tech, Securities Times | ⭐Samsung Electronics 2026Q2 at a record high: revenue 171.50 trillion KRW +130%, operating profit 89.49 trillion KRW +1813.83%, net profit 71.27 trillion KRW +1344.46%; the DS (memory + foundry) division's operating profit of 89.2 trillion KRW was a record high, with single-quarter earnings exceeding Nvidia's for the global top spot; guidance points to strong second-half server-chip demand and a continuing supply shortage | Overseas bellwether earnings | Memory pricing (hardest supply-side evidence) | S (but ⚠️ it was already public before the A-share open on 7/30, and A-share semiconductors still fell −7.08% that day — the most direct counter-evidence on whether external good news can drive A-shares) | Sina Tech, Securities Times |
| 3c | 07-30 05:00 (U.S. Eastern 7/29 after-hours, already public before the A-share 7/30 open) | Microsoft website, Sina Finance | Microsoft FY26Q4: revenue about $90 billion +18%, Azure +43% (fastest in four years), commercial RPO +84%, Azure annual revenue topping $100 billion for the first time; but it also cut its FY2027 capex expectation from $190 billion to $175 billion (capex discipline, itself part of the reason for the share-price surge); +8% after-hours, and the common stock +15.5% on 7/30 | Overseas bellwether earnings | AI compute | A (already priced once by A-shares on 7/30; and "cutting FY27 capex" runs opposite to Amazon "raising 2026 capex", so the two must be shown side by side) | Microsoft website, Sina Finance |
| 4 | 07-31 04:26 | Sina Finance | China ADRs and Asian semiconductors strong across the board: Nasdaq Golden Dragon China Index +1.05%; TSMC +7.63%, ASE Technology +11.79%, UMC +10.87%, Silicon Motion +21.66%, GDS +8.08%; Li Auto −2.25%, PDD −1.00% | Overseas industry | Packaging & testing, memory controllers, wafer foundry, IDC | A | Sina Finance |
| 5 | 07-30 14:36 press report / Xinhua filed 23:46 | Xinhua, Cailianshe | Politburo meeting maps out second-half economic work: macro policy to "keep exerting force and add force at the right time", "step up counter-cyclical adjustment, work harder to expand domestic demand and optimize supply", "solidly advance the planning and construction of the 'six networks'", "deeply implement the 'AI+' initiative", "stabilize the property market", "properly implement the package debt-resolution plan", "solidly advance reform and risk resolution at local small and medium financial institutions", "deepen comprehensive reform of capital-market investment and financing, and enhance the resilience and confidence of the capital market", "continue the comprehensive rectification of 'involutionary' competition" | Policy | Compute network/AI, domestic demand and consumption, property, brokers, banks, anti-involution | S (but only 22 minutes of trading remained after the press report, so today is the first full pricing day) | Xinhua, Cailianshe |
| 6 | 07-31 early morning (U.S. Eastern 7/30 after-hours) | 9to5Mac / MacRumors / CNBC | Apple FQ3 results: revenue $109.4 billion +16%, net profit $29.8 billion, EPS $2.02, all above consensus; but supply constraints led to soft revenue guidance for the next quarter, and the shares fell about 6% after hours at one point | Overseas bellwether earnings | Apple chain (negative), memory pricing (indirect corroboration) | A | MacRumors, CNBC |
| 7 | 07-31 (held today) | National Business Daily, Securities Times | The Price Supervision and Competition Bureau of the State Administration for Market Regulation is holding a "photovoltaic industry price-compliance guidance session" in Yancheng, Jiangsu today (7/31), with the China Photovoltaic Industry Association and PV companies attending; the aim is to implement the General Rules for the PV Industry Cost Accounting Model (formally effective from 7/27), curbing below-cost sales and malicious low-price bidding | Policy / anti-involution | Photovoltaics | A | NBD, Securities Times |
| 8 | 07-29 19:47 | Securities Times, Jrj.com | 永鼎股份 (Jiangsu Yongding, 600105): controlled subsidiary Suzhou Dingxin received high-power laser chip purchase orders from customers A and B totaling about RMB 1.133 billion over the past month (RMB 542 million including tax + $87 million including tax), with performance periods to 2027-12-10 / 2028-01-03 | Orders | Optical chips | A (⚠️ a 7/29 item; the stock already rose +7.33% on 7/30, so this is second-round fermentation, not brand-new good news) | Securities Times, Jrj.com |
| 9 | 07-30 20:42 | Jrj.com, TMTPost | 中际旭创 (Innolight, 300308): the controlling shareholder and parties acting in concert cut their stake from 17.99% to 16.53% (down 1.46%), selling 6.21 million A shares over the period (equity-change period 2025-05-28 to 2026-07-30, including H-share listing and equity-incentive exercise factors) | Stake reduction (negative) | Optical modules / CPO | B (negative) | Jrj.com |
| 10 | 07-29 evening / 07-30 morning report | Sina Finance | 兆易创新 (GigaDevice, 603986): Chairman Zhu Yiming proposed a buyback of RMB 1–2 billion of A shares, all for cancellation, pledged not to sell for 12 months, and plans to increase his stake by no less than RMB 1 billion. ⚠️ Controversy disclosed alongside: between 2026-05-06 and 06-12 Zhu Yiming sold about 11.11 million shares via centralized bidding and block trades, cashing out about RMB 4.4 billion | Buyback / governance | Memory | B+ | Sina Finance, Sina Finance · critical piece |
| 11 | 07-29 late night | Jrj.com, Sina Tech | 长江存储 (YMTC) issues a statement rebutting rumors: the circulating claim that "the USPTO ruled on 7/28 that all 19 claims of its core 3D NAND patent are invalid" is seriously misleading, and "YMTC concept stocks collapsed across the board" is maliciously fabricated; its patent litigation against Micron (27 patents, 295 claims) is still under adjudication | Clarification | Memory | B (it removes a negative, it does not add a positive) | Jrj.com |
| 12 | 07-28 / 07-29 | ICSmart, Tencent News | Samsung Electro-Mechanics raises MLCC prices about 30% across the board, effective 2026-08-01, and applies the new prices retroactively to placed-but-undelivered orders; Taiyo Yuden plans to follow on 9/1 | Price | Passive components / MLCC | B+ (⚠️ the A-share components sector fell −6.58% on 7/30 with 4 up and 58 down; the market has already rejected this once) | ICSmart, Tencent News |
| 13 | 07-30 evening | Yicai, Securities Times | 宇树科技 (Unitree)'s STAR Market IPO timetable is set: preliminary book-building 8/5, online and offline subscription 8/10, payment 8/12; issuing 40.446434 million shares for 10% of post-issue capital, raising RMB 4.202 billion; 2023–2025 revenue RMB 159 / 393 / 1,699 million, net profit RMB −11 / 95 / 278 million | Event | Humanoid robots | A | Securities Times, Yicai |
| 14 | 07-30 evening | Yicai | 赤峰黄金 (Chifeng Jilong Gold): gold metal content at its Sepon project in Laos raised from 107 tonnes to 260 tonnes, up 143% | Resource volume | Gold | A (single-stock level) | Yicai |
| 15 | 07-30 evening | Media-aggregated basis | 恒瑞医药 (Hengrui Pharma): its insulin degludec injection was approved for marketing for adult type 2 diabetes, the first domestically developed long-acting insulin analogue in China | Approval | Innovative drugs | A (⚠️ only a media-aggregated basis was obtained, with no original company filing; down-weighted) | Sina Finance search page |
| 16 | 07-30 evening | Yicai | Wave of buybacks / stake increases: 三环集团 (CCTC) plans a buyback of RMB 0.5–1.0 billion, 网宿科技 (Wangsu) RMB 300–600 million for cancellation, 东阳光 (HEC)'s controlling shareholder proposed a buyback of RMB 300–600 million together with a stake increase of RMB 300–600 million, 乐鑫科技 (Espressif) RMB 50–100 million, 甬矽电子 (Forehope) RMB 100–150 million, 中研股份 (Zhongyan) RMB 50–100 million | Buybacks | Whole-market sentiment | B | Yicai |
| 17 | 07-30 evening | Yicai | 容百科技 (Ronbay) invests RMB 4.723 billion in a Xiantao sodium-battery cathode materials project; 有研新材 (Grinm Advanced Materials) invests RMB 486 million in phase II of sputtering targets; 索菱股份 (Soling)'s subsidiary signs a $130 million (about RMB 883 million) electronic equipment supply contract | Investment / orders | Sodium batteries, sputtering targets, electronics | B | Yicai |
| 18 | 07-30 evening | Media-aggregated basis | 行云科技 (Xingyun Technology)'s supplementary agreement raises the compute-service contract value to RMB 3.053 billion (+201.14% versus the original agreement); 亿田智能 (Yitian Intelligent)'s subsidiary Gansu Yisuan signed an RMB 1.106 billion compute-resource service contract; 众合科技 (Unittec) jointly won an RMB 243 million Shenyang metro signaling-system upgrade | Orders | Compute leasing, rail transit | B (⚠️ all obtained on a media-aggregated basis only; Unittec's item must be distinguished from its RMB 238 million Shenyang metro project of 2024-12, and this report could not obtain the original filing for this one, so it is down-weighted) | Sina Finance search page |
| 19 | 07-29 | PBoC announcement | The PBoC will conduct RMB 600 billion of overnight reverse repos daily from 7/29 to 7/31 and RMB 300 billion on 8/3, at a fixed rate with quantity tendering | Liquidity | Whole market | B | Sina Finance |
| 20 | 07-30 after the close (obtained by this report) | East Money block-trade disclosure stock_lhb_detail_em. ⚠️ Three scope warnings must be read together: ① the disclosure covers only the top five buying and selling seats (including the institutional-dedicated seat) and is not whole-market flow — JCET's net buying of RMB 317 million against RMB 10.647 billion of turnover that day = 3.0%, Unisplendour's RMB 548 million against RMB 8.755 billion = 6.3%; ② Tongfu Microelectronics' −RMB 3.2 billion is on the "three-consecutive-day decline board" basis while −RMB 477 million is the "single-day decline board" basis; the two have different time windows and cannot be placed directly alongside the single-day figures for JCET/Unisplendour; ③ this report did not obtain the nature of the buying seats (hot money / institution / Stock Connect), so it describes them only as "the top-five buying seats were net buyers, direction positive" and does not characterize this as "institutional bottom-fishing" |
⚠️ Structural evidence where direction contradicts price: limit-up 德明利 (Demingli, 001309) showed net selling of −RMB 706 million (buy RMB 1.249 billion / sell RMB 1.955 billion, turnover ratio 18.57%), limit-up 江淮汽车 (JAC Motors, 600418) net selling of −RMB 393 million; while limit-down 紫光股份 (Unisplendour, 000938) showed net buying of +RMB 548 million, limit-down 长电科技 (JCET, 600584) +RMB 317 million, 星宸科技 (SigmaStar) +RMB 282 million, 有研硅 (GRINM Semiconductor) +RMB 132 million, 源杰科技 (Yuanjie) +RMB 125 million, 麦格米特 (Megmeet) +RMB 107 million; 通富微电 (TFME) net selling of −RMB 3.2 billion, 宏和科技 (Grace Fabric) −RMB 505 million, 利通电子 (Litong Electronic) −RMB 531 million | Flows | Whole-market structure | A | Data interface (East Money) |
| 21 | from 07-31 | Company filing | 山西汾酒 (Shanxi Fen Wine) goes ex-dividend today, paying a cash dividend of RMB 6.56 per share (7/30 was the record date) — its "decline" today includes a technical ex-dividend gap and cannot be read directly as weakness | Technical | Baijiu | C | Sina Finance |
| 22 | from 07-31 | Company filing | *ST 华闻 (*ST Huawen, 000793) resumes trading today with its delisting-risk warning removed, and its abbreviation changes to "ST 华闻" | Event | ST segment | C | Sina Finance |
| 23 | 07-31–08-03 | Sina Finance | ChinaJoy 2026 is held in Shanghai, themed "Travel with AI", with total exhibition area exceeding 140,000 square meters, a record high | Trade show | Gaming / AI applications | C | Sina Finance |
| 24 | 07-29 | MFA regular press conference | Responding to media reports that "the U.S. FCC is imposing import restrictions on foreign-made advanced robots" (⚠️ the FCC's statutory authority covers equipment authorization and the covered list, and "import restrictions" do not fully match its remit; this report did not obtain the original FCC text and flags it "to be verified"), MFA spokesperson Mao Ning expressed firm opposition to the over-generalization of national security | Geopolitics | Robotics (directionally negative for exports and positive for the domestic-substitution narrative; must not be read one way only) | B | Sina Finance |
| 26 | 07-30 02:00 (U.S. Eastern 7/29) | Xinhua, Sina Finance | ⭐The Fed holds rates for the fifth time this year (3.50%–3.75%), with 3 of 12 voters favoring a 25bp hike; Chair Warsh sharply reduces the use of forward guidance; the CME watch tool shows a 78.8% probability of a September hike; the backdrop is June CPI slowing to 3.5% YoY from May's 4.2%, but the U.S.–Iran situation pushing oil higher. U.S. equities fell sharply that day: Nasdaq −1.74% at 24442.94, S&P −1.52%, Dow −2.19% | Macro (counter-directional) | Global risk assets | S (negative) | Xinhua, Sina Finance |
| 25 | 07-30 15:00 (yesterday's close) | Obtained by this report | Market sentiment at a freezing point yesterday: 52 limit-ups < 74 limit-downs, 19 failed limits; on Tonghuashun industry data semiconductors −7.08% (7 up, 176 down), electronic chemicals −7.35%, components −6.58%, communications equipment −5.33%; baijiu +4.48% (19 up, 0 down, best in the market), banks +2.43% (42 up, 0 down, net inflow of RMB 11.975 billion on the Tonghuashun sector-summary basis; the Tonghuashun industry fund-flow table shows RMB 10.852 billion, a difference of RMB 1.123 billion between the two tables, with the cause unverified) | Market | Whole market | A | Data interface |
Sentiment temperature over the past 5 trading days (limit-ups / limit-downs / failed limits, East Money pool basis)
| Date | Limit-ups | Limit-downs | Failed limits | Reading |
|---|---|---|---|---|
| 07-24 | 40 | 25 | 11 | Weak |
| 07-27 | 111 | 6 | 13 | Climax |
| 07-28 | 61 | 49 | 20 | Divergence |
| 07-29 | 81 | 9 | 14 | Repair |
| 07-30 | 52 | 74 | 19 | Freezing point: limit-downs > limit-ups, the only such day this week |
2. Strongest positive branches, in descending order
| Rank | Branch | Strength | Core news | Logical hardness | Persistence | Beneficiary path | Representative stocks (main board first; research coverage only, not recommendations) | Risks |
|---|---|---|---|---|---|---|---|---|
| 1 | Memory price increases (the memory super-cycle gets an endorsement from a downstream buyer) | S | Amazon 2026 capex $200bn → $220 billion, with CNBC explicitly attributing it to "rising memory costs"; Apple gives weak guidance on supply constraints; overnight SanDisk +22%, Micron +15%, Western Digital +15%, SK Hynix +15%, Seagate +12%, Silicon Motion +21.66% | Extremely hard: this is the largest downstream buyer using its own financial guidance to acknowledge memory price increases, not an upstream price-increase letter | Medium term (quarters), but the A-share names have already fallen for a month, so the near term is a rebound rather than a reversal | Memory manufacturers/modules → packaging & testing → interface/controllers → distribution | 兆易创新 (GigaDevice, 603986), 德明利 (Demingli, 001309), 深科技 (Shenzhen Kaifa, 000021), 太极实业 (Taiji Industry, 600667), 紫光国微 (Unigroup Guoxin, 002049), 深圳华强 (Shenzhen Huaqiang, 000062), 中电港 (CECPort, 001287) | Price increases are a double-edged sword for module makers (they must buy wafers externally); Demingli showed net selling of RMB 706 million on the 7/30 block-trade disclosure; the sector has already retraced 50%+ from its high in July |
| 2 | Upward revisions to overseas cloud capex → AI hardware oversold rebound | A+ | AWS +37%, backlog $496 billion, Microsoft +15.5%, SOXX +8.5%, ASE +11.79%, TSMC +7.63%; Jassy says capacity will be insufficient in both 2026 and 2027 | Hard (at the industry-trend level), but for A-share names this is an indirect supply-chain positive rather than direct orders | Mainly 1–3 sessions: this is a rebound within a downtrend unless sustained volume expansion appears | Packaging & testing → PCB/copper-clad laminate → optical modules → servers/liquid cooling | 长电科技 (JCET, 600584), 通富微电 (TFME, 002156), 华天科技 (TSHT, 002185), 沪电股份 (WUS Printed Circuit, 002463), 生益科技 (Shengyi, 600183), 深南电路 (SCC, 002916), 东山精密 (DSBJ, 002384), 光迅科技 (Accelink, 002281), 剑桥科技 (CIG, 603083), 浪潮信息 (Inspur, 000977), 工业富联 (FII, 601138), 中科曙光 (Sugon, 603019), 紫光股份 (Unisplendour, 000938), 英维克 (Envicool, 002837) | Innolight's controlling shareholder is cutting its stake (the branch leader delivers a negative); Apple −6% after hours; cumulative July declines of 15–25%, so the rebound runs into dense trapped supply; semiconductors alone accounted for 10 of yesterday's 74 limit-downs |
| 3 | First full trading day after the Politburo meeting | A | The "AI+" initiative, the "six networks" (including the compute network), stepped-up expansion of domestic demand, stabilizing property, the package debt-resolution plan, reform of small and medium financial institutions, deepening comprehensive reform of capital-market investment and financing and enhancing capital-market resilience and confidence, continued rectification of "involutionary" competition | Moderately hard: the policy tone is clear, but getting to individual stocks requires grading "concept ≠ beneficiary", and the meeting named no A-share company | Medium term (policy is a slow variable); on the tape it usually means a first-day spike and second-day divergence | ① "AI+"/compute network → resonating in the same direction as branches 1 and 2 (this is today's most important structure) ② domestic demand/services consumption ③ property ④ brokers ⑤ banks ⑥ anti-involution | Property: 陆家嘴 (Lujiazui, 600663, limit-up yesterday), 香江控股 (Xiangjiang Holdings, 600162, limit-up yesterday), 保利发展 (Poly Developments, 600048), 万科 A (China Vanke A, 000002); grid: 国电南瑞 (NARI, 600406), 海兴电力 (Hexing Electrical, 603556, 2 consecutive limit-ups yesterday); brokers: 中信证券 (CITIC Securities, 600030), 东方财富 (East Money, 300059, not main board) | ⚠️ Strong counter-evidence: grid equipment fell −2.46% for the full session on 7/30, with net outflows of RMB 2.664 billion and 27 up versus 113 down (Tonghuashun industry basis, a full-session figure; this report did not obtain minute-level data for the 14:38–15:00 window after the press report, which is "no reliable data available", so no attribution is made) — full-session weakness itself shows the market did not front-run this policy. The "compute network" conflicts in direction with yesterday's AI hardware collapse (but the external environment has clearly improved today, so that contradiction is partly eased today — note "improved" does not equal "reversed"; see the qualifications on the 7/29 Fed decision in section 0) |
| 4 | Photovoltaic "anti-involution" price compliance (Yancheng meeting today) | A | Today the SAMR Price Supervision and Competition Bureau holds a PV price-compliance guidance session in Yancheng, Jiangsu; the General Rules for the PV Industry Cost Accounting Model took effect on 7/27; the Politburo meeting simultaneously stated "continue the comprehensive rectification of involutionary competition" | Moderate: the meeting is "price-compliance guidance", not production cuts and not quotas, and the path to earnings delivery is long | Event-driven, 1–2 sessions; unless a substantive cost floor follows the meeting | Polysilicon → wafers → cells → modules → auxiliary materials | 通威股份 (Tongwei, 600438), 隆基绿能 (LONGi, 601012), 弘元绿能 (Hongyuan Green Energy, 603185, limit-up yesterday), 福莱特 (Flat Glass, 601865), 协鑫集成 (GCL System Integration, 002506) | PV anti-involution has been traded several times already, with clear diminishing marginal effect; only Hongyuan Green Energy hit the limit-up in the PV equipment sector yesterday, so no echelon has formed |
| 5 | Baijiu / mass consumption (day 3, but the premise has changed) | B+ (downgraded) | Baijiu +4.48% yesterday, best in the market, 19 up 0 down, net inflow RMB 2.03 billion; 舍得酒业 (Shede Spirits) and 金徽酒 (Jinhui Liquor) both limit-up; Moutai raised the Feitian price to RMB 1639/bottle on 7/18; fund baijiu heavy-holding weight fell to a historic low of 0.97% in Q2 | Soft-leaning: the public attribution is "technology receding → safe-haven money rotating back + valuation repair" (liquor analyst Cai Xuefei explicitly called it "valuation repair rather than a fundamental turn"). Today the premise of technology receding has been overturned | Today is a falsification day: it must prove "baijiu can hold up even while technology rebounds" | Premium → sub-premium → regional baijiu | 贵州茅台 (Kweichow Moutai, 600519), 五粮液 (Wuliangye, 000858), 山西汾酒 (Shanxi Fen Wine, 600809, ex-dividend of RMB 6.56 today), 舍得酒业 (Shede Spirits, 600702), 金徽酒 (Jinhui Liquor, 603919) | The heavyweights already surged yesterday (Moutai +3.09%, Fen Wine +6.04%, Wuliangye +4.51%), so the risk/reward on day 3 is worse; money rotating back into technology is its biggest opposing force; Jinhui Liquor showed net selling of −RMB 8 million on yesterday's block-trade disclosure |
| 6 | Banks / insurance (allocation-style slow variable) | B | Banks +2.43% yesterday, 42 up 0 down, net inflow RMB 11.975 billion (sector-summary basis) / RMB 10.852 billion (industry fund-flow basis, a difference of RMB 1.123 billion between the two tables, cause unverified); three Politburo lines map directly: "package debt-resolution plan", "reform of small and medium financial institutions", "enhancing capital-market resilience and confidence" | Hard but slow: the policy fit is clear, but it produces neither limit-ups nor a thematic echelon | Trend-level, with the lowest intraday elasticity | Large state banks → joint-stock banks → insurers → brokers | 工商银行 (ICBC, 601398), 招商银行 (China Merchants Bank, 600036), 平安银行 (Ping An Bank, 000001), 中国平安 (Ping An Insurance, 601318), 浦发银行 (SPD Bank, 600000) | It is a seesaw with branches 1 and 2: a day when technology rebounds is usually a day when banks cool; yesterday's strength was itself the mirror image of the technology collapse |
| 7 | Humanoid robots (Unitree's IPO timetable is set) | B+ | Unitree: book-building 8/5, subscription 8/10, payment 8/12, raising RMB 4.202 billion; 2025 revenue RMB 1.699 billion, net profit RMB 278 million | Moderate: an IPO by itself adds not one cent of revenue to any A-share listed company, so this is an "event-driven mapping", and the vast majority of associated names are pure concept mappings | The event window is 8/5–8/10; today is a warm-up | Supplier mapping (each must be verified against filings; no filing, automatic Pass) | This report gives no list of "Unitree supplier" stocks — any supply relationship without company-filing or annual-report customer disclosure support is treated as a pure concept mapping and does not enter the list | U.S. FCC import restrictions on foreign-made advanced robots are a counter-variable; A-share "Unitree concept" names have repeatedly been falsified after vague replies on interactive platforms |
| 8 | Passive components / MLCC (price increase effective 8/1) | B | Samsung Electro-Mechanics raises MLCC prices 30% across the board, effective 8/1, with placed-but-undelivered orders repriced retroactively; Taiyo Yuden follows on 9/1; an AI server uses 13x the MLCC content of an ordinary server | The news itself is hard, but the market has already voted it down once with its feet | It needs extremely strong new evidence to be re-established | MLCC → aluminum electrolytic / film capacitors | 风华高科 (Fenghua Advanced, 000636, +6.29% yesterday but 6 failed limits), 三环集团 (CCTC, 300408, not main board; announced a RMB 0.5–1.0 billion buyback yesterday) | ⚠️ The components sector fell −6.58% yesterday with 4 up, 58 down and 7 limit-downs — the price-increase letter failed on yesterday's tape; 江海股份 (Jianghai)'s "product-category mismatch" has already been falsified |
| 9 | Gold (Chifeng Jilong Gold's resource volume +143%) | B (A at the single-stock level, C at the sector level) | Chifeng Jilong Gold's Sepon project in Laos: metal content 107 tonnes → 260 tonnes | An increase in resource volume is a hard fact, but it does not equal current-period profit | A single-stock event, it does not constitute a sector | — | 赤峰黄金 (Chifeng Jilong Gold) (single-stock event tracking); 紫金矿业 (Zijin Mining, 601899) rose +1.71% yesterday as a baseline sector move, unrelated to this news and not to be conflated | Resource volume ≠ reserves ≠ recoverable reserves ≠ current output; no profit figure can be derived from it |
2.1 The six questions every branch must answer
Branch 1 · Memory price increases
- Why is it positive: Amazon raised 2026 capex by $20 billion and publicly attributed it to memory costs, which amounts to the largest downstream buyer endorsing upstream prices; together with SanDisk +22%, Micron +15% and SK Hynix +15%, rising prices are no longer a rumor.
- Short term or medium term: the logic is medium term (quarters); the price action of A-share names is a short-term rebound — the two must be viewed separately.
- Is there a path to earnings delivery: yes, but the path forks. Manufacturers/packaging & testing (directly benefiting from price and capacity) > interface/controllers (volume and price both rising) > module makers (they buy wafers externally, so price increases hit costs first and selling prices later, and gross margin may fall before it rises). This is the layer most often misread in this branch.
- Is it fully priced in: it is not fully priced in; if anything it is priced in reverse — A-share memory leaders fell 50%+ in July, with the market previously pricing "AI capex has peaked".
- Is there one-day-wonder risk at highs: yes for Demingli (limit-up yesterday + net selling of RMB 706 million on the block-trade disclosure + 18.57% turnover); relatively small for GigaDevice (only +1.94% yesterday, with a buyback as support).
- Main-board coverage: ample, and this is the only branch in this report that is "S grade + ample main-board coverage" — 兆易创新 (GigaDevice, 603986, Shanghai main board), 德明利 (Demingli, 001309, Shenzhen main board), 深科技 (Shenzhen Kaifa, 000021), 太极实业 (Taiji Industry, 600667), 紫光国微 (Unigroup Guoxin, 002049).
Branch 2 · AI hardware oversold rebound
- Why is it positive: AWS backlog $496 billion + capex $220 billion + Microsoft +15.5% prove that the "AI capex has peaked" narrative was falsified this quarter; ASE +11.79% maps directly onto A-share packaging and testing.
- Short term or medium term: short term. This is an oversold rebound, not the start of a new cycle.
- Is there a path to earnings delivery: indirect. Amazon's and Microsoft's capex lands mainly in the North American supply chain, and the vast majority of A-share names are "industry-trend positives" or "indirect supply-chain positives"; only a handful of companies with actual overseas customer share can speak of direct orders — and this report cannot verify that layer name by name, so all of it is treated as indirect.
- Is it fully priced in: in reverse — it is priced with excessive pessimism.
- Is there one-day-wonder risk at highs: extremely high. The larger the opening gap up, the more likely it is a "window for distributing to trapped holders", and this is this report's main warning for today.
- Main-board coverage: ample (see the table above). In a rebound the main board's 10cm limit makes it easier to observe absorption strength than 20cm names — that is a difference in observability, not a judgment of quality.
Branch 3 · The Politburo meeting
- Why is it positive: the policy tone is "keep exerting force and add force at the right time", and "AI+" plus "the compute network" run in the same direction as today's overseas AI rebound — the most important "internal-external resonance" structure today.
- Short term or medium term: policy is medium term, and the tape usually shows a first-day spike.
- Is there a path to earnings delivery: layered — debt resolution / reform of small and medium financial institutions is a genuine asset-quality improvement for banks (medium term); the "six networks" means orders for grid equipment (medium term); "capital-market resilience" is sentiment for brokers (short term).
- Is it fully priced in: partly (the "six networks" was already raised at the April meeting, making this a re-confirmation of an old formulation rather than brand-new news, and that must be flagged).
- Is there one-day-wonder risk at highs: yes, especially for small caps that are pure policy mappings.
- Main-board coverage: ample, and dominated by large-cap blue chips.
Branch 4 · PV anti-involution: a short-term event; the path to earnings delivery is long (price compliance ≠ price increases ≠ profit); repeatedly anticipated already; high one-day-wonder risk; ample main-board coverage. Branch 5 · Baijiu: today is a falsification day rather than a continuation day, for the reasons in the table above; the path to earnings delivery must wait for Q3 results; ample main-board names. Branch 6 · Banks: medium-term allocation, no limit-ups, no echelon; ample main-board names.
2.2 ⭐ The biggest expectation gap this report sees today: interim pre-announcements across the whole chain have already exploded, while share prices fell 20–50% in July
This is the piece of evidence most deserving of its own table this issue. Every item below is a 2026 interim results pre-announcement already filed by the company — not a forecast and not a sell-side estimate; the right-hand two columns are their actual July share-price performance (Sina daily bars, forward-adjusted, 7/1 close → 7/30 close).
| Company (code) | Board | 2026H1 pre-announcement (net profit attributable to parent) | YoY | 7/1→7/30 share price | 7/30 single day |
|---|---|---|---|---|---|
| 兆易创新 (GigaDevice, 603986) | Shanghai main board | revenue about RMB 11.5bn, net profit attributable to parent about RMB 6.9bn, ex-non-recurring about RMB 4.85bn | +1099% (ex-non-recurring +791%) | 772.0 → 371.1, −51.9% | +1.94% |
| 通富微电 (TFME, 002156) | Shenzhen main board | RMB 1.6–1.8bn | +288.3%~+336.8% | — | −10.00% (limit-down) |
| 华天科技 (TSHT, 002185) | Shenzhen main board | RMB 750–850mn | +231.2%~+275.3% | — | −10.02% (limit-down) |
| 浪潮信息 (Inspur, 000977) | Shenzhen main board | RMB 2.6–3.1bn (above full-year 2025) | +226%~+288% | — | −7.82% |
| 东山精密 (DSBJ, 002384) | Shenzhen main board | RMB 2.9–3.0bn | +282.6%~+295.8% | — | −10.00% (limit-down) |
| 生益科技 (Shengyi, 600183) | Shanghai main board | RMB 3.099–3.298bn | +117%~+131% | — | −7.49% |
| 工业富联 (FII, 601138) | Shanghai main board | RMB 23.4–24.4bn | +93%~+101% | — | −6.84% |
| 紫光股份 (Unisplendour, 000938) | Shenzhen main board | RMB 1.91–2.32bn | +83.5%~+122.9% | — | −9.99% (limit-down) |
| 沪电股份 (WUS, 002463) | Shenzhen main board | RMB 2.83–3.00bn | +68.2%~+78.3% | 158.20 (7/1 intraday high) → 96.55, −39.0% | −8.27% |
| 长电科技 (JCET, 600584) | Shanghai main board | RMB 770–950mn | +63.5%~+101.7% | — | −10.00% (limit-down) |
| 深南电路 (SCC, 002916) | Shenzhen main board | RMB 2.1–2.3bn | +54.4%~+69.1% | — | −9.97% |
| 剑桥科技 (CIG, 603083) | Shanghai main board | — | +156.7%~+197.2% | — | −10.00% (limit-down) |
| 长飞光纤 (YOFC, 601869) | Shanghai main board | — | +711%~+914% | — | −6.76% |
| 亨通光电 (Hengtong, 600487) | Shanghai main board | — | +86.9%~+121.2% | — | −5.13% |
| 永鼎股份 (Jiangsu Yongding, 600105) | Shanghai main board | RMB 500–700mn | +57%~+120% | 62.5 → 33.0, −47.2% | +7.33% |
| 新易盛 (Eoptolink, 300502) | ChiNext | — | +77.6%~+102.9% | ⚠️ (this cell is the single-day decline, not the July interval) 421.20 → 371.10. Note: this closing price is exactly the same as GigaDevice's RMB 371.10 on 7/30; it has been re-checked against the Sina full-market snapshot and is a coincidence, not a row-shift error | −11.90% |
| 德科立 (Taclink, 688205) | STAR Market | — | +216.8%~+277.3% | — | — |
Data sources and scope warnings: the pre-announcement figures come from Securities Times ("Optical communications companies: pre-announcements explode as a group", "AI demand transmits to the packaging and testing end", "Tight supply-demand: broad positive pre-announcements across the optical communications chain in H1"), East Money, National Business Daily and company filings; of these, GigaDevice, WUS Printed Circuit and Demingli have been verified against the original company filings, while the rest are on a media-aggregated basis, and the company filings govern. ⚠️ Three scope limits on the share-price column: ① only 5 of the 17 rows have 7/1→7/30 interval data, with the rest shown as "—", so the "fell 20–50%" statement in the main text holds only for those 5 rows and the names listed in section 0, and must not be extrapolated to the whole table; ② the column mixes closing prices and intraday highs (WUS uses the 7/1 intraday high of 158.20, the rest are closing prices), so declines are not comparable across rows; ③ the Eoptolink row shows a single-day decline rather than an interval decline, and is flagged separately. Also note: Eoptolink's 7/30 close of RMB 371.10 is numerically identical to GigaDevice's 7/30 close of RMB 371.10; this report has re-checked it against the Sina full-market snapshot and it is indeed a coincidence, not a row-shift error.
What this table shows (judgment):
- The July decline in the A-share AI hardware chain was not driven by downward earnings revisions; it was driven by the valuation side plus flow behavior. During the week of 7/13–7/15 the entire chain disclosed historic interim pre-announcements as a group, and share prices then accelerated downward.
- All 5 main-board names that hit the limit-down yesterday (TFME, TSHT, DSBJ, Unisplendour, JCET) are companies whose interim pre-announcements doubled or nearly doubled. That is the direct basis for the "expectation gap" scoring item today.
- But the other side must be stated at the same time: a doubling pre-announcement does not mean the share price has bottomed. The lesson of July is precisely that "the day the pre-announcement is published is the interim high" — Jiangsu Yongding filed its positive pre-announcement on 7/5 and fell close to the limit on 7/6; Shenzhen Kaifa retraced 45% from its 7/1 high; Taiji Industry retraced 51% from its 7/1 high. Good news being realized as distribution has already happened repeatedly on this chain in July.
- This report's characterization for today is therefore: "the fundamental evidence is sufficient, but the market has already ignored it three times in July". The key today is not whether the news is strong enough, but whether this overseas slap can turn around the flow behavior on this A-share chain — which can only be verified by intraday signals, not inferred pre-market (see section 8).
3. Overall ranking of single-stock catalyst strength (main board, by total score, descending)
Grading rules restated: this list contains only the Shanghai and Shenzhen main boards. ChiNext (300/301), STAR Market (688) and BSE (920) names all go into the "watch zone, not primary coverage" in 3.2. There are only four labels: priority deep-dive / watch closely / watch only / Pass. A label is not a buy or sell recommendation; it is a ranking of research attention. Unified scope note (added after QC): "watch" in the 3.2 watch zone is equivalent to "watch only" (as a non-main-board name it cannot enter the primary list under the iron rules); the "fundamentals-dimension Pass / dual-dimension Pass" in section 6 both belong to the "Pass" label, with the sub-types used only to describe the strength of the Pass, not as a fifth or sixth label. 紫光国微 (Unigroup Guoxin, 002049) is recorded as "watch closely" in section 3, while "pure concept (with respect to memory)" in section 7.1 refers only to its role within the "memory price increase" branch; the two do not conflict — the section 3 label governs. Catalyst grade: S / A+ / A / B+ / B / C. In the "fundamentals" column of this section, any entry marked ⭐ is a score (0–15) assigned after this report used the
fundamentals-analystsub-agent to verify the company's annual report / filings line by line; the rest are inferred from public financial data.
| Rank | Code | Name | Shanghai/Shenzhen main board | Branch | Catalyst grade | Total score | Core news | Catalyst path | Directness of benefit | Fundamentals / industry position | Expectation gap | Technicals & sentiment | Risks | Conclusion |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 603986 | 兆易创新 (GigaDevice) | ✅ Shanghai main board | Memory | S | 71 | Amazon's capex raise attributed to memory costs; TrendForce: NOR Flash H1 contract prices up 100–120% cumulatively, DDR4 8Gb estimated +50% QoQ in Q3 | Memory chips are 71.3% of revenue and 76.0% of gross profit, so price increases flow straight into the income statement | Direct (with respect to price increases) / industry trend (with respect to the specific Amazon catalyst) | ⭐11/15: top three globally in NOR Flash, number one domestically in 32-bit MCUs; zero interest-bearing debt, 8.13% debt-to-asset ratio, OCF/net profit >1 for 4 consecutive periods, contract liabilities +283% | Very large: H1 net profit attributable to parent of about RMB 6.9bn (+1099%) already filed, while the share price is −55.8% from its 6/29 high | Limit-down on 7/28, −6.81% on 7/29, +1.94% on 7/30, turnover RMB 35.484bn and turnover ratio 14.53%, massive divergence | Governance (RMB 4.4bn cashed out vs a buyback ceiling of RMB 2bn, and the stake increase cannot start until 12/13); CXMT's 7/27 listing dismantled the "shadow stock" premium in one stroke; PB 10.30x, the 90.1st percentile of the past five years; RMB 2.05bn of the RMB 6.9bn H1 net profit is a non-recurring fair-value change | Priority deep-dive |
| 2 | 002463 | 沪电股份 (WUS Printed Circuit) | ✅ Shenzhen main board | AI PCB | A+ | 68 | AWS backlog $496bn, Microsoft +15.5%, SOXX +8.5% | Datacom PCB is 77.4% of revenue, of which high-speed switches and routers are 43.1% and AI servers plus HPC 15.9% | Industry-trend positive (strong, at the top of that tier) — the company has never disclosed a customer name and is Tier-2 to the cloud vendors | ⭐12.5/15: highest on the whole list. 2025 net profit attributable to parent +47.7%, gross margin 35.48%, ROE 28.57%, capacity utilization of 92.5%–99.7% across five bases | Large: the 7/14 pre-announcement put H1 net profit attributable to parent at RMB 2.83–3.00bn (+68.2%~78.3%), with Q2 alone at about RMB 1.673bn, the strongest sequential quarter in four years; yet after the pre-announcement the share price is −39.0% from its 7/1 high | −8.27% on 7/30, touched the limit-down at 94.73 intraday before recovering, turnover RMB 7.964bn | PB at the 95.6th percentile of its full history; 2026Q1 OCF/net profit collapsed to 0.41, inventory turnover 103 days; top-five customer concentration rose to 58.4% in 2026Q1; an H-share re-filing has been made, with dilution timing unknown | Priority deep-dive |
| 3 | 600584 | 长电科技 (JCET) | ✅ Shanghai main board | Packaging & testing | A+ | 67 | ASE Technology +11.79% overnight, TSMC +7.63%; packaging and testing pulled by both AI and memory | Packaging and testing is the shortest hop on the "cloud capex → chips → packaging" chain, and it eats from both the AI and the memory line | Industry-trend positive (strong) | 12/15 (inferred from public financials): number one in domestic packaging and testing, top three globally | Large: H1 pre-announcement of RMB 770–950mn (+63.5%~101.7%); limit-down on 7/30, yet net buying of +RMB 317mn on the block-trade disclosure | Limit-down on 7/30 with RMB 10.65bn of turnover and counter-trend net buying on the block-trade disclosure — the cleanest "price down, money in" name in the market | The lowest growth rate of the packaging-and-testing big three (its business spans consumer/memory/automotive, so compute profits are diluted); a cumulative decline of over 20% across three consecutive sessions | Priority deep-dive |
| 4 | 601138 | 工业富联 (FII) | ✅ Shanghai main board | AI server ODM | A+ | 64 | Amazon capex $220bn, AWS +37%; Microsoft Azure +43% | A global cloud-vendor ODM, and the direct recipient of overseas capex | Industry-trend positive (upper edge), and the name on this list whose financials come closest to "direct benefit" (basis: 2026Q1 revenue +56.5% with positive operating cash flow, the opposite direction from domestically-driven system makers). ⚠️ Scope correction: like WUS Printed Circuit, it has never disclosed customer names or the share of overseas revenue, so it cannot be classified as a "direct order positive" — using past quarterly financial results to infer the directness of today's news is correlation, not causation | 12/15: global cloud ODM leader, 2026Q1 net profit attributable to parent RMB 10.595bn (+102.5%) | Medium: H1 pre-announcement of RMB 23.4–24.4bn (+93%~101%); −6.84% on 7/30 | Its decline was smaller than the rest of the chain (−6.84% versus mostly limit-downs), which says the shareholder structure is steadier but also that elasticity is smaller | Enormous scale (2026Q1 revenue RMB 251.08bn) naturally caps share-price elasticity; gross margin is only 7.35% | Priority deep-dive |
| 5 | 600183 | 生益科技 (Shengyi) | ✅ Shanghai main board | Copper-clad laminate | A | 62 | Upstream of the AI PCB demand chain; AWS/Microsoft capex | CCL is the upstream material for PCB, and AI high-frequency high-speed laminates benefit directly | Industry-trend positive | 11/15: H1 pre-announcement of RMB 3.099–3.298bn (+117%~131%) | Large: a doubling pre-announcement versus −7.49% on 7/30 | −7.49% on 7/30, turnover RMB 8.25bn; −46.5% from its July high | Weaker pricing power at the material end than at the board end; PE (TTM) 61.9x, higher than WUS and Victory Giant | Watch closely |
| 6 | 002185 | 华天科技 (TSHT) | ✅ Shenzhen main board | Packaging & testing | A | 61 | ASE +11.79% | One of the packaging-and-testing big three | Industry-trend positive | 10.5/15: H1 pre-announcement of RMB 750–850mn (+231.2%~275.3%) | Large | Limit-down on 7/30, turnover RMB 6.19bn, turnover ratio 12.03% | Block-trade disclosure shows net selling of −RMB 203mn with 4 institutional sell seats; the share price is −41.9% from its July high | Watch closely |
| 7 | 002156 | 通富微电 (TFME) | ✅ Shenzhen main board | Packaging & testing | A+ | 61 | ASE +11.79%, TSMC +7.63% | Media describe its "compute business as pure" (no revenue breakdown obtained; to be verified) | Industry-trend positive (strong) | 11/15: H1 pre-announcement of RMB 1.6–1.8bn (+288.3%~336.8%), the fastest growth of the packaging big three | Very large: a near-4x pre-announcement versus a limit-down on 7/30 | Limit-down on 7/30, turnover RMB 9.19bn, cumulative decline over 20% across three consecutive sessions | ⚠️ Top-five buying seats were net sellers: −RMB 477mn on the single-day decline-board basis (5.2% of that day's turnover) / −RMB 3.2bn on the three-consecutive-day decline-board basis, with 4 institutional sell seats — the opposite direction from JCET (the comparable figure is the single-day basis; both net amounts are only single-digit percentages of turnover, so this is a weak signal) | Watch closely |
| 8 | 002916 | 深南电路 (SCC) | ✅ Shenzhen main board | PCB | A | 60 | AWS/Microsoft capex; AI PCB | An integrated platform of PCB + packaging substrates + electronic assembly | Industry-trend positive | 11/15: H1 pre-announcement of RMB 2.1–2.3bn (+54.4%~69.1%); 2025 revenue of RMB 23.65bn is the largest of the four | Medium: the lowest growth rate of the AI PCB big four | −9.97% on 7/30, turnover RMB 4.08bn | Gross margin 28.32%, clearly below WUS at 35.48% and Victory Giant at 35.22%, so its AI purity is a notch lower than the first two; PE (TTM) 53.3x | Watch closely |
| 9 | 000938 | 紫光股份 (Unisplendour) | ✅ Shenzhen main board | Servers / networking | A | 59 | Microsoft Azure +43%, AWS +37%; the Politburo's "AI+" and the compute network | H3C: networking equipment (switches/routers) plus servers, a two-engine model | Industry-trend positive | 10.5/15: H1 pre-announcement of RMB 1.91–2.32bn (+83.5%~122.9%); gross margin 12.88%, the highest of the four system makers | Large | Limit-down on 7/30, yet net buying of +RMB 548mn on the block-trade disclosure (the largest of all limit-down names), turnover RMB 8.76bn | −25.2% from its July high; 1 institutional sell seat | Watch closely |
| 10 | 001309 | 德明利 (Demingli) | ✅ Shenzhen main board | Memory modules | A | 58 | Silicon Motion (memory controllers) +21.66% overnight, SanDisk +22%, Micron +15%; the company's 7/15 pre-announcement put H1 revenue at RMB 16–18bn (+289%~338%), net profit attributable to parent at RMB 5.7–6.5bn (swing to profit, +4933%~5611%), ex-non-recurring at RMB 5.649–6.449bn, EPS RMB 25.35–28.91 | Memory modules/controllers, whose gross margin is a leveraged function of the rate of change of NAND/DRAM prices | ⚠️ Industry-trend positive (slope-sensitive), not a direct order positive — SanDisk/Micron/SK Hynix are upstream manufacturers, and Demingli is their customer, not their supplier; Amazon's $220bn capex corresponds to orders for the manufacturers and HBM makers, and Demingli has no verifiable related order filing | ⭐9/15: one of the module big three but not the leader — number 2 by H1 revenue (Longsys RMB 22–25bn > Demingli RMB 16–18bn > Biwin RMB 15–16bn), number 4 by profit (Longsys RMB 9.2–11.0bn > Biwin RMB 7.0–7.5bn > GigaDevice about RMB 6.9bn > Demingli RMB 5.7–6.5bn), with an implied net margin of 35.9%, the lowest; it has launched a self-developed PCIe/SATA dual-mode enterprise SSD controller | Large but converging: annualized H1 PE is only about 7.3x, essentially the same as Longsys at 7.1x and Biwin at 7.1x, neither a discount nor a premium; if gross margin reverts to the middle of its 2024 range, the normalized PE is 26–33x | The only limit-up among 176 decliners and 7 gainers as semiconductors fell −7.08% on 7/30, sealed at 13:22, turnover RMB 11.366bn, turnover ratio 18.57% (free-float basis) | ⚠️ This report's four biggest reservations: ① net selling of −RMB 706mn on the block-trade disclosure on the limit-up day (buy RMB 1.249bn / sell RMB 1.955bn, 28.19% of the day's turnover), with 3 institutional buy seats but a success rate of only 3.33%; ② backing out the company's own pre-announcement, 2026Q2 net profit attributable to parent is −5.7%~−29.7% QoQ and the net margin falls from 44.4% in Q1 to 27.8%~30.1% — the profit peak may already have occurred in 2026Q1; ③ inventory of RMB 12.192bn = 1.84x net assets and 66.3% of total assets, the highest among peers, while cumulative net profit attributable to parent from 2023 to 2026Q1 of RMB 4.41bn compares with cumulative operating cash flow of −RMB 4.76bn; ④ the number of shareholder accounts rose from 55,700 to 124,000 (+122.6%) in three months | Watch closely (position awareness must be corrected: RMB 390.04 is not a high; it is a rebound level after a −60.2% retracement from the 6/26 intraday high of RMB 980.00, with a 7/29 intraday low of RMB 334.00 and a maximum drawdown of −65.9%; the accurate description is "a high-volatility rebound after a 60% collapse, with the trend not yet repaired") |
| 11 | 000977 | 浪潮信息 (Inspur) | ✅ Shenzhen main board | AI servers | A | 57 | Microsoft/Amazon capex; the Politburo's "compute network" and "AI+" | Number one in domestic AI server share | ⚠️ Industry-trend positive (weak) + an indirect supply-chain negative — see the dedicated note below | ⭐10/15: H1 pre-announcement of RMB 2.6–3.1bn (+226%~288%), one half-year exceeding all of 2025; but gross margin is only 4.88%, cumulative conversion of net profit into cash over 6 years is just 15.8%, and 2026Q1 OCF was −RMB 7.772bn | Large | −7.82% on 7/30 on shrinking volume, turnover RMB 6.5bn | ⚠️ According to public reports, the company itself has been on the U.S. BIS Entity List since 2025-03-25 (the company responded in its 2024 annual report that "the impact is broadly manageable"; this report did not obtain the original BIS notice number, nor verify whether any license exemption or subsequent removal exists, and flags the basis as "to be verified"); if true, AWS/Azure orders are structurally unaddressable; 2026Q1 revenue −24.3% (versus FII at +56.5%); debt-to-asset ratio 73.64%, quick ratio 0.43, inventory of RMB 46.5bn is 55.5% of total assets | Watch closely |
| 12 | 002384 | 东山精密 (DSBJ) | ✅ Shenzhen main board | PCB / consumer electronics | A | 56 | The AI PCB chain; but Apple −6% after hours is a counter-variable | PCB + precision manufacturing | Industry-trend positive | 10/15: H1 pre-announcement of RMB 2.9–3.0bn (+282.6%~295.8%, low base) | Medium | Limit-down on 7/30, turnover of RMB 17.49bn, the largest on the whole main board, net selling of −RMB 236mn on the block-trade disclosure, 3 institutional sell seats | Apple-chain exposure is today's counter-risk (Apple gave weak guidance on supply constraints and fell 6% after hours); the high pre-announced growth comes from a low base | Watch closely |
| 13 | 000021 | 深科技 (Shenzhen Kaifa) | ✅ Shenzhen main board | Memory packaging & testing | B+ | 54 | Memory price increases; the surge in Micron/SanDisk/SK Hynix | Peyton Technology does DRAM/NAND packaging and testing | ⚠️ Industry-trend positive (lower edge, close to indirect supply chain) | ⭐10/15: the memory segment's gross margin jumped from 13.56% to 27.63% in 2025H2 (hard evidence); a healthy balance sheet (net cash of about RMB 2.1bn) | Medium | −9.29% on 7/30, closing just RMB 0.28 above the limit-down, turnover RMB 5.98bn | ⚠️ Peyton's two bases together contributed only about RMB 175mn of net profit attributable to parent in 2025, about 15% of Shenzhen Kaifa's total, at a net margin of just 3.8%; 41% of gross profit comes from metering smart terminals, which have nothing to do with memory; the RMB 1.47bn capacity expansion will not be completed until June/December 2027; the company's 7/1 filing states explicitly that "HBM is still at the R&D stage and generates no revenue in the near term"; PB at the 96.2nd percentile of the past five years; 2026Q1 OCF/net profit attributable to parent 0.47x | Watch closely |
| 14 | 002281 | 光迅科技 (Accelink) | ✅ Shenzhen main board | Optical modules | B+ | 53 | AWS backlog $496bn, Microsoft Azure +43% | Datacom optical modules (data and access are 70.9% of revenue) | ⚠️ Industry-trend positive, but the weakest fundamental elasticity in this round's North American capex narrative | ⭐9/15: H1 pre-announcement of RMB 558.55–614.97mn (+50.0%~65.15%); contract liabilities +154% QoQ in 2026Q1 to RMB 1.095bn (the hardest forward-looking demand evidence); net cash of about RMB 3.98bn plus RMB 3.5bn from a private placement received | Medium | Limit-down on 7/30, turnover RMB 6.15bn, net selling of −RMB 44mn on the block-trade disclosure, 4 institutional sell seats; 4 limit-downs already in July | ⚠️ Overseas revenue share has fallen for two straight years: 34.3% (2023) → 28.3% (2024) → 26.7% (2025); the company's own pre-announcement attributes the growth to "domestic cloud service providers" and never mentions North America; PE (TTM) 121.7x is 1.9x Innolight's 64.5x and 2.5x Eoptolink's 48.2x, while its revenue growth is only 1/8 of Innolight's; the placement price of RMB 167.55 is 9.08% underwater; inventory RMB 7.478bn with 293 days of turnover, and impairments widening year by year (RMB 138mn → 224mn → 310mn) | Watch closely (explicit downgrade: do not treat it as "the main-board version of Innolight") |
| 15 | 603019 | 中科曙光 (Sugon) | ✅ Shanghai main board | Compute | B+ | 52 | The Politburo's "compute network" and "AI+"; the overseas AI rebound | Systems + storage + software + liquid cooling + IT localization | Industry-trend positive | 10/15: gross margin 26.56%, the highest among system makers; 2026Q1 revenue +23.7% | Medium | −6.51% on 7/30, turnover RMB 5.56bn, a smaller decline than the rest of the chain | No H1 pre-announcement issued (the only one among the top 15 on this list); 2026Q1 net profit attributable to parent was only RMB 228mn, with a single-quarter ROE of 1.02% | Watch closely |
| 16 | 603083 | 剑桥科技 (CIG) | ✅ Shanghai main board | Optical modules | B+ | 51 | AWS/Microsoft capex | Optical modules | Industry-trend positive | 9/15: H1 pre-announcement of +156.7%~197.2% | Large | Limit-down on 7/30, turnover RMB 3.31bn, net buying of +RMB 62mn on the block-trade disclosure | −50.4% from its July high; a high PE | Watch closely |
| 17 | 600487 | 亨通光电 (Hengtong) | ✅ Shanghai main board | Optical fiber and cable | B+ | 50 | The whole optical communications chain is running hot; AI data-center fiber demand | Optical fiber and cable + submarine cable | Industry-trend positive | 10/15: H1 pre-announcement of +86.9%~121.2% | Medium | −5.13% on 7/30, turnover RMB 8.47bn, a clearly smaller decline than optical modules, with the resilience coming from the submarine cable business | The sustainability of fiber price increases needs Q3 verification | Watch closely |
| 18 | 002049 | 紫光国微 (Unigroup Guoxin) | ✅ Shenzhen main board | Specialty ICs | B | 48 | Spillover from memory price increases; the broader semiconductor rebound | Specialty integrated circuits + smart security chips | Indirect supply-chain positive | 10/15 | Small | Only −0.25% on 7/30, the most resilient main-board semiconductor name, turnover RMB 1.55bn | Resilience means there is no room for an oversold rebound; its linkage to memory price increases is weak | Watch closely |
| 19 | 002837 | 英维克 (Envicool) | ✅ Shenzhen main board | Liquid cooling / thermal | B | 47 | Cloud capex revisions → data-center thermal management | Liquid cooling is a must-have as AI rack power density rises | Industry-trend positive | 9.5/15 | Medium | Limit-down on 7/30, turnover RMB 2.25bn | −45.8% from its July high; small scale, high volatility | Watch closely |
| 20 | 601869 | 长飞光纤 (YOFC) | ✅ Shanghai main board | Optical fiber | B+ | 46 | The optical communications upcycle | Integrated preform–fiber–cable | Industry-trend positive | 10/15: H1 pre-announcement of +711%~914% (the highest on the chain) | Large | −6.76% on 7/30; block-trade net buying and selling essentially level (−RMB 0.00bn); Shanghai money was a seller | The extremely high pre-announced growth comes from a low base; at RMB 262.66 the absolute price level is high | Watch closely |
| 21 | 600438 | 通威股份 (Tongwei) | ✅ Shanghai main board | PV anti-involution | A | 44 | Today SAMR holds a PV price-compliance guidance session in Yancheng; the Politburo's "continue the comprehensive rectification of involutionary competition" | The polysilicon leader, directly matched to the cost-accounting rules | Policy positive (moderate directness: this is "price-compliance guidance", not production cuts and not price increases) | 9/15 | Medium | +1.30% on 7/30 (closing green against the trend), turnover RMB 885mn | PV anti-involution has been traded several times, with diminishing marginal effect; only Hongyuan Green Energy hit the limit-up in the PV equipment sector yesterday, so no echelon formed; the path to earnings delivery is long | Watch closely |
| 22 | 600663 | 陆家嘴 (Lujiazui) | ✅ Shanghai main board | Property / policy | B+ | 43 | The Politburo meeting's "stabilize the property market" | Core-district commercial and office property in Shanghai | Policy mapping (not direct orders) | 8/15 | Medium | Sealed at the limit-up at 14:30 on 7/30 (+9.96%), turnover RMB 428mn; +19.0% over the past 5 sessions, already day 2 of the policy fermentation | It completed a round of pricing around the time the Politburo report appeared, so day 2 is a high-level relay; property fundamentals show no improvement | Watch only |
| 23 | 601012 | 隆基绿能 (LONGi) | ✅ Shanghai main board | Photovoltaics | B | 42 | Same as branch 4 | The module leader | Policy positive | 8.5/15 | Small | +1.64% on 7/30, turnover RMB 2.0bn | Same as Tongwei; the module end benefits less from the cost-accounting rules than the polysilicon end | Watch only |
| 24 | 600702 | 舍得酒业 (Shede Spirits) | ✅ Shanghai main board | Baijiu | B | 41 | The baijiu sector rose +4.48% on 7/30, best in the market | Sub-premium baijiu | ⚠️ A sentiment/flow positive, not a fundamental positive | 8/15 | Small | Limit-up on 7/30 (sealed at 09:42, 1 failed limit), turnover only RMB 522mn (RMB 117mn of sealing money) | ⚠️ The public attribution for the rise is "safe-haven money rotating back as technology receded + valuation repair", and today that premise of a technology retreat has been overturned; turnover is tiny and absorption is questionable | Watch only |
| 25 | 601398 | 工商银行 (ICBC) | ✅ Shanghai main board | Banks | B | 40 | The Politburo's "package debt-resolution plan", "reform of small and medium financial institutions", "enhancing capital-market resilience and confidence" | Asset-quality improvement + high dividend | Policy positive (slow variable) | 12/15 (sound fundamentals, but no elasticity) | None | +2.52% on 7/30, with the bank sector 42 up 0 down and net inflows of RMB 11.975bn (sector-summary basis) / RMB 10.852bn (industry fund-flow basis), both bases the highest in the market, though the absolute values cannot be mixed | A seesaw with branches 1 and 2: a day when technology rebounds is usually a day when banks cool; intraday elasticity is close to zero | Watch closely (an allocation attribute, not a trading attribute) |
| 26 | 600105 | 永鼎股份 (Jiangsu Yongding) | ✅ Shanghai main board | Optical chips | A | 38 | 7/29 filing: Suzhou Dingxin received high-power laser chip orders totaling RMB 1.133bn | Optical chips (CW-DFB / EML, pointing to silicon photonics/CPO external light sources) | ⭐Direct order positive (the highest tier), but subject to a triple discount | ⭐7/15: ⚠️ Yongding holds only 44.8567% of Suzhou Dingxin (100% of revenue is consolidated, but only about 44.9% of profit is attributable to the parent); 2025 optical communications segment revenue actually fell −10.5%; in 4 of the past 5 years, 100%+ of net profit attributable to parent came from associate Dongchang Investment (auto dealership + property) | ⚠️ Already heavily front-run on 7/30: the single-day market-cap increase of RMB 3.289bn ÷ the equity-adjusted annualized net profit attributable to parent of about RMB 47mn ≈ 70x | +7.33% on 7/30, touching the limit-up at 33.78 without sealing, maximum drawdown 7.9%, turnover RMB 9.431bn, turnover ratio 19.75% — a heavy-volume divergence/realization pattern | ⚠️ Anonymous customers (customers A and B, with a confidentiality exemption procedure); the orders are recognized in installments across 2.4 years, and RMB 1.133bn including tax ≈ about RMB 1bn excluding tax; the company was publicly censured by the Shanghai Stock Exchange in 2024 for "inaccurate disclosure of a results pre-announcement", and the executives involved are still in post; the controlling shareholder cashed out RMB 788mn within six months, with a pledge ratio of 59.4%; PB 14.5x, PE (TTM) 470x | Watch only (the positive is real, but the price has already run ahead of it) |
| 27 | 600418 | 江淮汽车 (JAC Motors) | ✅ Shanghai main board | Autos | B | 33 | The auto OEM sector rose +2.85% on 7/30 | Vehicle OEM | Sentiment | 8/15 | Small | Limit-up on 7/30, turnover RMB 18.075bn, sealing orders of RMB 195mn | ⚠️ Limit-up but net selling of −RMB 393mn on the block-trade disclosure, with 2 institutional buy seats; a turnover ratio of only 3.52% against enormous turnover indicates high-churn wash trading rather than locked-up chips | Watch only |
| 28 | 603556 | 海兴电力 (Hexing Electrical) | ✅ Shanghai main board | Grid equipment | B | 32 | The Politburo's "six networks" includes the new-type power grid | Meters / distribution | Concept mapping (the "six networks" was already stated at the 2026-04-28 meeting; this is a re-confirmation, not brand-new news) | 9/15 | Small | Limit-up on 7/30 (2 consecutive), turnover RMB 650mn, net buying of +RMB 36mn on the block-trade disclosure, 3 institutional buy seats | ⚠️ Strong counter-evidence: for the full 7/30 session the grid equipment sector fell −2.46%, with net outflows of RMB 2.664bn and 27 up versus 113 down; at 2 consecutive limit-ups it has entered the relay stage | Watch only |
| 29 | 000636 | 风华高科 (Fenghua Advanced) | ✅ Shenzhen main board | MLCC | B | 30 | Samsung Electro-Mechanics raises MLCC prices 30%, effective 8/1, with existing orders repriced retroactively | Passive components | Industry-trend positive (but already rejected once by the market) | 8/15 | Medium | +6.29% on 7/30 but with 6 failed limits, turnover RMB 13.157bn, an extremely high turnover ratio | ⚠️ The components sector fell −6.58% on 7/30 with 4 up, 58 down and 7 limit-downs — the price-increase letter failed on yesterday's tape; the 7/29 block-trade disclosure showed northbound net selling of RMB 347mn and institutional net selling of about RMB 201mn (a lesson already back-filled in yesterday's recap) | Watch only |
| 30 | 600667 | 太极实业 (Taiji Industry) | ✅ Shanghai main board | "Memory" | C | 22 | The market classifies it as a memory-price beneficiary via Hitech Semiconductor (its DRAM packaging-and-testing JV with SK Hynix) | — | ⭐Pure concept mapping (with hard counter-evidence from company filings) | ⭐7/15 | Negative: the share price went from RMB 12.79 on 5/29 → RMB 32.10 on 7/1 (+151% in 26 trading days) → RMB 15.66 on 7/30 (−51.2%) | Closed limit-down on a one-line bar on 7/30, turnover RMB 5.245bn, turnover ratio 15.48% | ⚠️ See item 1 of the Pass list in section 6 — this is the most important falsification in this report | Pass |
3.1 Three pairs that must be read side by side (the most informative part of this section)
| Comparison | A | B | Note |
|---|---|---|---|
| Both packaging names hit the limit-down, but money moved in opposite directions | 长电科技 (JCET, 600584): limit-down, with the top-five buying seats net buying +RMB 317mn (single-day basis, 3.0% of turnover) | 通富微电 (TFME, 002156): limit-down, with the top-five buying seats net selling −RMB 477mn on the single-day basis (5.2% of turnover) / −RMB 3.2bn on the three-consecutive-day basis, and 4 institutional sell seats | TFME's pre-announced growth (+288%~337%) is far above JCET's (+63%~102%), but the money went the other way. Under the methodology established in yesterday's recap ("go by observable falsification conditions, not by ranking news strength"), on the comparable single-day basis JCET's flow direction is better than TFME's (⚠️ but the two net amounts are only 3.0% and 5.2% of that day's turnover, making this a weak rather than a strong signal, insufficient on its own to support the ranking) |
| Both are "memory beneficiaries", but one is direct and the other is diluted to 15% | 兆易创新 (GigaDevice, 603986): memory is 71.3% of revenue and 76.0% of gross profit, H1 net profit attributable to parent +1099% | 深科技 (Shenzhen Kaifa, 000021): Peyton's two bases together contributed only about RMB 175mn = 15% of net profit attributable to parent in 2025, at a net margin of 3.8%; 41% of gross profit comes from metering smart terminals unrelated to memory | The same "memory price increase" news hits the two income statements with 4–5x different force. This is exactly why "concept ≠ beneficiary" must be graded |
| Both are "main-board optical module leaders", but their overseas exposure runs opposite ways | 光迅科技 (Accelink, 002281): overseas revenue share 34.3%→28.3%→26.7%, down two years running; the company's own pre-announcement attributes growth to "domestic cloud service providers" | 中际旭创 (Innolight, 300308), 新易盛 (Eoptolink, 300502): North American CSP revenue is the overwhelming majority (ChiNext, observation only) | Treating Accelink as "the main-board version of Innolight" is a mismatch. If Accelink rises today, the driver is sector beta and sentiment repair, not its own exposure to AWS/Azure |
3.2 Watch zone (non-main-board, not primary coverage, registered only)
| Code | Name | Board | 7/30 performance | Reason for registration | Label |
|---|---|---|---|---|---|
| 300308 | 中际旭创 (Innolight) | ChiNext | −9.15%, turnover of RMB 59.773bn, the heaviest since 2026-06-05 | The global optical module leader, with the most direct overnight logic; but on the evening of 7/30 it filed that the controlling shareholder and parties acting in concert cut their stake from 17.99% to 16.53%, selling 6.21 million shares | Watch only (non-main-board + shareholder selling) |
| 300502 | 新易盛 (Eoptolink) | ChiNext | −11.90%, turnover RMB 36.131bn | H1 pre-announcement +77.6%~102.9%; PE (TTM) 48.2x and ROE 72.75%, the best on the chain | Watch |
| 300394 | 天孚通信 (T&S Communications) | ChiNext | −12.08% | Gross margin of 53.96%, the highest on the chain | Watch |
| 300476 | 胜宏科技 (Victory Giant) | ChiNext | −10.54% | An AI accelerator card / advanced HDI route; 2025 net profit attributable to parent RMB 4.31bn > WUS's RMB 3.82bn, ROE 35.56% | Watch |
| 301308 | 江波龙 (Longsys) | ChiNext | +2.78% (closed green against the trend) | Memory modules; the highest H1 profit growth of the memory big four (media basis, on the order of 622x) | Watch |
| 688525 | 佰维存储 (Biwin) | STAR Market | +3.76% (closed green against the trend) | Memory modules | Watch |
| 688008 | 澜起科技 (Montage Technology) | STAR Market | −5.17% | Memory interface chips, a direct DDR5 beneficiary | Watch |
| 688498 | 源杰科技 (Yuanjie Semiconductor) | STAR Market | −17.20%, net buying of +RMB 125mn on the block-trade disclosure (4 institutional sell seats) | Optical chips; H1 pre-announcement +1250.9%, the highest on the chain | Watch |
| 688825 | 长鑫科技 (CXMT) | STAR Market | — | ⚠️ On its 7/27 listing debut it opened at RMB 49.50, up 471.59%, with a total market cap of RMB 3.31 trillion, overtaking ICBC as the largest A-share by market cap; total funds raised of RMB 57.919bn (RMB 66.607bn with the over-allotment fully exercised; the RMB 29.5bn planned raise is the prospectus basis, and the two must be distinguished), the largest IPO in STAR Market history. It is the key variable for understanding July's memory-chain collapse (see the GigaDevice entry in section 5) | Watch (must be tracked, because it is the pricing anchor for the entire memory chain) |
| 688981 | 中芯国际 (SMIC) | STAR Market | −6.04% | The counterpart to TSMC +7.63% and UMC +10.87% | Watch |
| 688048 | 长光华芯 (Everbright Photonics) | STAR Market | −15.17% on 7/30 (yesterday's recap basis) | Optical chips, in the same branch as Jiangsu Yongding but moving the other way | Watch |
4. Single-stock scoring model (100 points total, with item-by-item detail)
Scoring dimensions and caps: authority of the news source 0–15 / directness of the catalyst 0–20 / earnings elasticity 0–15 / industry position and fundamentals 0–15 (referencing SEPA: revenue · net profit · gross margin · cash flow · leverage · barriers · whether it leads its niche) / expectation gap 0–10 / theme persistence 0–10 / A-share trading attributes 0–10 / risk deductions 0 to −15.
| Rank | Name (code) | News-source authority /15 | Catalyst directness /20 | Earnings elasticity /15 | Industry position and fundamentals /15 | Expectation gap /10 | Theme persistence /10 | A-share trading attributes /10 | Risk deduction | Total |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 兆易创新 (GigaDevice, 603986) | 14 | 18 | 15 | 11 | 8 | 8 | 9 | −12 | 71 |
| 2 | 沪电股份 (WUS, 002463) | 14 | 15 | 13 | 12.5 | 7 (cut from 9: the same pre-announcement has already been ignored by the market three times in July) | 8 | 8 | −9.5 | 68 |
| 3 | 长电科技 (JCET, 600584) | 14 | 15 | 11 | 12 | 7 (same decay) | 8 | 9 | −9 | 67 |
| 4 | 工业富联 (FII, 601138) | 14 | 16 (cut from 19, see the R8 scope correction: with no customer disclosure to support it, it cannot be classified as a "direct order positive") | 12 | 12 | 6 | 8 | 6 | −10 | 64 |
| 5 | 生益科技 (Shengyi, 600183) | 13 | 14 | 13 | 11 | 8 | 7 | 7 | −11 | 62 |
| 6 | 华天科技 (TSHT, 002185) | 13 | 14 | 13 | 10.5 | 8 | 7 | 7 | −11.5 | 61 |
| 7 | 通富微电 (TFME, 002156) | 13 | 15 | 14 | 11 | 7 (same decay) | 7 | 7 | −13 | 61 |
| 8 | 深南电路 (SCC, 002916) | 13 | 14 | 11 | 11 | 7 | 7 | 7 | −10 | 60 |
| 9 | 紫光股份 (Unisplendour, 000938) | 12 | 13 | 12 | 10.5 | 8 | 7 | 8 | −11.5 | 59 |
| 10 | 德明利 (Demingli, 001309) | 13 | 15 | 15 | 9 | 8 | 6 | 7 | −15 | 58 |
| 11 | 浪潮信息 (Inspur, 000977) | 12 | 9 | 14 | 10 | 8 | 7 | 7 | −10 | 57 |
| 12 | 东山精密 (DSBJ, 002384) | 12 | 12 | 12 | 10 | 6 | 7 | 8 | −11 | 56 |
| 13 | 深科技 (Shenzhen Kaifa, 000021) | 12 | 9 | 10 | 10 | 6 | 7 | 7 | −7 | 54 |
| 14 | 光迅科技 (Accelink, 002281) | 12 | 9 | 10 | 9 | 6 | 7 | 7 | −7 | 53 |
| 26 | 永鼎股份 (Jiangsu Yongding, 600105) | 13 | 19 | 9 | 7 | 2 | 6 | 5 | −23 → truncated to −15 | 38 |
| 30 | 太极实业 (Taiji Industry, 600667) | 8 | 2 | 3 | 7 | 1 | 3 | 4 | −6 | 22 |
Scoring notes (three points of scope that must be spelled out):
- "Catalyst directness" is the item with the widest dispersion this time. Jiangsu Yongding leads with 19 points (a named-amount order at its own subsidiary, the only "direct order positive" on the list with a filing-level named amount), while FII was cut from 19 in the first draft to 16 (closest to direct benefit on the financial side, but with no customer disclosure to support it); Inspur, Shenzhen Kaifa and Accelink all get only 9 points — respectively because "it is on the U.S. Entity List and overseas orders are unaddressable", "Peyton is only 15% of net profit attributable to parent and works on a packaging service-fee model", and "overseas revenue share has fallen for two straight years to 26.7%, with the company itself attributing growth to domestic cloud providers". Those three 9-point scores are the part of this report we least want overlooked.
- The "risk deduction" is capped at −15, and Jiangsu Yongding's actual deductions exceed the cap and are truncated to −15 (triple discount + governance sanction + valuation + a 70x single-day front-run). The truncation is itself a signal.
- Taiji Industry's "catalyst directness" gets 2 points rather than 0, because the path "SK Hynix expands capacity → Hitech makes additional investment → the agreed-return base rises" is theoretically valid; but as of today there is no expansion filing at all, 2026Q1 construction in progress is only RMB 47mn, and net fixed assets have declined year after year, so it gets only 2.
5. Detailed analysis of the top 10 stocks
Each name is laid out in the same six parts: related news → catalyst logic → industry branch stage → fundamental verification → industry position → technicals and sentiment → final judgment. Every fundamental conclusion marked ⭐ comes from this report's line-by-line verification of the original annual-report / filing PDFs using the
fundamentals-analystsub-agent; anything that could not be found is written as "no reliable data available".
51 兆易创新603986Shanghai main board · total score 71 · priority deep-dive · GigaDevice
① Related news
- 2026-07-31 early morning (CNBC): Amazon's 2026 capital expenditure $200bn → $220 billion, attributed to "rising memory costs". Link1
- 2026-07-30 23:27 (Sina Finance): SanDisk +22%, Micron +15%, Western Digital +15%, SK Hynix +15%, Seagate +12%. Link3
- 2026-07-29 20:10 (company filings 2026-061/062/063): Chairman Zhu Yiming proposed a buyback of RMB 1–2 billion, all for cancellation, pledged not to sell for 12 months, and plans to increase his stake by no less than RMB 1 billion.
- 2026-07-10 (filing 2026-054): H1 pre-announcement of revenue about RMB 11.5bn (+177%), net profit attributable to parent about RMB 6.9bn (+1099%), ex-non-recurring about RMB 4.85bn (+791%). Link31
- TrendForce 2026-06-16: NOR Flash contract prices rose 100%–120% cumulatively in H1 2026, SLC NAND 130%–150%; high-capacity NOR still has 60%–65% of room in H2, but low- and mid-capacity price momentum is slowing and partly entering consolidation. Link32
② Catalyst logic (direct or indirect? does it affect revenue or sentiment?) For the "memory super-cycle" it is a direct order positive, affecting revenue and gross margin rather than sentiment — memory chips are 71.3% of its revenue and 76.0% of gross profit, price increases flow straight into the income statement, and this is already realized in the financial data: 2026Q1 gross margin 57.08% (40.22% for full-year 2025), net margin 35.16%, and a Q2 single-quarter ex-non-recurring net margin of about 47% backed out from the pre-announcement. But with respect to the specific catalyst of "Amazon raising capex to $220bn", it is only an industry-trend positive — GigaDevice does not supply Amazon with a single chip. The real chain is: cloud capex ↑ → manufacturers push all capacity toward HBM/DDR5/high-layer 3D NAND → mature-node capacity is drained → structural shortages in niche DDR4/DDR3, NOR Flash and SLC NAND → price increases. This chain has already been verified once by the company's own financial statements, so its credibility sits at the top of the "industry-trend positive" tier.
③ Industry branch stage The first attempt at a rebound after the ebb, not a launch and not a climax. The memory branch completed its climax in June (GigaDevice closed at RMB 840 on 6/29) and its ebb in July (−55.8%), with CXMT's 7/27 listing accelerating the ebb. There is currently no limit-up echelon and no confirmed heavyweight — Demingli was the only memory stock to hit the limit-up across the whole market on 7/30, and GigaDevice itself rose only 1.94%. By the standard established in yesterday's recap, "a lone name does not constitute a branch", so today the second and third main-board memory names must follow before the branch is established.
④ Fundamental verification ⭐11/15
- Does the core business match? Yes. Memory chips RMB 6.566bn (71.3% of revenue / 76.0% of gross profit), MCUs RMB 1.910bn (20.8%), analog 3.6%, sensors 4.2%.
- Revenue and net profit trend? The strong-cycle character is extremely clear: 2023 revenue −29.1%, net profit −92.2%; 2024 +27.7%/+584.2%; 2025 +25.1%/+49.5%; 2026Q1 revenue +119.4%, net profit attributable to parent +522.8%.
- Gross margin, cash flow, leverage? Gross margin 34.42% (2023) → 40.22% (2025) → 57.08% (2026Q1); OCF/net profit attributable to parent >1 for 4 consecutive periods (1.22x in 2026Q1, with OCF +430.9% YoY); debt-to-asset ratio 8.13%, both short-term and long-term borrowings at 0, cash of RMB 14.694bn — a net-cash company. Inventory is only RMB 3.401bn, just +10.9% versus end-2025, so there is no abnormal stockpiling during a price-increase cycle; contract liabilities of RMB 360mn, +283.1% YoY, and a surge in customer prepayments is the hardest leading indicator of shortage.
- Real orders/customers/certifications/capacity? In DRAM it is deeply tied to CXMT through "exclusive distribution + exclusive foundry", and Chairman Zhu Yiming is also chairman of CXMT — this is both a capacity moat and a governance risk of related-party transactions and conflicts of interest.
- Niche leader? Top three globally and number one domestically in NOR Flash; number one domestically in 32-bit Arm general-purpose MCUs (for 8+ consecutive years). ⚠️ Specific market-share figures conflict across data providers (one says 18.5% and second globally in NOR, another 23.2% and third), so this report writes only "top three globally / number one domestically" and gives no specific percentage. Its niche DRAM global share is about 1.7% and seventh (2024, second-hand basis, to be verified) — a small share is both a source of elasticity and a weakness in bargaining power.
- Items with no reliable data available: the company has never split NOR Flash and DRAM revenue in its periodic reports; the business breakdown only goes down to the "memory chips" level.
⑤ Industry position: the heavyweight of the branch (not the leader — that is the newly listed CXMT; nor the most elastic — that is Demingli/Longsys). It is the combination of "the most solid fundamentals + main board + buyback support", which makes it a good anchor for observing whether the branch is established.
⑥ Technicals and sentiment
- Main board ±10%; limit-down on 7/28, −6.81% on 7/29, +1.94% on 7/30.
- 7/30 turnover RMB 35.484bn, turnover ratio 14.53% (equivalent to 14.30% of free-float market cap) — this is a violent collision of chips, not "inflows".
- −55.8% from the 6/29 high, with over RMB 320 billion of market cap evaporating in a month.
- No limit-up, no echelon, no one-line board; there is no "one-day-wonder" risk here, the risk is a "failed rebound".
⑦ Final judgment: priority deep-dive, but treat it as the "branch thermometer" rather than an elasticity name. Reason: it is the only name on this list that simultaneously has "direct memory benefit + Shanghai main board + H1 results already filed + zero interest-bearing debt + a chairman's buyback proposal". Three deductions must be kept in mind at all times: ① governance — RMB 4.4 billion cashed out (at an average of about RMB 396) versus a buyback ceiling of RMB 2 billion plus a stake-increase floor of RMB 1 billion, and the stake increase cannot start until 2026-12-13 (the six-month short-swing trading restriction), so it provides no actual buying for the next 4.5 months; ② the CXMT shadow-stock premium was dismantled in one stroke by the real company's 7/27 listing, which is structural and irreversible; ③ RMB 2.05bn of the RMB 6.9bn H1 net profit attributable to parent is a non-recurring fair-value change, leaving only RMB 4.85bn ex-non-recurring — anyone citing RMB 6.9bn must also give RMB 4.85bn. Valuation scope correction: the PE (TTM) of 90.59x shown on quote screens is distorted (it uses trailing net profit of RMB 2.875bn = FY2025 − Q1'25 + Q1'26, which does not reflect the RMB 6.9bn H1 already filed on 7/10). Recomputed on a basis including the interim pre-announcement: TTM net profit attributable to parent RMB 7.973bn → PE 32.7x; TTM ex-non-recurring RMB 5.775bn → PE 45.1x. But PB of 10.30x is at the 90.1st percentile of the past five years and total market cap at the 97.8th percentile of the past ten years — for a strongly cyclical stock, PB and market-cap percentiles are more reliable than PE percentiles; using "PE is only at the 21st percentile" to argue that it is cheap is the classic valuation trap for cyclicals.
52 沪电股份002463Shenzhen main board · total score 70 · priority deep-dive · WUS Printed Circuit
① Related news
- Amazon AWS revenue +37% (fastest in 18 quarters), backlog $496 billion (prior quarter $364 billion), 2026 capex $220 billion; Jassy: "even so, we won't be able to meet all demand in 2026, and the same goes for 2027". CNBC1
- Microsoft +15.5% (largest single-day gain in 18 years), Azure +43% (fastest in four years), commercial RPO +84%. Sina Finance2
- The company's 2026-07-14 pre-announcement: H1 net profit attributable to parent RMB 2.830–3.000 billion, +68.17%~+78.28%; ex-non-recurring RMB 2.730–2.880 billion. The filing attributes this to "structural demand in application areas such as high-speed switches, AI servers, high-performance computing and smart vehicles ... the Thailand subsidiary turned profitable on a single-quarter basis in Q2 2026". CFI33
② Catalyst logic An industry-trend positive at the top of that tier, but not a direct order positive. The company has never disclosed any customer name (the annual report lists only "customers one through five", and the H-share prospectus is likewise unnamed), and it is Tier-2 to the cloud vendors (CSP → switch makers/server ODMs → WUS). Any statement that "AWS's $220bn capex directly benefits WUS's orders" is an inference, not a public fact. But it is far stronger than an ordinary "indirect supply-chain positive", on five strands of evidence: ① the revenue mix is directly tied to it — data communications is 77.4%, of which high-speed switches and supporting routers are 43.1% and AI servers plus HPC 15.9%, so both destinations of cloud capex (compute + scale-out networking) land on its core business; ② 85.7% of sales are exports, the same coordinate system as "overseas cloud capex"; ③ in its official 7/14 filing the company named the transmission chain itself (filing-level evidence, not sell-side inference); ④ demand has already been realized as capacity utilization — all five bases ran at 92.5%–99.7% in 2026Q1 (Kunshan Qingsong >99%, Huli Micro >99%, Huangshi 99.7%, Thailand 99.1%, Changzhou Jintan 92.5%), and the annual report acknowledges "tight high-end capacity" and "high-end capacity is allocated to core customers first"; ⑤ the 2025 annual report explicitly names CSPs' sharp increase in capital spending as "bringing strong structural growth momentum to the PCB industry".
③ Industry branch stage: the AI PCB branch is in the divergence phase of "earnings already verified, share price ebbing". On 7/14 the whole industry released pre-announcements together (WUS, SCC, Shengyi and DSBJ on the same day or the next), and share prices then accelerated downward — this is the branch's "good news realized as distribution" phase. If it rebounds today, that is the first attempt at stabilization after the ebb, and it needs sector-level rather than single-stock signals.
④ Fundamental verification ⭐12.5/15 (highest on the list)
- Does the core business match? Highly. Datacom PCB RMB 14.656bn (77.4% of revenue, gross margin 39.68%), smart vehicles RMB 3.045bn (16.1%, gross margin 22.84%). Datacom gross margin is 16.84 percentage points above automotive boards, which is the mechanism behind "product-mix optimization = rising profitability".
- Revenue and net profit trend? 2023 → 2025 revenue RMB 8.938bn → 13.342bn → 18.945bn; net profit attributable to parent RMB 1.513bn → 2.587bn → 3.822bn; 2026Q1 revenue +53.9%, net profit attributable to parent +62.9%; backing out Q2 alone gives net profit attributable to parent of about RMB 1.673bn, +81.6% YoY and +34.7% QoQ, the steepest sequential jump in four years.
- Gross margin, cash flow, leverage? Gross margin 31.17% → 35.48%; ROE 16.84% → 28.57% (net margin/turnover/leverage resonating together, not simply added leverage); full-year 2025 OCF of RMB 3.872bn ≈ net profit attributable to parent of RMB 3.822bn, a 101% conversion rate — the profit is real cash. Debt-to-asset ratio 46.46%, with net interest-bearing debt of only about RMB 1.306bn against RMB 16.790bn of net assets, so short-term solvency is not a risk.
- ⚠️ But two signals to watch appeared in 2026Q1: OCF/net profit attributable to parent collapsed to 0.41 (1.87 in 2025Q1), inventory days rose from 86.8 to 103.0 and inventory from RMB 4.246bn to 4.913bn. The benign explanation is strategic stockpiling under full production and expansion (the annual report does mention "phased capacity constraints on high-end raw materials" and "implementing strategic safety inventory of key materials"), and accounts-receivable days actually improved from 90.7 to 86.8 (supporting the benign reading); but if the interim OCF conversion rate is still well below 1, the "growth funded by channel stuffing" critique becomes valid.
- Real orders/customers? Not verifiable — the company has never named them. The verifiable downstream types are "high-speed network switch and router makers, AI server and HPC vendors, and smart-driving domain-controller companies". This report names no customers.
- Risks: customer concentration 46.0% (2022) → 53.32% (2025) → 58.4% (2026Q1), so single-customer dependence is rising; the major shareholder 楠梓电子 (Nanzi Electronic) is simultaneously shareholder, customer, supplier and sales agent.
⑤ Industry position: one of the leaders in the domestic AI PCB first tier, but "the only leader" does not hold. Its niche positioning is "strength on the AI networking/switching side" (within datacom: high-speed switches and routers 43.1% > AI servers and HPC 15.9%); 胜宏科技 (Victory Giant) is stronger on the "AI accelerator card / advanced HDI" route (2025 net profit attributable to parent RMB 4.31bn > WUS's RMB 3.82bn, ROE 35.6% > 28.6%, though with higher leverage); 深南电路 (SCC) is more diversified but a notch lower in AI purity. The three are not perfect substitutes. ⚠️ A structural reminder that is easily misread: the market often treats WUS as "AI server PCB", but on its own reporting basis it looks more like the main supplier of 800G/1.6T high-speed switch and router PCB. If a cloud vendor's capex revision goes mainly into in-house ASICs/accelerator cards rather than network upgrades, the marginal elasticity for WUS will be weaker than market intuition suggests.
⑥ Technicals and sentiment: main board ±10%; on 7/30 −8.27%, touching the limit-down at 94.73 intraday before recovering, turnover RMB 7.964bn, turnover ratio 4.22%, volume ratio 1.48. −39.0% from the 7/1 intraday high of 158.20; but still +132.8% year to date in 2026 (from the 2025-07-01 low of 41.48). PE (TTM) 43.19x is at the 82.4th percentile of its full history but only the 57.5th percentile of the past year (showing that a fair part of this round's valuation expansion has been digested by July's drawdown); PB of 11.07x is at the 95.6th percentile of its full history, the tightest of the three valuation measures.
⑦ Final judgment: priority deep-dive, and the cleanest "fundamentals versus share price" divergence on this list. Between the far-above-expectation pre-announcement of 7/14 and touching the limit-down on 7/30, there was no negative company-level filing and no evidence of a fundamental reversal; on the same day Shengyi −7.49%, SCC −9.97%, Victory Giant −10.54% and DSBJ −10.00% all moved together, which is sector-level flow behavior, not a single-stock event. The attributable factors are the valuation side (PB at the 95.6th percentile of its full history) + sector flows + expected dilution from the H-share issue (a re-filing on 2026-06-05 with CICC/HSBC as sponsors; an RMB 17.6 billion expansion plan against RMB 4.392 billion of cash on the books, with issue size, price and timing all unknown).
53 长电科技JCET / 600584Shanghai main board · total score 69 · priority deep-dive
① Related news
- 2026-07-31 04:26 (Sina Finance): ASE Technology +11.79%, UMC +10.87%, TSMC +7.63%, Silicon Motion +21.66%. Link9
- The company's 2026-07-14 pre-announcement: H1 net profit attributable to parent RMB 770–950 million, +63.48%~+101.70%; ex-non-recurring RMB 740–910 million, +68.95%~+107.76%; Q2 pre-announced up over 65% QoQ. East Money34, National Business Daily35
- 2026-07-30 block-trade disclosure (obtained by this report): JCET hit the limit-down, yet showed net buying of +RMB 317 million.
② Catalyst logic An industry-trend positive (strong). Packaging and testing is the shortest hop on the "cloud capex → chips → packaging" chain, and it eats from both the AI and the memory line — a structural advantage that distinguishes it from pure-AI or pure-memory names. Overnight ASE +11.79% is a direct read-across for global packaging and testing, and is more comparable than using Microsoft/Amazon to map onto A-share system makers.
③ Industry branch stage: the packaging branch is in an extreme divergence phase of "all three pre-announced doublings while all three hit the limit-down". On 7/30 JCET, TFME and TSHT hit the limit-down on the same day, while their H1 pre-announcements were +63.5%~101.7%, +288.3%~336.8% and +231.2%~275.3% respectively. No limit-up echelon has appeared in the branch, and today is its first test.
④ Fundamental verification (public financials basis, 12/15)
- Does the core business match? Yes. IC packaging and testing, covering consumer, memory, automotive and compute.
- Earnings? H1 pre-announcement of RMB 770–950 million (+63.48%~+101.70%), with ex-non-recurring growing in line (+68.95%~+107.76%), so growth quality is better than the headline net-profit basis.
- ⚠️ But one point must be stated alongside: its growth rate is the lowest of the packaging big three — the explanation in public reports is that "its business spans consumer, memory and automotive, so compute profits are diluted by the other businesses", whereas TFME's "compute business is pure". That is: JCET is the one of the three with the smallest earnings elasticity, but the most diversified business and the strongest cycle resistance.
- No reliable data available: this report did not obtain JCET's 2026H1 revenue breakdown by business or the share of AI/memory packaging, and makes no estimate.
⑤ Industry position: the common description is number one in domestic packaging and testing and top three globally (in the second tier alongside Amkor and ASE). ⚠️ This report did not obtain a current third-party share report, so the ranking is flagged "to be verified" and used only as a qualitative reference, not as an independent scoring input. It is the heavyweight of this branch rather than an elasticity name.
⑥ Technicals and sentiment
- Limit-down on 7/30 (−10.00%, closing at RMB 64.51), with turnover of RMB 10.647 billion and a turnover ratio of 8.93%.
- The top-five buying seats net bought +RMB 317 million (single-day decline-deviation board basis; ordinary seats were sellers, success rate 41.56%) — with the stock at the limit-down, the top-five buying seats were positive in direction. ⚠️ But it must also be stated: the net buy amount is only 3.0% of that day's RMB 10.647 billion turnover, the disclosure covers only the top five buying and selling seats and is not whole-market flow, and the nature of the buying seats (hot money / institution / Stock Connect) was not obtained and is flagged to be verified, so it is described only as "direction positive" and not characterized as "institutional bottom-fishing".
- −20.2% over the past 5 sessions, with the last 5 sessions at +2.59 / −0.66 / −6.70 / −6.70 / −10.00%, so the declines are widening day by day, a late-stage pattern of accelerating downside.
- −39.5% from its recent-month high.
⑦ Final judgment: priority deep-dive, and in this report's view the name with the clearest falsification conditions. It has four attributes at once: Shanghai main board, limit-down, large turnover, and net buying on the block-trade disclosure. Under the methodology established in yesterday's recap ("go by observable falsification conditions, not by ranking news strength"), the signal "on a limit-down day the top-five buying seats were positive" is more verifiable today than the already-digested news of "a 3x interim pre-announcement" (this is a comparison of verifiability, not of importance — the evidence grade of a company filing is always above a cross-sectional block-trade snapshot) — the latter was already ignored once by the market in July (TFME pre-announced +288%~337% and still hit the limit-down on 7/30 with RMB 3.2 billion of net selling on the disclosure). Risks: a cumulative decline of over 20% across three consecutive sessions (already triggering the disclosure threshold); if it cannot hold up today after a gap higher, yesterday's "net buying" will prove to be trapped buyers rather than effective bottom-fishing.
54 工业富联FII / 601138Shanghai main board · total score 67 · priority deep-dive
① Related news: same as 5.2 (Amazon capex $220bn / AWS +37% / backlog $496bn / Microsoft Azure +43%). The company's 2026H1 pre-announced net profit attributable to parent is RMB 23.4–24.4 billion, +93%~+101%.
② Catalyst logic An industry-trend positive (upper edge); the first draft of this report classified it as a "direct order positive" and cut that after QC — because it too has no public disclosure of customer names or the share of overseas revenue. But its evidence on the financial side is still the strongest on the chain, based on two comparable sets of hard data:
- 2026Q1 revenue of RMB 251.08 billion, +56.5% YoY, net profit attributable to parent of RMB 10.595 billion, +102.5% YoY, with positive operating cash flow (OCF per share +RMB 1.26);
- In the same quarter, Inspur, whose customers are mainly domestic, had revenue −24.3% and OCF per share of −RMB 5.29. Same quarter, same "AI server" label, opposite directions — this is the hardest empirical evidence that "overseas capex ≠ orders for domestic system makers", and the only reason this report lists FII rather than Inspur as a "direct beneficiary".
③ Industry branch stage: it is not on the "oversold" line. It fell only −6.84% on 7/30, clearly less than the rest of the chain (mostly limit-downs), and is −31.8% from its recent-month high. The shareholder structure is steadier, but the elasticity is smaller too.
④ Fundamental verification (public financials basis, 12/15): 2026Q1 gross margin 7.35%, single-quarter ROE 6.18%, positive operating cash flow; H1 pre-announcement +93%~+101%. Its enormous scale is a double-edged sword: full-year revenue is 7x Inspur's, so revenue elasticity is real but share-price elasticity is naturally capped; a gross margin of just 7.35% marks it as a classic scale-driven ODM business.
⑤ Industry position: the common description is global cloud-vendor ODM leader (no current third-party share report obtained, to be verified). The verifiable facts on the financial side are: 2026Q1 revenue RMB 251.08 billion +56.5%, net profit attributable to parent RMB 10.595 billion +102.5%, positive operating cash flow.
⑥ Technicals and sentiment: −6.84% on 7/30, turnover RMB 8.349 billion; no limit-down and no block-trade disclosure. Resilience = no explosive power for an oversold rebound, but it also = if the sector rebounds today, its persistence may be stronger than a one-day snap-back in limit-down names.
⑦ Final judgment: priority deep-dive, positioned as the "branch direction confirmer". If today's AI hardware rebound is real (money genuinely rotating back rather than a sentiment snap-back), FII should strengthen in step and hold into the close; if it opens high and fades while only limit-down names snap back, that says today is merely a technical repair in oversold names, not a return of money. This is the single most important "authenticity detector" this report has designed for today.
55 通富微电TFME / 002156Shenzhen main board · total score 63 · watch closely
① Related news: ASE +11.79%, TSMC +7.63%; the company's H1 pre-announced net profit attributable to parent is RMB 1.6–1.8 billion, +288.26%~+336.80% (NetEase/media aggregation36, Securities Times37).
② Catalyst logic: an industry-trend positive (strong). Media describe its "compute business as pure", making its earnings elasticity the largest — ⚠️ but this report did not obtain a revenue breakdown of its AI/compute business, so "number one in compute purity" is a media paraphrase rather than company disclosure, is flagged "to be verified", and is not used as an independent input to the ranking. The only verifiable fact is that its H1 pre-announced growth (+288.3%~336.8%) is the highest of the packaging big three.
③ Industry branch stage: it sits in the same "3x pre-announcement + limit-down" extreme divergence phase as JCET, but its flow signal is the exact opposite of JCET's.
④ Fundamental verification (11/15): H1 net profit attributable to parent RMB 1.6–1.8 billion, +288%~337%, the fastest of the three. No reliable data available: this report did not obtain the revenue share of its AI/compute business or its customer structure, and makes no estimate.
⑤ Industry position: number two in domestic packaging and testing (the industry ranking is a common description; no current third-party share report was obtained, flagged to be verified); "number one in compute purity" is likewise a media basis, to be verified.
⑥ Technicals and sentiment (this is the key section)
- Limit-down on 7/30, turnover RMB 9.191 billion, turnover ratio 10.61%; a cumulative decline of over 20% across three consecutive sessions.
- ⚠️ Both block-trade disclosure records point to net selling, but they have different time windows and cannot be placed alongside other stocks' single-day figures: net selling of −RMB 3.2 billion on the three-consecutive-day decline-board basis (4 institutional sell seats, success rate 20.00%); net selling of −RMB 477 million on the single-day decline-board basis (3 institutional sell seats, 5.2% of the day's RMB 9.191 billion turnover). The figure comparable with JCET is the latter (−RMB 477 million), not the former.
- Last 5 sessions: +9.77 / +0.04 / −10.00 / −10.00 / −10.00% — three consecutive limit-downs, the steepest decline on this list.
⑦ Final judgment: watch closely, but explicitly ranked behind JCET. There is only one reason, and it is the methodological point this report most wants to stress: TFME's news strength (+288%~337%) is far above JCET's (+63%~102%), but the direction of the top-five buying seats is inverted (−RMB 477 million versus +RMB 317 million on the comparable single-day basis; the three-day cumulative basis of −RMB 3.2 billion cannot be compared directly with JCET's single-day figure). Yesterday's recap used an entire section to prove that "ranking by news strength gets it wrong, ranking by observable falsification conditions gets it right". Today we do not repeat yesterday's mistake: between two stocks in the same branch that both hit the limit-down, we take the one whose flow direction is positive as the priority.
56 生益科技600183Shanghai main board · total score 62 · watch closely · Shengyi
① Related news: upward revisions to AWS/Microsoft capex; the company's H1 pre-announced net profit attributable to parent is RMB 3.099–3.298 billion, +117%~+131%. ② Catalyst logic: an industry-trend positive. Copper-clad laminate (CCL) is the upstream material for PCB, and AI high-frequency high-speed laminates benefit directly from volume growth in high-layer-count boards. ③ Industry branch stage: in step with AI PCB, in the phase where earnings are verified but the share price is ebbing. ④ Fundamental verification (11/15): 2025 revenue of RMB 28.43 billion, the largest of the AI PCB big four, net profit attributable to parent RMB 3.33 billion, gross margin 26.47%, ROE 21.25%, debt-to-asset ratio 42.29%. H1 pre-announcement +117%~131% (with a clear low-base factor). ⑤ Industry position: commonly described as the domestic CCL leader (no current third-party share report obtained, to be verified). But the material end's bargaining power over board makers is weaker than board makers' power over their downstream — a structural disadvantage relative to WUS. ⑥ Technicals and sentiment: −7.49% on 7/30, turnover RMB 8.252 billion; −46.5% from its recent-month high, a deeper drawdown than WUS (−39.0%). PE (TTM) 61.9x, clearly above WUS at 43.2x and Victory Giant at 37.9x. ⑦ Final judgment: watch closely. Its position is "it fell more but it is more expensive" — when both hold at once, it usually means the market applies a bigger discount to its earnings quality. If the AI PCB branch is established today, it is a source of elasticity; if this is only a one-day snap-back, it will also be among the fastest to fall back.
57 华天科技TSHT / 002185Shenzhen main board · total score 61 · watch closely
① Related news: ASE +11.79%; H1 pre-announced net profit attributable to parent RMB 750–850 million, +231.16%~+275.31%. ② Catalyst logic: an industry-trend positive, one of the packaging big three. ③ Industry branch stage: same as 5.3/5.5. ④ Fundamental verification (10.5/15): H1 pre-announcement +231%~275%, the middle of the three. No reliable data available: the AI/memory packaging revenue breakdown was not obtained. ⑤ Industry position: commonly described as number three in domestic packaging and testing (no current third-party share report obtained, to be verified). ⑥ Technicals and sentiment: limit-down on 7/30, turnover RMB 6.187 billion, turnover ratio 12.03%; net selling of −RMB 203 million on the block-trade disclosure, with 4 institutional sell seats (success rate only 6.77%); −41.9% from its recent-month high. ⑦ Final judgment: watch closely, ranked below JCET and level with TFME. Flow direction is negative (−RMB 203 million) and the institutional sell seats' success rate is only 6.77% (meaning those seats' recent selling decisions have had a very low historical hit rate; this indicator must be read with reverse caution and must not be treated as a positive). Within the packaging big three, this report ranks them: JCET (positive flow) > TFME (largest elasticity but −RMB 3.2bn of flow) ≈ TSHT (small negative flow).
58 深南电路SCC / 002916Shenzhen main board · total score 60 · watch closely
① Related news: AWS/Microsoft capex; H1 pre-announced net profit attributable to parent RMB 2.100–2.300 billion, +54.4%~+69.1% (2026-07-14). ② Catalyst logic: an industry-trend positive. An integrated interconnect platform of PCB + packaging substrates + electronic assembly. ③ Industry branch stage: same as AI PCB. ④ Fundamental verification (11/15): 2025 revenue RMB 23.65 billion, net profit attributable to parent RMB 3.28 billion, gross margin 28.32%, ROE 20.79%, debt-to-asset ratio 43.82%. Its H1 pre-announced growth of +54.4%~69.1% is the lowest of the AI PCB big four. ⑤ Industry position: the strongest in integrated capability (packaging substrates are a scarce asset), but its AI purity is below WUS and Victory Giant — a gross margin of 28.32% versus WUS at 35.48% and Victory Giant at 35.22% is 7 percentage points lower, and that gap is the price of AI purity. ⑥ Technicals and sentiment: −9.97% on 7/30 (close to the limit-down), turnover RMB 4.078 billion; PE (TTM) 53.3x, above WUS's 43.2x on lower growth. ⑦ Final judgment: watch closely. The combination of "lowest growth + higher valuation" places it behind WUS in the internal AI PCB ranking; but the packaging substrate business gives it better downside protection.
59 紫光股份000938Shenzhen main board · total score 59 · watch closely · Unisplendour
① Related news: Microsoft Azure +43%, AWS +37%; the Politburo meeting's "deeply implement the 'AI+' initiative" and "solidly advance the planning and construction of the 'six networks' (including the compute network)"; H1 pre-announced net profit attributable to parent RMB 1.91–2.32 billion, +83.5%~+122.9%. ② Catalyst logic: an industry-trend positive. H3C = networking equipment (switches/routers) plus servers, a two-engine model, and the networking business is the pillar of its gross margin. Azure +43% and the AWS backlog of $496 billion point to two lines, "compute + scale-out networking", and the networking line matches H3C's product mix. ③ Industry branch stage: ebbing in step with the systems/networking branch, −25.2% from its recent-month high in July (one of the smallest drawdowns among the AI hardware names on this list). ④ Fundamental verification (10.5/15): 2026Q1 revenue RMB 27.98 billion (+34.6%), net profit attributable to parent RMB 788 million (+126.1%), gross margin 12.88%, the highest of the four systems/networking makers (Inspur 6.64%, FII 7.35%, Sugon 26.56% — Sugon is high but only 1/9 of Unisplendour's scale); OCF per share −RMB 1.08 (negative, but far smaller in magnitude than Inspur's −RMB 5.29). ⑤ Industry position: commonly described as first-tier in domestic enterprise networking equipment (H3C) and top three domestically in servers (no current third-party share report obtained, to be verified). ⑥ Technicals and sentiment (this is the main reason it made the list)
- Limit-down on 7/30 (−9.99%), turnover RMB 8.755 billion, turnover ratio 8.88%;
- Net buying of +RMB 548 million on the block-trade disclosure — the largest of all limit-down names on 7/30 (1 institutional sell seat, success rate 39.06%). ⑦ Final judgment: watch closely, paired with JCET as a "limit-down + net inflow" dual-signal name. The difference: JCET corresponds to the direct read-across from ASE +11.79% overnight, while Unisplendour corresponds to the indirect read-across from Microsoft Azure +43%; JCET's news mapping is harder, while Unisplendour's shallower drawdown (−25.2%) means thinner trapped supply overhead. The two can serve as a cross-check on whether the AI hardware rebound is real.
510 德明利001309Shenzhen main board · total score 58 · watch closely · Demingli
① Related news
- Overnight Silicon Motion (memory controllers) +21.66%, SanDisk +22%, Micron +15%, SK Hynix +15%.
- The company's 2026-07-15 pre-announcement: H1 revenue RMB 16–18 billion (+289.39%~+338.06%), net profit attributable to parent RMB 5.7–6.5 billion (a swing to profit, +4932.74%~+5611.02%), ex-non-recurring RMB 5.649–6.449 billion, EPS RMB 25.35–28.91; non-recurring gains and losses of about RMB 50.98 million, share-based payment expense of about RMB 160 million.
- ⚠️ TrendForce (2026-07-03/07-06): 2026Q3 DRAM contract prices are estimated at only +13%
+18% QoQ and NAND at +10%+15%, versus conventional DRAM at +58%~63% and NAND Flash at +70%~75% in 2026Q2. Sina Tech38
② Catalyst logic (the passage in this report that most needs precise wording) An industry-trend positive, and a "slope-sensitive" one — not a direct order positive.
- Why it is not a direct order positive: SanDisk, Micron and SK Hynix that rallied are upstream manufacturers, and Demingli is their customer, not their supplier; a rally in manufacturers' shares generates no orders for Demingli. Amazon's $220 billion capex attributed to "rising memory costs" corresponds to orders for the manufacturers and HBM makers, not for module makers. This report found no order filing at Demingli related to it — no reliable data available.
- Why it is also not merely "indirect supply chain": memory prices are the direct determinant of Demingli's selling prices and the fair value of its inventory, the transmission is first-order and lands in the current period's accounts, and it has already been quantitatively confirmed by a 2026Q1 gross margin of 57.42% and an H1 pre-announcement of +5000%.
- ⚠️ The core mechanism (this must be spelled out): Demingli benefits from the "acceleration" of memory price increases, not from the price "level". A module maker's costs are wafers bought months earlier — when prices accelerate, selling prices run ahead of costs and gross profit explodes; once the rate of increase converges, costs catch up with selling prices and gross margin falls back. Single-quarter gross margin evidence: −1.89% (2023Q2) → 37.23% (2024Q1) → 24.11% → 14.14% → 1.32% (2024Q4) → 4.69% (2025Q2) → 27.14% (2025Q4) → 57.42% (2026Q1). In the previous cycle gross margin collapsed from 37.23% to 1.32% in 4 quarters while revenue was still growing — revenue growth cannot cover for a gross-margin collapse.
③ Industry branch stage: the first rebound after a 60% collapse, and a high-beta performance within a sector-wide rebound, so it carries no "independent fundamental" meaning — on the same day, 7/30, GigaDevice +1.94%, Longsys +2.78% and Biwin +3.76% all closed green. The four triggers of July's collapse have still not been falsified: on 7/2 Meta announced it would sell surplus AI compute externally (breaking the "compute is absolutely scarce" narrative); on 7/13 Morgan Stanley warned of a triple peak in memory prices/inventory/profits and explicitly noted that "inventory levels at module makers and distributors are rising" and that phone and PC customers were starting to cut orders; UBS calculated that if cloud capex slows, DRAM demand growth could fall from 36% to 18%; and HBM4 has not shipped at scale, leaving ASP below expectations.
④ Fundamental verification ⭐9/15
- Does the core business match? Yes. In 2025: SSDs RMB 4.582bn (42.5%, gross margin 12.69%), embedded storage RMB 3.663bn (34.0%, 11.86%), removable storage RMB 1.474bn (13.7%, 20.74%), memory modules RMB 1.051bn (9.7%, 25.91%). A counter-intuitive point: the two largest lines (SSD and embedded) happen to have the lowest gross margins. The mix transition is real — removable storage fell from 58.3% of revenue in 2023 to 13.7%, so it is no longer a "USB-drive maker".
- ⚠️ Q2 has already turned (backed out from the company's own pre-announcement, not company disclosure): Q1 actuals were revenue RMB 7.538bn, net profit attributable to parent RMB 3.346bn, net margin 44.4%; backing out the H1 pre-announcement gives Q2 revenue of RMB 8.46–10.46bn (+12.2%
+38.8% QoQ), net profit attributable to parent of RMB 2.35–3.15bn (−29.7%−5.7% QoQ) and a net margin falling to 27.8%~30.1%. That is: in a Q2 when memory prices were still rising, revenue grew sequentially while profit fell sequentially — the profit peak may already have occurred in 2026Q1. - ⚠️ Earnings quality (the point that most needs to be called out): cumulative net profit attributable to parent from 2023 to 2026Q1 was RMB 4.41bn, against cumulative operating cash flow of −RMB 4.76bn (2023 −RMB 1.015bn / 2024 −1.263bn / 2025 −2.241bn / 2026Q1 −0.241bn). All the profit has settled into inventory: inventory went from RMB 4.436bn at end-2024 → RMB 7.058bn at end-2025 → RMB 12.192bn at end-2026Q1 (up RMB 5.134bn in a single quarter). Inventory/net assets of 1.84x and inventory/total assets of 66.3% are the highest among peers (Longsys 1.47x/50.0%, Biwin 1.42x/49.3%, GigaDevice 0.13x/12.2%). Debt-to-asset ratio 64.0%, net interest-bearing debt about RMB 3.94bn. The business model in one sentence: it is essentially a long position in RMB 12.2 billion of NAND/DRAM inventory levered with a 64% debt ratio, wrapped in a module-manufacturing shell. That explains both why its elasticity is the largest and why its risk is the largest.
- Niche leader? No. Cross-comparison of H1 pre-announcements: number 2 by revenue (Longsys RMB 22–25bn > Demingli RMB 16–18bn > Biwin RMB 15–16bn), number 4 by profit (Longsys RMB 9.2–11.0bn > Biwin RMB 7.0–7.5bn > GigaDevice about RMB 6.9bn > Demingli RMB 5.7–6.5bn), with an implied net margin of 35.9%, the lowest (GigaDevice 60.0% > Biwin 46.8% > Longsys 43.0%). Revenue is about 10% above Biwin's while profit is about 16% below Biwin's — showing that in the same price-increase environment, its product mix and ability to pass through costs are weaker than Biwin's. The accurate description is "one of the main domestic players in consumer removable storage + a second-tier module maker", not "the niche leader in memory modules".
- In-house development progress (real and verifiable): it has launched a self-developed PCIe/SATA dual-mode enterprise SSD controller and mass-produced several QLC-based embedded storage products. But the R&D expense ratio is falling: 6.08% (2023) → 4.25% (2024) → 2.71% (2025) → 2.23% (2026Q1). The revenue share of enterprise storage, its customer structure and controller shipments have never been broken out by the company — no reliable data available.
⑤ Industry position: one of the module big three, first in elasticity, second in scale, fourth in profit, last in net margin.
⑥ Technicals and sentiment (position awareness must be corrected)
- RMB 390.04 is not a high. From the 6/26 intraday high of RMB 980.00 to the 7/29 intraday low of RMB 334.00 is −65.9% in 22 trading days, including 7 limit-downs across 12 days (with three consecutive one-line boards on 7/15–7/17). The 7/30 close of RMB 390.04 is still −60.2% from the high.
- 7/30: the semiconductor sector fell −7.08% and it was the only limit-up among 176 decliners and 7 gainers, first sealed at 13:22 with 1 failed limit, turnover RMB 11.366 billion, turnover ratio 18.57% (free-float basis).
- ⚠️ Net selling of −RMB 706 million on the block-trade disclosure on the limit-up day (buys RMB 1.249bn / sells RMB 1.955bn / disclosed turnover RMB 3.203bn, 28.19% of the day's total), with the interpretation column reading "3 institutional buy seats, success rate 3.33%".
- The shareholder structure has deteriorated markedly: shareholder accounts went from 55,715 on 2026-04-10 to 124,036 on 2026-07-20 (+122.6%), with average holding value per account falling from RMB 1.82 million to RMB 880,000; over the past 8 trading days the turnover ratio has run at 11%~25%, so cumulative turnover far exceeds 100% and the chips at this level have already changed hands completely.
⑦ Final judgment: watch closely — not upgraded because it was "the only limit-up in semiconductors", and not downgraded because it "fell 60%".
It has the largest earnings elasticity on the whole list (H1 net profit attributable to parent +4933%~5611%) and simultaneously the largest risk deduction (−15, the full amount). Three mutually contradictory pieces of evidence must be laid on the table together: ① annualized H1 PE is only about 7.3x, exactly in line with Longsys at 7.1x and Biwin at 7.1x, neither a discount nor a premium; ② but if gross margin reverts to the 15%20% historical mid-range at the 2024 pace, the normalized PE is 26x33x — the gap between 7.3x and 30x consists entirely of assumptions about how much longer the price-increase cycle can last; ③ net selling of RMB 706 million on the limit-up day + Q2 profit already down sequentially + inventory of RMB 12.2bn at 66.3% of total assets + cumulative operating cash flow of −RMB 4.76bn.
One sentence for today: it is in a state where "most of the earlier gains have been given back, but the current profit level still requires the cycle to continue in order to be sustained". This report makes no valuation characterization of any price level; the comparison of 7.3x with 26–33x above is only a sensitivity note on normalized cyclical earnings, and constitutes neither a valuation conclusion nor a target price.
6. Pass list
Per the methodology required by item 2 of section 6.4 of yesterday's recap, this section adds a "Pass type" column: [Fundamentals-dimension Pass] = the fundamentals do not support this catalyst, but it does not assert that the stock will not rise (a pure sentiment name can rise without any fundamentals); [Dual-dimension Pass] = neither the earnings nor the sentiment dimension supports it, and only this type asserts "do not participate". Yesterday's recap recorded that on 7/30 there were 4 Passes judged wrong (广电电气 / Guangdian Electric, 神雾节能 / Shenwu Energy Saving, 爱丽家居 / Aile Home, 明新旭腾 / Mingxin Xuteng), and all four were pure sentiment names and all four hit the limit-up — which is exactly why the types must be separated.
| Code | Name | Concept | Why it was associated | Reason for Pass | Pass type | Continue to observe? |
|---|---|---|---|---|---|---|
| 600667 | 太极实业 (Taiji Industry) | "Memory price beneficiary" | Its 55%-owned 海太半导体 (Hitech Semiconductor) is a DRAM back-end packaging and testing JV with SK Hynix, and SK Hynix rose +15% overnight | ⭐There is hard counter-evidence in company filings, and this is the most important falsification in this report: Hitech operates on a cost-plus foundry model of "all costs + an agreed return (10% of total investment + excess return)", and in the original text of the company's 2026-06-26 Announcement on Abnormal Share Price Movements (Interim 2026-039) it states that it is "relatively little affected by the industry upcycle" and "does not involve HBM products". Three years of evidence spanning a full cycle: in 2023 (SK Hynix's huge loss), 2024 (SK Hynix's record) and 2025, Hitech's net profit was RMB 219.1 / 237.1 / 239.1 million respectively, a swing of under 10% — price elasticity ≈ 0. Separately: minority interests of RMB 107.5 million ≈ Hitech's RMB 239.1 million × 45%, cross-checking that the consolidation relationship holds — but "consolidated" ≠ "able to transmit price increases". 94% of profit comes from 十一科技 (Shiyi Technology, engineering design), whose order backlog has fallen −47.6% over two years (RMB 43.286bn → 22.663bn); in 2026 there is also RMB 149.84 million of tax late-payment penalties charged directly to profit or loss; and cumulative litigation and arbitration of RMB 920 million equals 10.66% of the most recent audited net assets | Dual-dimension Pass | Yes (the only condition that would overturn this is a Hitech capacity-expansion / additional-investment filing, since the agreed return = total investment × 10%; but 2026Q1 construction in progress is only RMB 47 million and net fixed assets have declined year after year, so public information is currently zero) |
| 300308 | 中际旭创 (Innolight) | Optical module / CPO leader | The overnight AWS backlog of $496bn and Microsoft +15.5% | ① Not main board (ChiNext), so under the iron rules it cannot enter the primary list; ② on the evening of 7/30 it filed that the controlling shareholder and parties acting in concert cut their stake from 17.99% to 16.53%, selling 6.21 million A shares over the period — the branch leader issues a selling filing on the eve of a rebound; ③ turnover of RMB 59.773 billion on 7/30 was the heaviest since 2026-06-05, and a rebound after record volume usually needs a second test to confirm | Fundamentals-dimension Pass (the fundamentals are not weak; what is being passed is "main board" and "the timing of the stake reduction") | Yes (it is still the direction indicator for the branch) |
| 002484 | 江海股份 (Jianghai Capacitor) | Passive components / MLCC | Samsung Electro-Mechanics' 30% MLCC price increase on 8/1 | The product-category mismatch was already verified on 7/30: Jianghai mainly makes aluminum electrolytic capacitors, film capacitors and supercapacitors, not MLCC; −9.61% on 7/30 with the components sector at 4 up, 58 down | Dual-dimension Pass | No |
| 603221 | 爱丽家居 (Aile Home) | Pure sentiment high-flyer | Already 8 consecutive limit-ups | Turnover of only RMB 68 million, turnover ratio 1.37% (yesterday's recap basis), block-trade net buying of about 0 on 7/30; the echelon is completely broken at 5–7 consecutive boards; the risk is not whether it will rise, but that it may be impossible to sell once the board opens | Fundamentals-dimension Pass (explicitly does not assert that it will not rise today — the same type of judgment was wrong 4 times yesterday) | Yes (as a sentiment thermometer) |
| 605068 | 明新旭腾 (Mingxin Xuteng) | Auto interiors / pure sentiment | Already 4 consecutive limit-ups | 1 failed limit on 7/30 with thinning sealing orders (yesterday's recap basis); no news-flow support | Fundamentals-dimension Pass | Yes |
| 003032 | 传智教育 (Chuanzhi Education) | AI applications / education | 4 consecutive limit-ups; the ChinaJoy theme "Travel with AI" | An AI business mapping with no filing support; block-trade net buying of only +RMB 16 million on 7/30 with 1 institutional buy seat; the education sector rose 2.78% on 7/30 but had net outflows of RMB 92 million — price up with net flows negative, the "risk aversion rather than allocation" pattern defined in yesterday's recap | Fundamentals-dimension Pass | Yes |
| 600162 | 香江控股 (Xiangjiang Holdings) | Property / the Politburo's "stabilize the property market" | Limit-up at 09:34 on 7/30 | The policy wording is "stabilize the property market" (more concise than the April meeting's wording), names no company and involves no specific funding arrangement; a pure policy mapping; turnover of RMB 332 million with RMB 33 million of sealing money, so volume is thin | Fundamentals-dimension Pass | Yes (to observe whether the property branch produces a sector-level advance) |
| 601616 | 广电电气 (Guangdian Electric) | Grid equipment / the "six networks" | Limit-up on 7/30 | The "six networks" is wording already put forward at the 2026-04-28 Politburo meeting, so this is a re-confirmation rather than brand-new news; the grid equipment sector fell −2.46% for the full 7/30 session with net outflows of RMB 2.664 billion and 27 up versus 113 down (a full-session figure, not the 22 minutes after the press report); the stock's 7/30 block-trade disclosure showed net selling of −RMB 12 million, i.e. a limit-up with net outflows | Fundamentals-dimension Pass | Yes |
| 000820 | 神雾节能 (Shenwu Energy Saving) | Pure sentiment | Limit-up on 7/30 with turnover of only RMB 22 million | A RMB 3.20 low-priced stock, no news flow, extremely small turnover | Fundamentals-dimension Pass | No |
| 605388 | 均瑶健康 (Junyao Health) | Beverages and dairy / mass consumption | Limit-up on 7/30 (2 consecutive) | The dairy branch already entered its ebb yesterday (limit-up counts fell from 7 on 7/29 to 2 on 7/30); a limit-up on 7/30 but with net outflows (yesterday's recap basis); today it faces the opposing force of "money rotating back into technology" | Fundamentals-dimension Pass | Yes |
| 605179 | 一鸣食品 (Yiming Food) | Dairy | 3 consecutive limit-ups | Same as above; turnover of only RMB 49 million, consecutive boards without volume | Fundamentals-dimension Pass | Yes |
| 002049 | 紫光国微 (Unigroup Guoxin) | Specialty ICs / "memory" | The semiconductor rebound mapping | Not a Pass, but one point must be clarified: it fell only 0.25% on 7/30, the most resilient main-board semiconductor name — resilience means there is no room for an oversold rebound; and its linkage to memory price increases is weak (its core business is specialty ICs and smart security chips) | Not a Pass (listed here only to clarify the association) | Yes |
| — | "宇树科技 (Unitree) concept stocks" | Humanoid robots | Unitree's book-building on 8/5 and subscription on 8/10 | ⚠️ This report gives no list of "Unitree supplier" stocks. Reasons: ① an IPO by itself adds not one cent of revenue to any A-share listed company; ② any supply relationship without company-filing or annual-report customer disclosure support is treated as a pure concept mapping; ③ A-share "Unitree concept" names have repeatedly been falsified after vague replies on interactive platforms; ④ a potential counter-variable: the media-reported "U.S. FCC import restrictions on foreign-made advanced robots" (original FCC text not obtained, to be verified) | Dual-dimension Pass (this refers to the whole class of "unverified supplier mappings", not to any specific company) | Yes (8/5 book-building and 8/10 subscription are real event windows) |
| — | "Gold stocks associated with 赤峰黄金 (Chifeng Jilong Gold)" | Gold | The Sepon project in Laos: metal content 107 → 260 tonnes | Resource volume ≠ reserves ≠ recoverable reserves ≠ current output ≠ profit, and no profit figure can be derived from it; moreover this is a single-stock event at Chifeng Jilong Gold and does not constitute a sector positive. 紫金矿业 (Zijin Mining, 601899) rose +1.71% on 7/30 as a baseline sector move, unrelated to this news and not to be conflated | Fundamentals-dimension Pass (referring to "the other gold stocks associated by extension") | Yes |
7. Ranking within each branch
Role definitions: leader = the branch's pricing anchor | heavyweight = the largest and most solid on fundamentals, determining the branch's persistence | niche leader = number one in a sub-track | elasticity name = the largest earnings/share-price elasticity | catch-up | back row | pure concept = a mapping with no fundamental support.
7.1 Branch one: memory price increases (S grade)
| Rank | Stock | Shanghai/Shenzhen main board | Role | Catalyst directness | Fundamental support | Trading identifiability | Conclusion |
|---|---|---|---|---|---|---|---|
| 1 | 兆易创新 (GigaDevice, 603986) | ✅ Shanghai main board | Heavyweight | Direct (with respect to price increases) | ⭐11/15, H1 net profit attributable to parent about RMB 6.9bn (+1099%), zero interest-bearing debt, OCF/net profit >1 for 4 consecutive periods | High (turnover RMB 35.484bn, turnover ratio 14.53%) | The hardest evidence, and the most observable name within the branch |
| 2 | 德明利 (Demingli, 001309) | ✅ Shenzhen main board | Elasticity name | Industry trend (slope-sensitive) | ⭐9/15, H1 net profit attributable to parent RMB 5.7–6.5bn but Q2 already down QoQ, inventory/net assets 1.84x | Extremely high (the only limit-up, but net selling of RMB 706mn on the block-trade disclosure) | Largest elasticity, largest risk; whether to follow is decided by the signals in section 8 |
| 3 | 深科技 (Shenzhen Kaifa, 000021) | ✅ Shenzhen main board | Back row | ⭐Industry trend (lower edge) | ⭐10/15, but Peyton is only about 15% of net profit attributable to parent, at a 3.8% net margin, with expansion only landing in 2027 | Medium | Who is the follower: this one. Memory exposure is diluted to 15%, so it should not represent the branch |
| 4 | 紫光国微 (Unigroup Guoxin, 002049) | ✅ Shenzhen main board | Pure concept (with respect to memory) | Indirect supply chain | 10/15 | Low (only −0.25% on 7/30) | Weak linkage to memory price increases, not treated as a name in this branch |
| — | 太极实业 (Taiji Industry, 600667) | ✅ Shanghai main board | Pure concept (already falsified) | — | ⭐7/15 | — | Pass (see section 6) |
| Watch | 江波龙 (Longsys, 301308), 佰维存储 (Biwin, 688525), 澜起科技 (Montage, 688008), 长鑫科技 (CXMT, 688825) | ❌ Not main board | Leader/elasticity | — | Longsys has the highest H1 profit among module makers | — | Observation only. CXMT is the pricing anchor for the entire chain and must be tracked |
Conclusion within the branch: hardest evidence = GigaDevice; most observable within the branch = GigaDevice; follower = Shenzhen Kaifa; Pass = Taiji Industry. Key reminder: this branch's true leader (CXMT) is on the STAR Market and just listed, so there is no real leader on the main board — meaning the branch inherently lacks a pricing anchor on the main board, a structural disadvantage relative to the "AI hardware" branch.
7.2 Branch two: AI hardware oversold rebound (A+ grade)
| Rank | Stock | Shanghai/Shenzhen main board | Role | Catalyst directness | Fundamental support | Trading identifiability | Conclusion |
|---|---|---|---|---|---|---|---|
| 1 | 长电科技 (JCET, 600584) | ✅ Shanghai main board | Heavyweight (packaging) | Industry trend (strong, a direct read-across from ASE +11.79%) | 12/15, H1 +63.5%~101.7% | Extremely high: limit-down + RMB 10.65bn turnover + net buying of +RMB 317mn on the block-trade disclosure | The hardest trading signal |
| 2 | 沪电股份 (WUS, 002463) | ✅ Shenzhen main board | Heavyweight (PCB) | Industry trend (strong, upper edge) | ⭐12.5/15, highest on the whole list, H1 +68.2%~78.3%, capacity utilization of 92.5%–99.7% across five bases | High (touched the limit-down then recovered, turnover RMB 7.964bn) | The hardest fundamentals |
| 3 | 工业富联 (FII, 601138) | ✅ Shanghai main board | Leader (direct beneficiary) | ⭐Direct order positive (the only one in this branch) | 12/15, 2026Q1 revenue +56.5% with positive OCF | Medium (only −6.84%, no limit-down) | Direction confirmer: if it strengthens, money is genuinely rotating back; if it does not, this is only an oversold snap-back |
| 4 | 紫光股份 (Unisplendour, 000938) | ✅ Shenzhen main board | Heavyweight (networking + servers) | Industry trend | 10.5/15, gross margin 12.88%, the highest among system makers | High: limit-down + net buying of +RMB 548mn on the block-trade disclosure (the largest among limit-down names) | Forms a cross-check with JCET |
| 5 | 通富微电 (TFME, 002156) | ✅ Shenzhen main board | Elasticity name (highest H1 pre-announced growth of the big three; "number one in compute purity" is a media basis, to be verified) | Industry trend (strong) | 11/15, H1 +288%~337% | Contradiction: limit-down + net selling of −RMB 3.2bn on the block-trade disclosure, 4 institutional sell seats | Strongest news, worst flows — the only reason this report ranks it behind JCET |
| 6 | 生益科技 (Shengyi, 600183) | ✅ Shanghai main board | Upstream materials | Industry trend | 11/15, H1 +117%~131% | Medium | Fell more (−46.5%) but is more expensive (PE 61.9x) |
| 7 | 华天科技 (TSHT, 002185) | ✅ Shenzhen main board | Back row (number three in packaging) | Industry trend | 10.5/15, H1 +231%~275% | Medium (limit-down, net selling of −RMB 203mn on the block-trade disclosure) | Last of the packaging big three |
| 8 | 深南电路 (SCC, 002916) | ✅ Shenzhen main board | Integrated platform | Industry trend | 11/15, H1 +54.4%~69.1% (lowest of the four) | Medium | Lowest growth + higher valuation |
| 9 | 东山精密 (DSBJ, 002384) | ✅ Shenzhen main board | Elasticity / Apple chain | Industry trend | 10/15, H1 +282.6%~295.8% (low base) | High (turnover of RMB 17.49bn, the largest on the main board, but net selling of −RMB 236mn on the block-trade disclosure) | Apple −6% after hours is its counter-risk today |
| 10 | 浪潮信息 (Inspur, 000977) | ✅ Shenzhen main board | The one most easily misjudged | ⚠️ Industry trend (weak) + indirect supply-chain negative | ⭐10/15, H1 +226%~288%, but per public reports it is on the U.S. BIS Entity List (basis to be verified), 2026Q1 revenue −24.3%, OCF −RMB 7.772bn | Medium (declining on shrinking volume) | Who most deserves a downgrade: this one. Its exposure to overseas capex is structurally zero |
| 11 | 光迅科技 (Accelink, 002281) | ✅ Shenzhen main board | Another one easily misjudged | ⚠️ Industry trend (the weakest exposure in this round) | ⭐9/15, overseas revenue share down two straight years to 26.7%, the company's own pre-announcement attributes growth to "domestic cloud service providers"; PE (TTM) 121.7x is 1.9x Innolight's while its growth is 1/8 of Innolight's | Medium (limit-down, net selling of −RMB 44mn on the block-trade disclosure, 4 institutional sell seats) | Do not treat it as "the main-board version of Innolight" |
| 12 | 中科曙光 (Sugon, 603019) | ✅ Shanghai main board | High-margin compute | Industry trend | 10/15, gross margin 26.56%, the highest, but no H1 pre-announcement issued | Low | Lacks an earnings anchor |
| 13 | 剑桥科技 (CIG, 603083) | ✅ Shanghai main board | Elasticity (optical modules) | Industry trend | 9/15, H1 +156.7%~197.2% | Medium (limit-down + net buying of +RMB 62mn on the block-trade disclosure) | Small-cap elasticity |
| 14 | 英维克 (Envicool, 002837) | ✅ Shenzhen main board | Niche leader (liquid cooling) | Industry trend | 9.5/15 | Medium (limit-down) | Small scale, high volatility |
| Watch | 中际旭创 (Innolight, 300308), 新易盛 (Eoptolink, 300502), 天孚通信 (T&S, 300394), 胜宏科技 (Victory Giant, 300476) | ❌ Not main board | Leaders | Close to a direct order positive (North American CSP revenue is the bulk) | Eoptolink ROE 72.75%, PE 48.2x, the best on the chain | — | Observation only; Innolight filed a stake-reduction notice last night |
Conclusion within the branch: hardest evidence = JCET (flow signal) and WUS (fundamentals); most observable within the branch = JCET; followers = TSHT and SCC; the two most deserving of a downgrade = Inspur and Accelink (both "look the most like it, but have the weakest actual exposure"); Pass = none (this branch needs no Pass, though DSBJ's Apple-chain exposure needs a separate counter-risk flag).
7.3 Branch three: Politburo meeting mappings (A grade)
| Rank | Stock | Shanghai/Shenzhen main board | Role | Catalyst directness | Fundamental support | Trading identifiability | Conclusion |
|---|---|---|---|---|---|---|---|
| 1 | 工商银行 (ICBC, 601398), 招商银行 (CMB, 600036), 中国平安 (Ping An, 601318) | ✅ | Heavyweight (allocation) | Direct policy fit (debt resolution + reform of small and medium financial institutions + capital-market resilience) | 12/15 | Low (no limit-ups, no echelon) | Hardest but bluntest; a seesaw with branches one and two |
| 2 | 陆家嘴 (Lujiazui, 600663) | ✅ Shanghai main board | Elasticity (property) | Policy mapping | 8/15 | High (limit-up at 14:30 on 7/30) | Already day 2 of the fermentation, watch only |
| 3 | 中信证券 (CITIC Securities, 600030) | ✅ Shanghai main board | Heavyweight (brokers) | Policy mapping ("deepening comprehensive reform of capital-market investment and financing") | 12/15 | Extremely low (−0.04% on 7/30) | The broker sector showed no reaction at all on 7/30, which says the market does not treat this line as a trading opportunity |
| 4 | 国电南瑞 (NARI, 600406), 特变电工 (TBEA, 600089), 海兴电力 (Hexing Electrical, 603556) | ✅ | Pure concept (the "six networks") | ⚠️ A re-confirmation of old wording, not brand-new news | 9/15 | Hexing Electrical at 2 consecutive limit-ups | Strong counter-evidence: grid equipment fell −2.46% for the full 7/30 session, with net outflows of RMB 2.664bn and 27 up versus 113 down (full-session figures) |
| 5 | 通威股份 (Tongwei, 600438), 隆基绿能 (LONGi, 601012), 弘元绿能 (Hongyuan Green Energy, 603185) | ✅ | Policy elasticity (anti-involution) | Today's Yancheng meeting is a real event | 8.5–9/15 | Hongyuan Green Energy limit-up on 7/30 (a lone name) | Today's meeting is the only substantive catalyst, but "price-compliance guidance" ≠ production cuts ≠ price increases |
Conclusion within the branch: hardest evidence = banks (but with no intraday elasticity); most observable within the branch = photovoltaics (because the meeting really is being held today); follower = grid equipment (already rejected by the tape on the day the meeting was reported); Pass = Xiangjiang Holdings, Guangdian Electric.
8. Verification signals at today's open
Per the methodology required by item 3 of section 6.4 of yesterday's recap, this is stated up front: all 5 failure signals in section 8 of yesterday's pre-market report hit, while the branch ranking in section 9 scored 0/3. Therefore: the observable conditions in this section rank above the branch ranking in section 2 and the single-stock list in section 3. Where they conflict, this section governs.
What was recalibrated in the model this issue (in response to yesterday's "branch ranking 0/3"): ① section 9.1 is now ordered by the clarity of verification conditions rather than by total score (see the note in 9.1); ② a decay rule was added to the "expectation gap" item in the scoring table — if the same pre-announcement has already been ignored by the market N times, its expectation-gap score decays each time — so this issue WUS, JCET and TFME had their expectation-gap scores cut from 9 in the first draft to 7 (three falsifications in July); ③ a 1-point difference in score has no discriminating power, so please read the scores as "the 70s band / the 60s band / the 50s band" rather than by rank.
8.1 Auction signals (09:15–09:25)
⚠️ Item 0, which ranks above the A-share auction itself: look at Korea and Taiwan first. The Korean and Taiwanese markets open about 60–90 minutes before A-shares, and are the most direct and timely verification of "whether the overnight memory gains were an ADR premium", "whether Samsung's guidance can carry through" and "whether SK Hynix's earnings miss has been digested". Three specific things to watch: ① SK Hynix's open on the Korean main board (it closed −0.79% on the Korean main board on 7/30 while its U.S. ADR was quoted +15%, a clear gap between the two venues, and the Korean main board governs today); ② whether Samsung Electronics can extend its strength after the record results of 7/30; ③ the opens of TSMC and ASE in Taiwan (their U.S. ADRs were +7.63% and +11.79% respectively). If Korean/Taiwanese semiconductors open high and fade, then the A-share gap up at 09:25 is very likely the high of the day.
| Signal | Meaning | Threshold |
|---|---|---|
| The overall gap-up size of the AI hardware chain | Overnight SOXX +8.5%, ASE +11.79%, Microsoft +15.5% theoretically correspond to a 3%–6% gap up | A gap up within 2% = the external positives are insufficiently digested and the rebound is low quality; a gap up above 7% = sentiment overshoot, historically raising the probability of a spike and fade; a 3%–5% gap up is the most balanced between "digestion" and "overshoot" (the above are observation bands, not action thresholds) |
| Auction volume in 长电科技 (JCET, 600584) / 紫光股份 (Unisplendour, 000938) | Both are "limit-down + net buying on the block-trade disclosure", and auction volume will determine whether yesterday's net buying was bottom-fishing or getting trapped | Auction turnover markedly enlarged with a gap up above 4% = yesterday's net buying was effective; a gap down or a flat open = yesterday's net buying was passive absorption, downgrade immediately |
| Whether 德明利 (Demingli, 001309) opens one-line or sharply higher | It was the only limit-up name, with net selling of RMB 706 million on yesterday's block-trade disclosure | A one-line board = yesterday's net selling has been covered by today's buying, and the branch strength is revised up; opening high and fading = the RMB 706 million of net selling is confirmed and this report's "watch closely" label on it is void |
| The gap-up size of 工业富联 (FII, 601138) | It is the only "direct order positive" name and was resilient yesterday | Its gap up ≥ the average gap up of the limit-down names = money is genuinely rotating back; a clearly smaller gap up = today is only a technical snap-back in oversold names |
| Baijiu heavyweights (贵州茅台 / Kweichow Moutai 600519, 五粮液 / Wuliangye 000858) | The public attribution for yesterday's rise was safe-haven money rotating in | A gap down, or opening high and fading = safe-haven money is rotating back into technology and branch five is downgraded; continued advance on volume = the safe-haven logic is intact, meaning the market does not buy the overnight overseas positives |
| ⚠️ 山西汾酒 (Shanxi Fen Wine, 600809) goes ex-dividend today (RMB 6.56 per share) | A technical gap | Its decline includes the ex-dividend effect and must not be read directly as weakness — this is the number most easily misread today |
8.2 Sector signals (09:30–10:30)
| Confirmation signal | Failure signal |
|---|---|
| Semiconductors, components and communications equipment all turn green at the same time, with more than 100 advancers within the semiconductor sector (yesterday it was 7 up, 176 down, on the Tonghuashun industry basis; use same-source data for intraday verification — East Money's industry classification counts are not comparable across sources) | Advancers in the semiconductor sector fall back below 50 quickly after a gap up |
| At least 2 of the packaging big three (JCET 600584, TFME 002156, TSHT 002185) hit the limit-up or rise more than 7% — this is the most direct read-across from ASE +11.79% overnight, and the single most important sector criterion this report has designed | All three open high and fade → the ASE read-across does not hold in A-shares, and branch two is downgraded immediately |
| A second and third main-board limit-up appear in the memory chain (yesterday there was only Demingli). Under the standard established in yesterday's recap, a lone name does not constitute a branch | Only Demingli hits the limit-up all day → memory is still a single-stock event, not a branch |
| All four AI PCB names (WUS 002463, Shengyi 600183, SCC 002916, DSBJ 002384) close green together on expanding volume | Only the most oversold names rise while the leaders do not → this is valuation repair, not logic repair |
| Whole-market limit-ups return above 70 and limit-downs fall back below 20 (yesterday: 52 limit-ups vs 74 limit-downs) | Limit-downs still above 40 → yesterday's freezing point has not ended |
| The PV sector produces a sector-level advance (rather than Hongyuan Green Energy alone), corresponding to today's price-compliance guidance meeting in Yancheng | PV is still a lone limit-up → the meeting adds nothing substantive |
8.3 Single-stock signals
| Observation point | Confirmation | Failure |
|---|---|---|
| Absorption in the first 5 minutes | JCET, Unisplendour and WUS do not break their auction opening price within 5 minutes of the open | They fall back more than 2% within 5 minutes of the open |
| A quick limit-up seal | The first limit-up in AI hardware/memory appears before 10:00, and does not break | No name on the chain hits the limit-up all day, or the board breaks after sealing |
| Volume without breaking | If Demingli seals the board, the sealing amount must be markedly larger than yesterday's and the board must not open all day | The board breaks (it already broke once yesterday) |
| Large-order net inflows | GigaDevice, WUS and JCET show net inflows of major funds with turnover no lower than yesterday's (RMB 35.484bn / 7.964bn / 10.647bn respectively) | Price up but turnover halved → this is short covering, not new money |
| Being crowded out by a stronger name | — | If 中际旭创 (Innolight, 300308, ChiNext 20cm) gaps up sharply and absorbs all the money, main-board names will be "crowded out" and get only the leftovers — this is an inherent disadvantage of a main-board-first strategy on days when 20cm names rebound, and must be recognized in advance |
8.4 Risk signals (any one of these requires downgrading all of today's judgments)
- Gap up then plunge: the AI hardware chain gaps up more than 4% but falls back to flat before 10:00 — this is the classic pattern of a "window for distributing to trapped holders", and in a sector that has already fallen for a month with extremely dense trapped supply overhead, this report considers it the single most likely risk.
- A lone limit-up: only 1 name on the chain hits the limit-up all day (a repeat of yesterday's "Demingli alone" and "Jiangsu Yongding alone" structure). Yesterday's recap already proved with two cases that a lone name does not constitute a branch.
- The heavyweights do not follow: if FII, GigaDevice and WUS — the three heavyweights — are weaker than oversold small caps all day, this is a sentiment snap-back rather than a return of money.
- Yesterday's strong names ebb: if baijiu and banks fall sharply today while AI hardware fails to catch the baton, it is a double-kill structure — when that structure appears on the last trading day of the month, index risk exceeds single-stock opportunity.
- The block-trade direction diverges again: if the day's biggest gainers show "price up on shrinking volume" or "limit-up sealing orders thinning rapidly" intraday, recall yesterday's lessons from Demingli (limit-up but net selling of RMB 706 million) and JAC Motors (limit-up but net selling of RMB 393 million).
- ⚠️ The Apple chain weakens on its own: Apple fell about 6% after hours (weak guidance on supply constraints), so 东山精密 (DSBJ, 002384), 鹏鼎控股 (Avary Holding, 002938), 蓝思 (Lens Technology) / 立讯 (Luxshare) and other Apple-chain names may diverge from the AI chain. If the Apple chain drags down the whole electronics sector index, it will mask the true strength of AI hardware — watching the sector index will mislead you; you must watch individual stocks.
- The flow characteristics of the last trading day of the month: today is the last trading day of July, with month-end ranking and position-adjustment behavior, so the extrapolability of any single-day conclusion is lower than usual. The PBoC has already conducted RMB 600 billion of overnight reverse repos daily from 7/29 to 7/31 and will conduct RMB 300 billion on 8/3, so cross-month liquidity is arranged; but this report did not obtain DR007/Shibor, which is "no reliable data available".
- ⚠️ The official July manufacturing PMI is usually released by the National Bureau of Statistics at 09:30 on the last day of each month; if it is released as usual today, it lands at the same moment as the open. This report could not verify today's exact release schedule or the market's expected value, so this is "no reliable data available" — please confirm it yourself before the open, as it could change the tape at the instant of the open.
- ⚠️ A systemic deterioration in cloud vendors' free cash flow is the counter-variable being overlooked in the S-grade branch: Meta fell about 8% against the trend on 7/30 on weak guidance and a plunge in free cash flow; Amazon's own free cash flow is also under pressure even as it raises capex. "Raising capex" is orders for the upstream but cash outflow for the cloud vendors themselves — if the market's pricing focus switches from "who benefits" to "who is burning cash", the whole AI hardware chain's rebound will be interrupted. Microsoft (which cut FY27 capex) surging and Meta (high capex + deteriorating FCF) slumping on 7/30 is the signal that this switch has already begun.
- Geopolitics and FX: the escalation of the U.S.–Iran situation has already pushed oil higher and reinforced rate-hike expectations (see item 26 of section 1); this report did not obtain the RMB exchange rate or the dollar index, which is "no reliable data available".
9. Final conclusions
Stated again: what follows is a ranking of research attention and a set of observable verification conditions. It contains no buy or sell recommendations, no position sizing, no target prices and no ratings.
9.1 The 5 main-board stocks most worth watching today
Note on the reordering rule (added after QC): the ordering in this section differs from the overall list in section 3, because the two have different ordering objectives — section 3 ranks by total score of "catalyst strength + fundamentals", while this section ranks by the clarity of today's observable verification conditions. This is precisely the adjustment required by item 3 of section 6.4 of yesterday's recap ("move the verification-signals section ahead of the single-stock list"). Hence JCET (limit-down + positive flow direction, the cleanest verification conditions) rises to number 1, TFME (strongest news but negative flow direction, with two non-comparable block-trade bases) drops out of the top five, and Unisplendour (also "limit-down + top-five buying seats net buying") enters the top five.
| Rank | Stock | Branch | Reason for attention | Biggest risk | Verification point today |
|---|---|---|---|---|---|
| 1 | 长电科技 (JCET, 600584) | AI hardware (packaging) | Four signals hold at once: Shanghai main board + limit-down on 7/30 + turnover of RMB 10.647 billion + net buying of +RMB 317 million on the block-trade disclosure; overnight ASE +11.79% is the most direct peer read-across; H1 pre-announcement +63.5%~101.7% | A cumulative decline of over 20% across three consecutive sessions; if it cannot hold today, yesterday's net buying will prove to be trapped absorption | A gap up above 4% at the 09:25 auction + not breaking the opening price in the first 5 minutes; at least 2 of the packaging big three up more than 7% |
| 2 | 沪电股份 (WUS, 002463) | AI PCB | ⭐Fundamentals of 12.5/15, the highest on the whole list; after the 7/14 pre-announcement of H1 +68.2%~78.3% (with Q2 alone +34.7% QoQ, the strongest in four years) the share price instead fell −39.0%; capacity utilization of 92.5%–99.7% across five bases, so demand has already been realized as actual shipments | PB at the 95.6th percentile of its full history; 2026Q1 OCF/net profit collapsed to 0.41; the timing and size of the H-share issue are entirely unknown | Whether it closes green in step with Shengyi, SCC and DSBJ (in step = logic repair; only it rising = single-stock behavior) |
| 3 | 兆易创新 (GigaDevice, 603986) | Memory | Shanghai main board + memory at 71.3% of revenue + H1 net profit attributable to parent of about RMB 6.9bn (+1099%) + zero interest-bearing debt + the chairman's buyback proposal; −55.8% from its 6/29 high | The CXMT shadow-stock premium has been irreversibly dismantled by the real company's 7/27 listing; governance blemishes (RMB 4.4bn cashed out vs a buyback ceiling of RMB 2bn, with the stake increase waiting until 12/13); PB at the 90.1st percentile of the past five years; RMB 2.05bn of the RMB 6.9bn H1 net profit is non-recurring | Whether the memory chain can produce a second and third main-board limit-up — if Demingli is the only one, this branch is not established |
| 4 | 工业富联 (FII, 601138) | AI server ODM | The only name on this list qualifying as a "direct order positive" (2026Q1 revenue +56.5% with positive OCF, a hard contrast against Inspur's −24.3%); H1 pre-announcement +93%~+101% | Excessive scale (Q1 revenue RMB 251.08bn) naturally caps share-price elasticity; gross margin is only 7.35% | It is the "authenticity detector": if its gap up and persistence are weaker than the limit-down names, today is only an oversold snap-back rather than a return of money |
| 5 | 紫光股份 (Unisplendour, 000938) | Servers / networking | Limit-down on 7/30 but net buying of +RMB 548 million on the block-trade disclosure, the largest of all limit-down names; gross margin of 12.88%, the highest among systems/networking makers; H1 pre-announcement +83.5%~122.9%; a drawdown of only −25.2%, so trapped supply overhead is relatively thin | Negative OCF per share; the systems business has a structurally low gross margin | It forms a cross-check with JCET: only if both "limit-down + net inflow" names strengthen together does the rebound's flow character hold |
9.2 The 3 strongest branches today
| Rank | Branch | Core catalyst | Persistence | Representative stocks (main board) |
|---|---|---|---|---|
| 1 | AI hardware oversold rebound (packaging > PCB > servers > optical modules) | Overnight SOXX +8.5%, ASE +11.79%, TSMC +7.63%, Microsoft +15.5%; AWS backlog $496 billion, capex $220 billion; interim pre-announcements across the whole chain generally doubled while July share prices fell 20%–50% | 1–3 sessions. This is a rebound within a downtrend unless sustained volume and a limit-up echelon appear | 长电科技 (JCET, 600584), 沪电股份 (WUS, 002463), 工业富联 (FII, 601138), 紫光股份 (Unisplendour, 000938), 通富微电 (TFME, 002156), 生益科技 (Shengyi, 600183) |
| 2 | Memory price increases | Amazon's capex raise explicitly attributed to memory costs; SanDisk +22%, Micron +15%, SK Hynix +15%, Silicon Motion +21.66% | The logic is medium term, the A-share names are a short-term rebound. ⚠️ The strongest counter-evidence is also in this branch: TrendForce estimates Q3 DRAM contract prices at only +13%~18% QoQ and NAND at +10%~15%, versus about +60% in Q2 — the slope of the price increase is converging sharply, and module makers' gross margins depend on the slope, not the level | 兆易创新 (GigaDevice, 603986), 德明利 (Demingli, 001309); ⚠️ 太极实业 (Taiji Industry, 600667) has been falsified by company filings, see section 6 |
| 3 | Photovoltaic "anti-involution" price compliance (Yancheng meeting today) | Today the SAMR Price Supervision and Competition Bureau holds a PV price-compliance guidance session in Yancheng, Jiangsu; the General Rules for the PV Industry Cost Accounting Model took effect on 7/27; the Politburo meeting simultaneously stated "continue the comprehensive rectification of involutionary competition" | Event-driven, 1–2 sessions. It is "price-compliance guidance", not production cuts and not price increases, and the path to earnings delivery is long | 通威股份 (Tongwei, 600438), 隆基绿能 (LONGi, 601012), 弘元绿能 (Hongyuan Green Energy, 603185), 福莱特 (Flat Glass, 601865) |
9.3 Directions this report excludes from its watch scope today, and why
| Direction | Reason for exclusion |
|---|---|
| 太极实业 (Taiji Industry, 600667) and every "Hitech Semiconductor = memory price beneficiary" mapping | ⭐The company's 2026-06-26 filing states in its own words that it is "relatively little affected by the industry upcycle" and "does not involve HBM products"; Hitech operates on a cost-plus foundry model of "all costs + an agreed return (10% of total investment + excess return)"; across three years spanning a full cycle its net profit was RMB 219.1 / 237.1 / 239.1 million, so price elasticity ≈ 0. This is the only falsification in this report backed by hard counter-evidence from a company filing |
| 浪潮信息 (Inspur, 000977) as an "AWS/Microsoft capex beneficiary" | ⚠️ According to public reports, the company itself has been on the U.S. BIS Entity List since 2025-03-25 (the company responded in its 2024 annual report that "the impact is broadly manageable"; this report did not obtain the original BIS notice number, nor verify whether any license exemption or subsequent removal exists, and flags the basis as "to be verified"); if true, AWS/Azure orders are structurally unaddressable; 2026Q1 revenue −24.3% (versus FII at +56.5%), the exact opposite direction in the same quarter. Its real drivers are domestic compute demand and inventory price spreads during the memory price-increase window; there is no causal link to the AWS backlog. Attributing the 7/8 limit-up to the results pre-announcement is correct; attributing today's potential rebound to AWS is wrong |
| 光迅科技 (Accelink, 002281) as "the main-board version of Innolight" | ⭐Overseas revenue share has fallen for two straight years: 34.3% → 28.3% → 26.7%; the company's own H1 pre-announcement attributes growth to "domestic cloud service providers" and never mentions North America; PE (TTM) 121.7x is 1.9x Innolight's 64.5x and 2.5x Eoptolink's 48.2x, while its revenue growth is only 1/8 of Innolight's; the placement price of RMB 167.55 is already 9.08% underwater |
| 永鼎股份 (Jiangsu Yongding, 600105) (the positive is real, but the price has run ahead of it) | The positive is real (a direct order positive, RMB 1.133 billion), but there is a triple discount: it holds only 44.8567% of Suzhou Dingxin (only about 44.9% of the profit is attributable to the parent), the orders are recognized in installments across 2.4 years, and the company was publicly censured by the Shanghai Stock Exchange in 2024 for "inaccurate disclosure of a results pre-announcement" with the executives involved still in post. More important is the pricing: the 7/30 single-day market-cap increase of RMB 3.289 billion ÷ the equity-adjusted annualized net profit attributable to parent of about RMB 47 million ≈ 70x; it spiked to the limit-up without sealing that day, with a maximum intraday drawdown of 7.9% and a turnover ratio of 19.75% — a heavy-volume divergence/realization pattern |
| Next-day relays of yesterday's limit-up names (Aile Home at 8 boards, Mingxin Xuteng at 4, Chuanzhi Education at 4, JAC Motors, Shede Spirits, Jinhui Liquor) | Among yesterday's 52 limit-ups, several showed "limit-up but net selling on the block-trade disclosure" (Demingli −RMB 706mn, JAC Motors −RMB 393mn, Guangdian Electric −RMB 12mn, Jinhui Liquor −RMB 8mn); and if money rotates back into technology today, yesterday's safe-haven mainline and the sentiment high-flyers lose their funding at the same time |
| The "8/1 price increase takes effect" mapping in passive components / MLCC | The price-increase letter is real (Samsung Electro-Mechanics +30% across the board, effective 8/1, with existing orders repriced retroactively), but the market has already voted it down once with its feet: on 7/30 the components sector fell −6.58%, with 4 up, 58 down and 7 limit-downs; Fenghua Advanced had 6 failed limits that day, and its 7/29 block-trade disclosure showed northbound net selling of RMB 347 million and institutional net selling of about RMB 201 million |
| Unverified "宇树科技 (Unitree) supplier" mappings | An IPO by itself adds no revenue to any A-share company; any supply relationship without company-filing support is treated as a pure concept; and the media-reported "U.S. FCC import restrictions on foreign-made advanced robots" (original FCC text not obtained, to be verified) is a counter-variable |
| 东山精密 (DSBJ, 002384) as an "AI PCB rebound" name | ⚠️ It is the only name on this list with significant Apple-chain exposure, and Apple gave weak guidance on supply constraints and fell about 6% after hours. If the Apple chain weakens on its own today, it will diverge from the AI chain; and its H1 pre-announced growth of +282.6%~295.8% also comes mainly from a low base |
| 中际旭创 (Innolight, 300308) | Not main board; on the evening of 7/30 it filed that the controlling shareholder and parties acting in concert cut their stake from 17.99% to 16.53%, selling 6.21 million shares; turnover of RMB 59.773 billion that day was a record high |
9.4 Final one-sentence judgment
Today is "the day with the most sufficient evidence" and also "the day it is easiest to overestimate because of that": the overnight overseas evidence chain (Amazon's $220 billion capex explicitly attributed to memory costs, the AWS backlog of $496 billion, Microsoft +15.5%, SOXX +8.5%, ASE +11.79%, Samsung Q2 operating profit +1813.83%) points in exactly the same direction as the already-filed interim pre-announcements across the whole A-share chain (packaging +63%~337%, AI PCB +54%~296%, memory +1099%~5611%); but three things must be remembered at the same time: ① the 7/30 U.S. rally was mainly a recovery of the 7/29 hawkish Fed shock (the Nasdaq's two-day net gain is about 1.0% and the Dow has not recovered), and the CME already shows a 78.8% probability of a September hike; ② Samsung's record results and Microsoft's Azure +43% were already public before the A-share open on 7/30, and A-share semiconductors still fell −7.08% that day — the marginal impact of external supply-side evidence on this chain was empirically measured as weak yesterday; ③ the only genuinely new information not yet priced by A-shares is Amazon's capex and Apple's soft guidance. But precisely because this A-share chain has now fallen for a month with extremely dense trapped supply overhead, and because the slope of memory price increases has converged from about +60% in Q2 to an estimated +10%~18% in Q3, this report characterizes today as "a high-quality rebound attempt within a downtrend" rather than "a trend reversal" — so the real basis for decisions is not the branch ranking in section 2 but the six observable falsification conditions in section 8, especially "at least 2 of the packaging big three up more than 7%", "whether the memory chain can produce a second main-board limit-up" and "whether FII's strength is no weaker than the limit-down names"; until they are satisfied, yesterday's "Demingli hit the limit-up but showed net selling of RMB 706 million on the block-trade disclosure" remains the truest picture of this market right now.
⚠️ Risk warning: this list is pre-market information triage and observation only and does not constitute investment advice. A-share volatility risk is extremely high, and automatically generated content may contain stale information or errors in industry-chain mapping; it must not be used directly as a basis for trading.
Sources38
Every external link cited in the body, numbered in order of appearance. · 21 domains
- 1CNBCcnbc.com
- 2Sina Financefinance.sina.com.cn
- 3Sina Financefinance.sina.com.cn
- 4Securities Times · SK Hynix earnings missstcn.com
- 5Sina Techfinance.sina.com.cn
- 6Securities Timesstcn.com
- 7Microsoft websitenews.microsoft.com
- 8Sina Financefinance.sina.com.cn
- 9Sina Financefinance.sina.com.cn
- 10Xinhuanews.cn
- 11Cailianshecls.cn
- 12MacRumorsmacrumors.com
- 13CNBCcnbc.com
- 14NBDnbd.com.cn
- 15Securities Timesstcn.com
- 16Securities Timesstcn.com
- 17Jrj.com24h.jrj.com.cn
- 18Jrj.comstock.jrj.com.cn
- 19Sina Financefinance.sina.com.cn
- 20Sina Finance · critical piecefinance.sina.com.cn
- 21Jrj.comfinance.jrj.com.cn
- 22ICSmarticsmart.cn
- 23Tencent Newsnews.qq.com
- 24Securities Timesstcn.com
- 25Yicaiyicai.com
- 26Sina Finance search pagefinance.sina.com.cn
- 27Sina Financefinance.sina.cn
- 28Sina Financefinance.sina.com.cn
- 29Xinhuanews.cn
- 30Sina Financefinance.sina.com.cn
- 31LinkPDFstockmc.xueqiu.com
- 32Linktrendforce.cn
- 33CFIcfi.net.cn
- 34East Moneyfinance.eastmoney.com
- 35National Business Dailymrjjxw.com
- 36NetEase/media aggregation163.com
- 37Securities Timesstcn.com
- 38Sina Techfinance.sina.com.cn