A-Share Pre-Market Brief | 2026-08-28 Friday

Scan window: 2026-08-27 (Thursday) 15:00 close → 2026-08-28 (Friday) 08:05 Beijing time. Data read timestamps: quotes / fund flows / limit-up-limit-down pools are a snapshot taken after the 2026-08-27 close (15:00); the dragon-tiger list (LHB) is on the exchange disclosure basis after 8/27 18:00 (yesterday's recap went out at 15:30 and could not include it; this brief has added it); overnight US equities are as of the 2026-08-27 16:00 ET close (= 8/28 04:00 Beijing time), with after-hours prices separately timestamped. Basis boundaries for this brief (please judge conclusion strength accordingly): ① the per-stock fund column uses the THS "net fund amount" basis (inflow − outflow, across all 5,209 listed names), which is a different basis from the Eastmoney "main-force net inflow" cited in several earlier briefs and cannot be netted against them across briefs; ② the LHB "net buy" counts only the top five seats, a different denominator from market-wide net fund amount, so for conflicting names this brief gives both numbers rather than choosing one; ③ the limit-up / limit-down pools have been validated with tc == len(pool) (8/27: 77/77; 8/26: 52/52; 8/25: 65/65); ④ sector moves / fund flows use the THS 90-industry basis; ⑤ all PE/PB figures cited here are recomputed on the post-interim-report basis — the interim reports of 德明利 (Demingli, 001309), 胜宏 (Victory Giant) and 金安 (Jin'an) all landed on the evening of 8/27, so any valuation percentile scraped before 8/28 is wrong for these three (see §5).

⚠️ Data-fetch failure log (this section is not sent to clients):

• Eastmoney's push2/clist endpoint has now returned 502 Bad Gateway for a second full trading day (tested and failed three times: 8/27 pre-market, 8/27 post-close, 8/28 pre-market). Capabilities affected: sector quotes, sector fund flows, market-wide paged snapshots, ulist.np batch fund flows.
• Substitute routes that actually worked for this brief (recommend hard-coding them into the scripts): ① quotes via Tencent qt.gtimg.cn (stable, includes PE/PB/turnover/float market cap; pulling 33 names at once is no strain); ② limit-up / broken-limit pools via push2ex (a different cluster, unaffected); ③ sector quotes + fund flows via ak.stock_board_industry_summary_ths(); ④ per-stock fund flows via ak.stock_fund_flow_individual(symbol="即时"); ⑤ LHB via ak.stock_lhb_detail_em() (succeeded this time, 66 rows); ⑥ the disclosure-appointment calendar via ak.stock_report_disclosure(market="沪深京", period="2026半年报") — the period literal is "2026 半年报", writing "2026 年中报" raises a KeyError; the returned column names are 首次预约/初次变更/…/实际披露, without the word "时间".
• Announcement route: CNINFO's hisAnnouncement/query needs a code,orgId combination for its stock parameter, and passing only the 6-digit code silently returns 0 records (no error) — we hit this; switched to Eastmoney np-anotice-stock (stock_list=<code>), which returns art_code that can be assembled into https://pdf.dfcfw.com/pdf/H2_<art_code>_1.pdf to fetch the source PDF. That CNINFO endpoint also throws intermittent 504s.
• PDF parsing: WebFetch returns only "binary, cannot parse" for announcement PDFs, but it does drop the file into the local tool-results directory, and the local pdfplumber extracts the full text fine. The source quotes from 深南电路 (Shennan Circuits, 002916)'s investor-relations record and 德明利 (001309)'s dividend proposal both came through this route. The 澜起科技 (Montage Technology, 688008) H-share announcement PDF of 8/18 came back as 0 bytes; its board / disclosure date was cross-confirmed via the disclosure-appointment calendar instead.
• Sina US quotes hq.sinajs.cn returns 403; overnight US equities were switched to stockanalysis single-stock pages, with a "close / previous close − 1" reverse-check on every name (NVDA 227.98/209.66−1 = +8.737% matches the page's +8.74%; SMH 573.00/555.77−1 = +3.10% matches). No staleness found.
• Valuation-basis trap (two sub-agents independently fell into the same one): Baidu's stock_zh_valuation_baidu on 8/27 gave 德明利 (001309) PE 23.76 / PB 14.72, 金安 PE 105.44, and 胜宏 PB 14.85 — all of them pre-interim-report denominators; Tencent's qt.gtimg.cn had already updated to the H1 basis. This brief uses Tencent throughout, independently reverse-checked.

---

0. One-Sentence Takeaway

The most worthwhile thing to look at today is not any single news item, but a directional fork revealed by the batch of interim reports that landed last night: gross margins at the upstream materials end are expanding, while profits at some PCB manufacturers are collapsing. The evidence is three sets of company-stated numbers from the same reporting period, the same supply chain, pointing in exactly opposite directions — **金安国纪 (Jin'an Guoji, 002636) copper-clad laminate (CCL) Q2 quarter-on-quarter gross margin +10.64pct to 37.08% (year-ago base was only 11.31%), volume up just 10.77% while revenue rose 68.46% (implied ASP +52.1%); 胜宏科技 (Victory Giant, 300476) core PCB business Q2 quarter-on-quarter gross margin −2.35pct, single-quarter ex-non-recurring net profit −7.7% QoQ; and the most extreme sample is one disclosed in the same batch last night and missed entirely in this brief's first draft — 奥士康 (Aoshikang, 002913) — a pure PCB manufacturer, H1 revenue +16.45% while ex-non-recurring net profit fell −57.17%. (Basis note: the above mixes single-quarter QoQ with H1 cumulative bases and cannot be lined up as "three directions within the same Q2".) The interim report of 深南电路 (Shennan Circuits, 002916) (already disclosed 8/26 18:41 — not last night's news) records the same transmission: "the tight-supply condition in key upstream materials such as CCL fiberglass electronic cloth and copper foil has continued to intensify, driving sharp cost increases and exerting a certain impact on the Company's profitability". The strongest branch is therefore upstream CCL / electronic cloth materials and package substrates, not PCB manufacturing. ⚠️ But this must be said at the same time: this brief's sample cannot rule out an equally natural alternative explanation — export / FX exposure. Ranking the three companies by export exposure coincides exactly with ranking them by supply-chain position, and 深南电路 (002916)'s own interim report shows a domestic gross margin +7.36pct versus overseas of only +0.73pct — a 6.63pct fork inside a single company. "Profit is stuck upstream" should be treated as a hypothesis awaiting verification, not a proven conclusion; see the counter-evidence passage below §2. The driver type is earnings (Day 1) primarily, overseas industry (Day 2) secondarily: NVIDIA's Q2 FY27 (revenue $96.22B, Q3 guidance $108.0B, AWS commitment to purchase 2 million GPUs) was released after the US close on 8/26, A-shares have already traded it for a full day on 8/27, so today is pricing Day 2, not Day 1; overnight NVDA +8.74% and SMH +3.10% are confirmation, but MU closed −0.32% and fell another −1.43% after hours, so the memory leg was absent overseas last night. There is one split on the money side that must be stated clearly: every one of yesterday's CCL and PCB limit-up names showed "limit-up + market-wide net fund outflow" (金安国纪 (002636) −RMB 557 million, 长飞光纤 (YOFC, 601869) −RMB 708 million, 宏昌电子 (Hongchang Electronics, 603002) −RMB 73 million, 嘉立创 (JLC Technology, 001232) −RMB 66 million, 东材科技 (Sichuan EM Technology, 601208) −RMB 49 million), while the real net inflows went to 生益科技 (Shengyi Technology, 600183), which failed to hold its limit (+RMB 1.251 billion), and to the memory heavyweights 长鑫科技 (CXMT, 688825) (+RMB 3.230 billion) and 澜起科技 (688008) (+RMB 2.095 billion). Pre-market state call: today is an earnings-verification day, not a news-driven day — the 698 companies whose actual disclosure date is already marked 8/28 landed last night, while 951 companies market-wide have still not disclosed and must all finish before 8/31 (next Monday, the statutory deadline); and the appointed interim-report disclosure date for both 澜起科技 (688008) and 长鑫科技 (688825) is 8/29 — 8/29 is a Saturday, which means these two interim reports will be published between this Friday's close and the weekend, with the entire share-price reaction landing next Monday (8/31). These two happen to be the #1 and #2 net fund inflows of yesterday; holding them into today's close = taking on an event exposure that spans the weekend and cannot be adjusted during the week.

⚠️ Two names in this brief had their draft conclusions overturned by pre-publication fundamentals checks, and the direction was downward, out of the recommendation slots: 德明利 (001309) and 胜宏科技 (300476). The first draft treated their interim reports as fresh in-window positives; the check found that 德明利 (001309)'s H1 numbers were already public in the 7/15 earnings pre-announcement and had already been fully priced by three one-tick limit-down days, and that the 20.8pct gap between 胜宏科技 (300476)'s net profit attributable to parent +33.30% and ex-non-recurring +12.50% comes entirely from an equity mark-to-market gain the company itself labels "not sustainable". The full revision trail is in §10.

---

1. News Overview

All times are Beijing time. Items tagged (post-close) are incremental news falling inside the scan window; items tagged (out-of-window / priced N days) are old news the market has already traded, listed to mark pricing progress, not as fresh positives.

1.1 The pricing-progress dimension (implementing item 1 of §6.4 of yesterday's recap)

Yesterday's recap required that "whenever a sector shows ≥3 limit-ups on ≥3 consecutive days, it is force-entered into the candidate pool regardless of the original catalyst date". This brief has done so, aggregating the official limit-up pools of 8/25, 8/26 and 8/27 by Shenwan industry:

This table corrects an easy mistake to make: the index of the AI hardware chain has been rising for many days, but its limit-ups only appeared in a batch for the first time yesterday. "The sector has been up for a long time" and "the sector is starting to print limit-ups" are two different pricing stages — the former is money pushing the index, the latter is sentiment entering individual names. But this must be said at the same time: Day 1 also means there is no historical sealing behaviour to reference, so the money quality of yesterday's limit-up batch has to be checked separately — and the check says most of them were distributing (see §0 and §2), which is why this brief gives none of those limit-up names a recommendation slot merely because it is "Day 1".

One methodology self-check: aggregating by Shenwan industry breaks the AI materials chain apart — 联瑞新材 (688300) is classified under "non-metallic materials" and 大普微 (DapuStor, 301666) under "computer equipment", so neither appears in any row above. Aggregating by industry code systematically understates the breadth of a cross-industry main theme, which is why this brief's branch split is a manual grouping by supply-chain position rather than the Shenwan industry taxonomy taken as given.

1.2 Main-theme names on today's (8/28) disclosure calendar — check "which time is this being said" before you look at the numbers

Today (first-appointed date 8/28) has 719 companies, of which the 698 whose actual disclosure date is already marked 8/28 mostly landed last night. The four with the highest market cap / main-theme relevance are below. All four report today, but only one is genuinely saying it for the first time.

What this table is really about is something bigger than any single stock: 1,490 companies disclose interim reports today and tomorrow, and a substantial share of their H1 cumulative numbers were already made public via earnings pre-announcements, flash results or H-share announcements. This brief has verified that on four names: 德明利 (001309, 7/15 pre-announcement), 金安国纪 (002636, 7/14 pre-announcement), 中天科技 (600522, 7/15 pre-announcement) and 中芯国际 (688981, 8/14 H-share). Before reading any interim report today, first ask "which time is this being said", then ask "are the numbers good".

---

2. Strongest Positive Branches, Descending

⚠️ One piece of counter-evidence against this brief's core thesis (added by QA; please weight it equally with the table above)

This brief's core thesis is "profit is stuck in upstream materials". But there is an equally natural alternative explanation this brief's sample cannot exclude: export / FX exposure. Ranking the three companies by export exposure coincides exactly with ranking them by supply-chain position —

Company	Overseas revenue share	Gross margin change	Supply-chain position
金安国纪 (002636)	domestic 97.70% (almost no FX exposure)	H1 +21.83pct	Upstream materials
深南电路 (002916)	overseas 37.60%	Up overall	Midstream substrates / PCB
胜宏科技 (300476)	export-led	H1 −3.00pct	Downstream PCB manufacturing

Table scrolls horizontally

The strongest counter-evidence comes from 深南电路's own by-region table in its interim report: domestic sales gross margin 33.19%, +7.36pct YoY; overseas sales gross margin 30.93%, only +0.73pct YoY — the same company, the same reporting period, the same supply chain, the same exposure to CCL price hikes, and a 6.63pct difference in gross margin expansion between domestic and overseas. "Upstream price hikes" cannot explain that fork, but "export exposure / the RMB exchange rate" can: export sales are priced in foreign currency while costs are in RMB, so RMB appreciation directly lowers the RMB-denominated selling price and compresses gross margin — this channel runs through revenue, not through finance expense (this brief's §5⑩ was right to break out FX as a "second headwind", but in doing so it missed this revenue-side channel). Corroboration: 深南's finance expense was +605.44% YoY and 胜宏's +694.35%, and both interim reports attribute this to the exchange rate, showing this is a shared environment across the export chain, not a 胜宏-specific trait.

This brief's characterisation must therefore be narrowed to: "gross margins are moving in opposite directions at the upstream materials end and at part of the PCB manufacturing end" is a directly observable fact; while "profit is migrating from downstream to upstream" is a hypothesis awaiting verification, not a proven conclusion. Same direction does not establish causation — a common third factor (FX, the split between AI and traditional demand) can equally produce same-direction disclosures.

A second piece of wording that needs narrowing: for 金安国纪's gross margin expansion, the moat the company itself cites is "relying on the Company's own electronic fiberglass cloth production" — which means at least part of its profit comes from vertically integrating the fiberglass-cloth layer, not from taking profit out of PCB makers' hands. The magnitudes do not line up either: 胜宏's Q2 revenue of RMB 6.110 billion × a 2.35pct gross margin difference ≈ RMB 140 million of gross-profit impact, while 金安's entire H1 revenue is only RMB 3.399 billion — the two magnitudes cannot support an accounting-sense "transfer" of "downstream profit ending up upstream".

The biggest divergence between this brief and yesterday's recap, and why: yesterday's recap merged "PCB / CCL" into a single branch ranked #2. This brief splits it into three layers with three different rankings — CCL materials #1, package substrates #2, PCB manufacturing #6. The basis for the split is not judgement but the three opposite-direction single-quarter gross margins in last night's three interim reports: 金安 (materials) Q2 +10.64pct QoQ, 深南 (substrates) +16.20pct YoY, 胜宏 (PCB manufacturing) Q2 −2.35pct QoQ. Merging them into one branch would bury that fork entirely, and the fork is precisely the most tradable information today.

---

3. Master Ranking of Per-Stock Catalyst Strength (26 names, descending)

Column notes: "Yesterday's net fund amount" is on the THS market-wide basis; "Yesterday's LHB" is on the exchange's top-five-seat basis — different denominators, so both numbers are given where they conflict. "Open→Close %" = the move from yesterday's open to yesterday's close, measuring how much room is left if you execute the list at the open; it is the column to read first in this table.

---

4. Per-Stock Scoring Model (100 points)

Score breakdown for four samples (so readers can check scoring consistency):

---

5. Detailed Analysis of the Top 10 Names

1深南电路002916SZ Main Board ±10% | Watch closely | 70 pts (first draft was 84 pts and "priority deep-dive"; cut in QA, reasons below)

• ⚠️ The correction that has to come first: this name's draft rating rested on a factual error. The first draft treated the "Investor Relations Activity Record" of 8/27 18:28 as "first-hand incremental information inside the window", and on that basis gave it 9 points for expectation gap and a total of 84, first on the whole list. QA cross-checking confirmed two things:

  1. The company's 2026 interim report was actually released at 2026-08-26 18:41 (verified against the exchange disclosure stream; the interim report's timestamp precedes the record by a full trading day), so 8/27 was the first full trading day after the interim report, and that day's +4.90% was already the reaction to it;
  2. The three sets of data the first draft treated as "beyond the report body" appear verbatim in the body of the 8/26 interim report. Page 13 source text: "The printed circuit board business achieved principal operating revenue of RMB 8.552 billion, up 36.32% year on year, accounting for 55.91% of the Company's total operating revenue; gross margin was 36.85%, up 2.43 percentage points year on year." "The package substrate business achieved principal operating revenue of RMB 3.667 billion, up 110.76% year on year, accounting for 23.97% of the Company's total operating revenue; gross margin was 31.35%, up 16.20 percentage points year on year."; the same page also contains "successfully drove volume production of high-end products at multiple customers"; and page 12 already states that "key raw materials such as CCL and fiberglass cloth are showing tightening supply and accelerating price increases on the pull of AI compute demand". Conclusion: that record is a restatement of the interim report body and contains nothing beyond it. This brief's first-draft characterisation of it was wrong. Interim report PDF¹⁴

  The nature of this error deserves its own note: this brief used the "trace it back to the first disclosure date" test to downgrade 德明利 (001309, 7/15 pre-announcement), 联瑞新材 (688300, 8/15 interim report) and 大普微 (301666, 8/26 contract), yet it was only for the one name it most wanted to recommend that it never went back to check the interim report's disclosure timing. The test was skipped on the favoured name.

• The parts that still hold after correction (these were not overturned): H1 revenue RMB 15.296 billion (+46.33%), attributable net profit RMB 2.251 billion (+65.55%); package substrate revenue RMB 3.667 billion +110.76%, gross margin 31.35%, +16.20pct YoY (note: this is the H1 cumulative YoY basis; the interim report does not disclose Q2 single-quarter segment data, and the year-ago base was only about 15.15%, so this is low-base recovery); multi-customer volume production of high-end Guangzhou FC-BGA products; the Thailand plant and Nantong phase four are ramping, and the Wuxi AI compute electronic-circuit project has a 1-year construction period. The industrial logic has not changed; what changed is whether this information is new today — it is not.

• Asset impairment provisions (first-hand announcement, No. 2026-048, released 2026-08-26 18:41): H1 total provisions of RMB 363.5750 million (RMB 363.6 million), broken down as receivables RMB 137.4 million, inventory RMB 206.6 million and fixed assets RMB 19.6 million, with the source text stating it "will reduce the Company's total profit for January–June 2026 by approximately RMB 363.5750 million". These provisions are already included in the H1 statements — the RMB 2.251 billion of attributable net profit is the figure after deducting this RMB 363.6 million, not a fresh negative still to land; the first draft's description of it as "this brief's biggest information gap" was likewise wrong. What is worth noting is the provisioning attitude: it equals 16.2% of H1 attributable net profit, a sharp contrast with 德明利 (001309) provisioning only 0.194% against RMB 21.45 billion of inventory. The announcement notes it is unaudited. The content of the same-day "Announcement on Change of Accounting Policy" was not obtained by this brief and remains an unverified item.

• Valuation (omitted in the first draft, now added): PE (TTM) 59.68 / PB 13.79 (Tencent basis, post-interim-report). In a table sorted by catalyst strength, the first draft gave no valuation for precisely the highest-scoring name — a presentation bias.

• Technical sentiment: yesterday +4.90%, opened +3.15%, only +1.70% open→close; turnover 1.45%, the lowest of the 33 names in the candidate pool; funds +RMB 144 million, low-ranked within the AI chain.

• Final call: downgraded from "priority deep-dive" to "watch closely", total 84 → 70. Once "there is in-window incremental information" is deleted, the only reason it has left is one: yesterday's turnover of 1.45% was the lowest in the pool, i.e. the interim report has been out for two days and the share register has barely loosened. That still holds and still has value, but it is a share-register reason, not an information reason, and is not enough to support the top score and the sole "priority deep-dive". One contradiction also to be kept on record: its +RMB 144 million of fund flow yesterday was among the weakest in the chain (versus 长鑫科技 (688825)'s RMB 3.230 billion).

2金安国纪002636SZ Main Board ±10% | Watch closely · do not chase a gap-up | 76 pts

• Related news: 2026 interim report disclosed 2026-08-27 18:36. Source PDF¹
• First-hand data: revenue RMB 3.399 billion +65.74%, attributable net profit RMB 765.9 million +986.66%, ex-non-recurring RMB 770.4 million +955.34%, gross margin 33.14% (+21.83pct YoY), weighted ROE 19.39%, gearing ratio 47.38%.

  ⚠️ Base-effect note (added by QA): the year-ago base behind +21.83pct was only 11.31%, and the Q1 base behind the +10.64pct QoQ move was 26.44%; the +986.66% attributable growth is measured against a year-ago base of just RMB 70.5 million. These are CCL recovering off a cycle bottom, not a leap from a normal level — the direction is readable, the magnitude cannot be extrapolated. By the same token, the base behind 胜宏科技 (300476)'s Q2 gross margin of −6.72pct YoY is 2025Q2's 38.83%, a cycle-high base.

• Four pieces of evidence this brief regards as high quality:
  1. Ex-non-recurring (RMB 770.4 million) > attributable (RMB 765.9 million), with non-recurring items at −RMB 4.5 million. There is no one-off gain dressing up the profit — the exact opposite of 胜宏科技 (300476) the same evening.
  2. The volume/price split is unusually clear: the interim report source text reads "sold 30.7430 million sheets of various CCL, up 10.77% year on year; achieved operating revenue of RMB 3,326.6040 million, up 68.46% year on year" — volume up only 10.77% while revenue rose 68.46%, implying ASP +52.1%; roughly four fifths of this growth came from price.
  3. The company names its own moat: the interim report says "facing tight supply and rising prices in the key raw material electronic fiberglass cloth, the Company relied on its own electronic fiberglass cloth production to strengthen the security and stability of CCL raw material supply". This explains why, at the tightest point in electronic cloth, its operating cost rose only 24.95% (while revenue rose 65.74%).
  4. Zero interest-bearing debt (short-term and long-term borrowing line items are empty), with finance expense of −RMB 48.1 million (a net gain); CCL is 96.76% of revenue, and domestic revenue is 97.70% (almost no FX or tariff exposure); contract liabilities of RMB 152.6 million are +119% versus the start of the year, pointing to Q3 orders still being tight.
• ⚠️ But three negatives must be written just as prominently:
  1. The H1 cumulative numbers were already public in the 7/14 earnings pre-announcement (attributable net profit guided at RMB 730–820 million; actual RMB 765.9 million, landing at the 39.9 percentile, 1.18% below the midpoint). Yesterday's limit-up was not buying this interim report (which was only disclosed at 18:36), and the profit it delivers was announced 44 days ago. The only genuinely new information becomes visible after splitting out the quarter: Q2 single-quarter gross margin of 37.08%, +10.64pct QoQ — that was not in the pre-announcement.
  2. Limit-up yesterday but RMB 557 million of market-wide net fund outflow (inflow RMB 1.461 billion / outflow RMB 2.019 billion), while LHB top-five-seat net buying was +RMB 275 million. These two numbers cannot be divided by each other, and this brief does not infer "who was selling" from them — the THS "net fund amount" is a constructed quantity that infers active buy/sell direction from order-size tiers, while the LHB is the exchange's disclosed actual turnover of the top five seats, with different construction and different denominators. The conflict itself is an uncertainty signal. The only conclusion retained is one: on the day of the limit-up, the market-wide inferred basis recorded a large net outflow, which warrants caution; as to whether it was "distribution", this brief takes no position.
  3. PE (TTM) 50.7 sits at the 16.8% three-year percentile while PB 11.79 sits at the 93.2% three-year percentile — a 76pct gap. A low PE is not a cushion here; it is a reading that "earnings are standing at a cycle peak". Also: the private placement received CSRC registration approval on 2026-08-12 (CSRC License [2026] No. 1819), but the approval does not state issue size or price, and this brief has no reliable data on them, so it constitutes unpriced equity supply.
• Final call: watch closely, but explicitly do not chase a gap-up. This is the name in this brief with "the cleanest fundamentals and the dirtiest share register" — its interim report ranks in the top two of the pool on quality, yet yesterday's trading behaviour was distribution. The suggested handling is to wait for it to digest yesterday's loose chips, rather than taking a gap-up today.

3生益科技600183SH Main Board ±10% | Watch closely | 75 pts

• Related news: the 深南 record names "CCL… the tight-supply condition has continued to intensify", and 生益 is the undisputed domestic CCL leader, hence the most direct beneficiary of that naming. Its own interim report was disclosed 8/15 (old).
• Technical sentiment and funds: yesterday +8.43%, touched limit-up intraday then broke it, turnover value RMB 14.314 billion, net fund amount +RMB 1.251 billion. Across the whole CCL/resin group it is the only name with a large net inflow (金安 −RMB 557 million, 东材 −RMB 49 million, 宏昌 −RMB 73 million).
• Risk: no in-window news of its own; PE 65.92 / PB 18.51.
• Final call: watch closely. This brief regards it as more worth following than the batch of materials names that sealed their limits yesterday — precisely because it did not seal, yet real money came in. Verification points in §8.

4工业富联601138SH Main Board ±10% | Watch closely | 74 pts

• Related news: NVIDIA Q3 guidance $108.0B; AWS committed to buying 2 million GPUs and adopting the Vera CPU, partly integrated with Rubin. Overnight NVDA +8.74%.
• Bull logic: the core global AI server ODM, and the A-share whole-machine mapping closest to cloud capex. But A-shares already traded this news for a full day yesterday, so the expectation gap is limited.
• Technical sentiment: yesterday +5.43%, turnover only 0.75% — a trillion-yuan market cap (RMB 1,267.2 billion of float) rising 5.43% on 0.75% turnover means very light selling pressure.
• Final call: watch closely. The "highest certainty, lowest elasticity" name in this pool, better used as a yardstick for judging "is today an index-level compute rally or a thematic rally" than as a source of elasticity.

5星网锐捷002396SZ Main Board ±10% | Watch closely | 73 pts

• Related news: AWS's 2 million GPUs and Rubin integration → data-centre networking demand; limit-up +9.99% yesterday.
• Bull logic: switches are mandatory supporting equipment for GPU clusters, and the beneficiary path is more direct than optical modules (switch count scales linearly with cluster size).
• Technical sentiment and funds: limit-up, turnover value RMB 3.237 billion, turnover 13.51%, funds +RMB 363 million and LHB +RMB 226 million, positive on both bases — the only one of yesterday's 5 communications-equipment limit-ups where that is true (长飞光纤 (601869) −RMB 708 million; 楚天龙 (Chutian Dragon, 003040) is a digital-currency high-flyer; 长江通信 and 汇源通信 are too small).
• Final call: watch closely. The only name in the communications-equipment branch with clean money quality.

6澜起科技688008STAR Market ±20% | Watch closely (holding = betting on the reveal) | 72 pts

• Related news: the CNINFO disclosure-appointment calendar shows the 2026 interim report is scheduled for 2026-08-29, with "actual disclosure" blank as of writing — i.e. it will be published between this Friday's close and the weekend, and since 8/29 is a Saturday, the share-price reaction lands next Monday (8/31). Yesterday +10.06%, funds +RMB 2.095 billion, turnover value RMB 14.463 billion (#2 in the pool).
• ⚠️ A split in basis you must know before the reveal: the attributable range of +63.90%~+81.20% in the 7/17 earnings-increase announcement is widely quoted, but the ex-non-recurring range in the same announcement is only +14.5%~+32.9%, with non-recurring items about 32% of attributable profit. Yesterday's +10.06% priced the attributable basis; if tonight's ex-non-recurring lands at the bottom of the range, there is a risk of being falsified on the ex-non-recurring basis. Also, per public reporting on 7/17, the company said at the time that it was "fully cooperating with all investigation requirements of the Korean prosecutors' office"; this brief obtained no update on that matter, so it is an unverified item.
• A warning from this brief's sister sample: 德明利 (001309)'s H1 pre-announcement, released on 7/15, triggered three consecutive one-tick limit-downs — "earnings pre-increase" and "share price up" have already been proven not to be the same thing in this memory cycle. 澜起's own shares also fell more than 8% on the day it released its 7/17 pre-increase announcement (recorded in yesterday's pre-market brief).
• Final call: watch closely, but this must be said plainly: buying it today is not buying the logic, it is betting on an ex-non-recurring number that will only be published over the weekend and only priced next Monday. That single reason is why this brief does not put it in priority deep-dive.

7联瑞新材688300STAR Market ±20% | Watch only · do not chase a gap-up | 70 pts

• Related news: no in-window announcement of its own (both the interim report and the bonus-issue plan date to 8/15, priced 9 days ago). It is included because of its industrial position: electronic-grade silica micropowder is a core filler for CCL and package substrates, sitting at the intersection of branches 1 and 2.
• An evidence gap that must be stated: the shortage list named in the 深南 record is "CCL fiberglass electronic cloth and copper foil", and silica powder is not on it. This brief has no first-hand evidence that silica powder is equally tight; its rally is same-chain mapping, not being named as a beneficiary.
• Technical sentiment: yesterday +20.00%, opened only +1.89%, +17.77% open→close — 89% of the move happened intraday. A reader who executed the list at yesterday's open captured almost the entire move; entering today faces a completely different cost basis.
• Final call: downgraded from the first draft's "watch closely" to "watch only · do not chase a gap-up". The downgrade follows from confirming, after further checks, that it has no in-window news, is not on the shortage list, and already realised its move yesterday — when all three hold at once, a good industrial position is not a reason to buy.

8德明利001309SZ Main Board ±10% | Watch only | 58 pts (first draft was 78 pts and "watch closely"; cut after checking)

• Related news: on 2026-08-27 after the close it disclosed the 2026 interim report, the non-recurring items review report, the interim audit report, and the "2026 Interim Profit Distribution and Capital Reserve Conversion Plan" (RMB 15.00 cash per 10 shares + 4 bonus shares per 10 shares from capital reserve; total share capital 226,842,870 → 317,580,018 shares).

• ⚠️ The first point that determines the rating: this interim report contains no new information on the H1 cumulative basis, and that "old information" once produced three limit-downs. The company already disclosed an earnings pre-announcement on 2026-07-15:

  Item	7/15 guided range	Interim actual	Landing point
  Operating revenue	RMB 16–18 billion	RMB 16.911 billion	46 percentile
  Net profit attributable to parent	RMB 5.7–6.5 billion	RMB 6.017 billion	40 percentile, −1.36% versus the midpoint
  Ex-non-recurring attributable	RMB 5.649–6.449 billion	RMB 5.966 billion	40 percentile

  Table scrolls horizontally

  And after the pre-announcement: three consecutive one-tick limit-downs on 7/15, 7/16 and 7/17 (only 8,321 lots traded all day on 7/16), with the July high of 954.19 → the 7/29 low of 334.00, a peak-to-trough of −65.0%; the 8/27 close of 429.76 is still 55.0% below the July high. The earnings this interim report delivers were already fully priced by the market with three limit-downs back in mid-July.

• ⚠️ Second point: the only significant new information in the interim report is negative — Q2 collapsed. Derived by subtracting Q1 from H1 (and cross-checked as consistent with the interim report body's "Q2 alone revenue RMB 9.372 billion, +24.33% QoQ"):

  Metric	Q1 2026	Q2 2026	QoQ
  Operating revenue	RMB 7.538 billion	RMB 9.372 billion	+24.33%
  Gross margin	57.42%	38.84%	−18.58pct
  Net profit attributable to parent	RMB 3.346 billion	RMB 2.671 billion	−20.18%
  Ex-non-recurring attributable	RMB 3.334 billion	RMB 2.632 billion	−21.04%

  Table scrolls horizontally

  Revenue accelerating, profit down 20% QoQ and gross margin collapsing 18.58 percentage points QoQ is the textbook shape of "volume up, price down / costs catching up" (Q1 filled high-priced orders with low-cost inventory; in Q2 the expensive wafers entered cost of sales). It is worth noting that the company voluntarily disclosed Q2 revenue +24.33% QoQ in the body, but nowhere disclosed Q2 single-quarter profit or single-quarter gross margin.

• ⚠️ Third point: inventory of RMB 21.45 billion with a 0.194% impairment provision. Inventory carrying value is RMB 21.454 billion (+203.4% versus the start of the year; for the basis behind that versus revenue +311.55%, see below), 64.95% of total assets; raw materials plus work in progress are 81.77% of it; the impairment provision is only RMB 41.69 million = 0.194%. The company states its provision ratios for 2023/2024/2025/2026H1 were 1.34% / 1.50% / 0.50% / 0.19%, the lowest in four years. Inventory turnover days lengthened from 211 days to 286.6 days. The company itself writes in its risk factors: "if raw material prices fall sharply in the future, the Company may need to make large inventory impairment provisions, sharply reducing profits, and in extreme circumstances this could cause the Company to record a loss".

• ⚠️ Fourth point: cash and debt. Net operating cash flow was −RMB 6.736 billion (cash conversion of net profit at −1.12×), and the 2023/2024/2025/2026H1 series is −RMB 1.015 / 1.263 / 2.241 / 6.736 billion, negative for four straight periods and accelerating; net interest-bearing debt is RMB 10.740 billion (RMB 3.835 billion at the start of the year); quick ratio 0.60 (excluding inventory, current assets do not cover current liabilities).

• The positives that must be given at the same time: ① non-recurring items are only RMB 51.11 million, 0.85% of attributable net profit — clean profit quality, with ex-non-recurring and attributable profit from the same source; ② enterprise-grade storage H1 revenue of RMB 4.815 billion, 28.5% of the total, quantified and disclosed for the first time, the one substantive positive in this interim report; ③ controlling shareholder Li Hu has undertaken not to reduce his holdings for 12 months from 2026-07-27.

• Valuation (recomputed on the post-interim-report basis): PE (TTM) 14.29, at the 0.0% percentile since listing — below the stock's own historical minimum of 18.8 since listing; while PB 9.97 sits at the 60.7% percentile since listing. "PE at a historical low + PB above the historical median" is the textbook shape of a cyclical whose earnings have peaked: the low PE comes from a cycle-peak denominator, not a cheap numerator. Peers 江波龙 (301308) at 13.85/7.46 and 佰维存储 (688525) at 13.64/8.85 show this is a sector-level cycle-pricing state, not an individual trait.

• Final call: downgraded from the first draft's "watch closely" to "watch only". Yesterday's agreement across volume, price, funds and the LHB is real, but it is built on earnings that were made public 44 days ago and produced three limit-downs at the time; and of the three genuinely new pieces of information in the interim report, two are negative. The 4-for-10 bonus issue is a theme, not value (a pure accounting reclassification; per-share metrics dilute proportionally after the ex-rights adjustment).

9长鑫科技688825STAR Market ±20% | Watch only (wait for the interim report) | 57 pts

• Related news: yesterday's net fund amount was +RMB 3.230 billion, #1 market-wide; at the same time, on 8/27 it announced the exercise of the IPO over-allotment option (greenshoe); the appointed interim-report disclosure date is 8/29 (Saturday, so the reaction lands next Monday).
• Two facts pointing in opposite directions: the money is the strongest in the market, but exercising the greenshoe means the free float increases, a supply-side negative; and the interim report is revealed tonight. PE 142.34 / PB 27.11.
• Final call: watch only. It is the cleanest gauge of "whether the memory line is getting incremental money" (large cap, no thematic character, no high-flyer sentiment distortion). Today the suggestion is to use it as a dashboard, not as a position.

10胜宏科技300476ChiNext ±20% | Watch only | 52 pts (first draft was 71 pts and "watch closely"; cut after checking)

• Related news: 2026 interim report disclosed 2026-08-27 20:12. This stock has no earnings pre-announcement (verified: none of the 4,879 records in Eastmoney's full pre-announcement table is 300476; nor is there any pre-announcement/flash result among the 25 announcements listed between 7/1 and 8/28), so the interim report is a first disclosure and its information increment should have been larger than 金安国纪 (002636)'s — but the content points negative. Source PDF³

• ⚠️ First point: the growth in attributable profit is "borrowed". Revenue RMB 11.629 billion +28.77%, attributable net profit RMB 2.857 billion +33.30%, but ex-non-recurring only RMB 2.418 billion +12.50%. The entire 20.8pct scissors gap comes from RMB 439 million of non-recurring items, the single largest of which is RMB 587.9 million of "gains from fair value changes on financial assets and financial liabilities held by a non-financial enterprise". Page 20 of the interim report gives the cause: "mainly due to the increase in fair value of shares held from a placement", with the company self-assessing "Sustainable: No", equal to 18.06% of total profit. This is a mark-to-market gain on a shareholding unrelated to the core PCB business.

• ⚠️ Second point: Q2 single-quarter ex-non-recurring turned negative.

  	2026Q1	2026Q2	QoQ	2025Q2	YoY
  Revenue	RMB 5.519 billion	RMB 6.110 billion	+10.7%	RMB 4.719 billion	+29.5%
  Gross margin	34.46%	32.10%	−2.35pct	38.83%	−6.72pct
  Attributable net profit	RMB 1.288 billion	RMB 1.568 billion	+21.7%	RMB 1.222 billion	+28.3%
  Ex-non-recurring net profit	RMB 1.257 billion	RMB 1.161 billion	−7.7%	RMB 1.225 billion	−5.3%
  Non-recurring items	RMB 31 million	RMB 408 million	—	—	—

  Table scrolls horizontally

  Q2 single-quarter ex-non-recurring was −7.7% QoQ and −5.3% YoY, the first single-quarter decline in ex-non-recurring profit since this AI upcycle began; the apparent "Q2 attributable +21.7% QoQ" is entirely propped up by that RMB 408 million mark-to-market gain. Cash flow moved in step: Q2 operating cash flow as a share of attributable profit was only 50.7% (Q1: 164.3%), and Q2 cash received from goods sold as a share of revenue was 79.7% (Q1: 99.3%) — H1 on a full-period basis shows OCF/attributable profit of 101.9%, which looks healthy, but all the increment sits in Q1.

• ⚠️ Third point: it is on the cost-bearing side of this round of price hikes, not the benefiting side. The change table in the interim report: operating cost +34.83% outpaced operating revenue +28.77%, and that 6.06pct difference is precisely the source of the −3.00pct in gross margin; the risk chapter reads: "the principal raw materials required for the Company's production are CCL, prepreg, copper balls and copper foil… where raw material prices rise sharply, the Company will raise product prices accordingly, but there is a certain lag in passing this downstream". 胜宏's raw material is exactly 金安国纪 (002636)'s product — two interim reports corroborated the destination of the profit on the same evening. There is a second headwind as well: finance expense of RMB 296.7 million versus RMB 37.3 million a year earlier, +694.35% (FX losses and interest expense), against a high export share. ⚠️ Basis note: this interim report ticks "not applicable" for "products or services accounting for more than 10%", so it nowhere quantifies the export share, disclosing only qualitatively that "export revenue is substantial in amount, with export customers and imported regions concentrated in Hong Kong, Taiwan China, Japan, Korea, Europe and the US"; the frequently quoted 76.83% is on the 2025 annual report basis, has not been re-verified for this period, and is flagged as to be verified.

• ⚠️ Fourth point: PB 6.82 is an accounting appearance, not a de-rating. Share capital rose from 873 million shares at the end of Q1 to 983 million at the end of H1 (+110.2 million shares), capital reserve rose from RMB 5.410 billion to RMB 25.328 billion (+RMB 19.919 billion), and Q2 single-quarter "cash received from investment absorbed" was RMB 20.064 billion, which the interim report explains as "proceeds from the overseas issuance and listing of shares" (cross-verified: the H-share line 02476.HK exists, and the H-share capital matches the A-share capital increment exactly). Net assets jumped from RMB 17.404 billion to RMB 37.909 billion, and of that RMB 20.5 billion increment, RMB 20 billion is cash just raised, not accumulated from operations — computed on pre-raise net assets, PB = 14.9× (about the 90th percentile). By the same token, weighted ROE falling from 21.48% to 11.80% is mainly a denominator effect.

• One information gap: the interim report ticks "not applicable" for "products or services accounting for more than 10%", i.e. the company chooses not to break out its product mix; and Eastmoney's revenue-composition database has its latest record for this stock stopping at the 2025 annual report with only a single "PCB manufacturing" category. The AI-related revenue share therefore has no basis — any circulating "AI accounts for X%" does not come from this interim report.

• Final call: downgraded from the first draft's "watch closely" to "watch only". The first draft treated "the interim report only landed last night = a positive expectation gap" as a bullish reason, but the sign of an expectation gap depends on content, not on timing; the content of this first-disclosure interim report points negative.

---

6. Pass List

On the limits of using "net fund outflow" as a Pass reason: this brief uses "limit-up + market-wide net fund outflow" to identify distribution, but this is a single-day reading. Single-day extremes in main-force fund flow tend to mean-revert and cannot serve as a directional test on their own; this brief gives a second, independent piece of evidence in every such case (old catalyst / high turnover / gap leaving no room), so no stock here is passed on the fund reading alone. By the same token, 金安国纪 (002636) in §3 is likewise "limit-up + net outflow", but because the quality of its interim report constitutes independent positive evidence, this brief leaves it in a watch slot rather than passing it — the test must be applied symmetrically.

---

7. Intra-Branch Ranking

7.1 CCL / electronic cloth / epoxy resin

• Hardest names: 金安国纪 (002636) (fundamentals) / 生益科技 (600183) (funds). That these two attributes sit on two different stocks today is what most needs watching in this branch.
• Who is following: 东材, 宏昌 (both of whose own announcements are out of window).
• The core contradiction in this branch: the logic is hardest (a first-hand naming from the downstream buyer plus first-hand delivery from the upstream seller), but yesterday's money was dirtiest (3 of the 4 limit-up names saw net outflows). The handling is to keep the branch's high ranking (logic) while downgrading the sealed limit-up names (money), leaving only the clean-money 生益科技 (600183) at the top.

7.2 Package substrates / advanced packaging materials

7.3 Memory

• Why the whole branch is down-weighted: MU was −0.32% overnight and −1.43% after hours while SMH was +3.10% — overseas has priced compute and memory separately, so the A-share memory line got no external confirmation last night; layered on top of 德明利 (001309)'s Q2 single-quarter gross margin of −18.58pct, a first-hand negative reading from a domestic leader.

---

8. Opening-Bell Verification Signals for Today

8.1 Call-auction signals

• 深南电路 (002916): opening +2%~+5% with turnover expanding to 2%~4% would be healthy (yesterday's 1.45% turnover was the lowest in the pool, so it has the most room to expand). If it gaps up more than +7%, the record has already been priced by sentiment and priority deep-dive falls to watch.
• 金安国纪 (002636): the RMB 557 million of net outflow yesterday is the sole reason this brief downgraded it. Today's test is whether funds can turn positive — if flows turn net positive after the open and turnover does not exceed yesterday's 7.15%, yesterday's distribution has been digested and the call can be raised; if it again shows "up/flat + net outflow", the distribution is confirmed as unfinished.
• 联瑞新材 (688300) / 赛微电子 (300456) / 晶丰明源 (688368): 73%–92% of yesterday's gains in these three happened intraday, so a gap-up today is yesterday's buyers' exit window; do not chase any gap-up above 5%.
• 德明利 (001309): watch how much the bonus-issue proposal amplifies sentiment. If it opens one-tick limit or gaps up more than +7% with a sealing order / turnover value below 0.15, treat it as a sentiment seal rather than a money seal.

8.2 Sector signals

• The divergence-return test for the CCL group (this brief's first observation point): whether 生益科技 (600183) can re-seal its limit after breaking it yesterday. The next-day re-seal rate after a broken limit is the single most sensitive indicator of this branch's strength. If 生益 seals while 金安国纪 (002636), 东材 (601208) and 宏昌 (603002) — the three limit-up names with net outflows yesterday — break theirs, that confirms "real logic, dirty register" and money concentrating into the leader, which is healthy; the reverse would be an ebb.
• Verifying the PCB manufacturing vs materials fork: if 胜宏科技 (300476) and 沪电股份 (002463) weaken today while 生益科技 (600183) and 金安国纪 (002636) strengthen, the "profit is stuck upstream" reading from last night's three interim reports is confirmed by the tape, and that fork can serve as next week's main framework; if PCB manufacturing leads instead, the market is still pricing "AI-PCB as one undifferentiated chain" and this brief's split has not yet been accepted.
• Limit-up counts in components / semiconductors: 4 and 5 respectively yesterday (both 0→0→N Day 1). A combined total of ≥6 today confirms diffusion; ≤2 means yesterday's Day 1 was the only day.
• Volume in the core names: the semiconductor industry saw net inflows of RMB 25.193 billion yesterday (32.5% of the market-wide total). If that falls below RMB 10 billion today, the money behind the main theme is leaving.

8.3 Single-name signals

• Whether 工业富联 (601138) turnover can break 1% (0.75% yesterday, 0.37% the day before). This is the test for "index-level compute rally vs thematic rally", carried over for a second day.
• 星网锐捷 (002396): yesterday it was the only one of the 5 communications-equipment limit-ups positive on both fund bases; watch whether it can post a second consecutive day of net inflow today.
• 长鑫科技 (688825): net inflow of RMB 3.230 billion yesterday, #1 market-wide; watch whether it stays positive for a second consecutive day today — the cleanest gauge of incremental money in the memory line.
• 沪电股份 (002463): open→close +0.02% yesterday, a second consecutive day of lagging. If it lags again on heavy volume today, "export-heavy PCB makers with no package-substrate offset are being squeezed" gains a third independent sample (the first two being last night's batch: 奥士康 (002913) revenue +16.45% / ex-non-recurring −57.17%, and 胜宏科技 (300476)'s Q2 single-quarter ex-non-recurring −7.7% with gross margin −2.35pct QoQ).

  A self-imposed limit is required: 深南电路 (002916)'s PCB segment gross margin rose +2.43pct YoY, which is a counter-example to that claim, and its absolute level of 36.85% is above 胜宏's H1 33.22%, while its revenue of RMB 8.552 billion is the same order of magnitude as 胜宏's RMB 11.629 billion — this is not a peripheral business. So "the entire PCB manufacturing layer is being squeezed" does not hold within the sample, and this brief's downgrade targets only the practice of buying that layer indiscriminately. One place where a self-imposed limit is required: 深南电路 (002916)'s PCB segment gross margin is +2.43pct YoY (rising) and cannot be used as evidence that "PCB manufacturing is being squeezed" — it is precisely a counter-example. This brief's squeeze judgement rests solely on 胜宏科技 (300476)'s single-quarter data, a sample size of 1; 深南's existence shows there is differentiation within the same link, and manufacturers with pricing power can still expand gross margin. Hence §2 demoting PCB manufacturing to #6 targets the practice of "buying PCB manufacturing indiscriminately", not an assertion that every company in that link is deteriorating.

8.4 Risk signals

1. ⚠️ Two reveals, but the payoff lands next Monday: the appointed interim-report disclosure date for both 澜起科技 (688008) and 长鑫科技 (688825) is 8/29, and 8/29 is a Saturday — which means the interim reports will be published between this Friday's close and the weekend, with the entire share-price reaction landing next Monday (8/31); they are not a driver during today's session. These two happen to be the #1 and #2 net fund inflows of yesterday, so holding them into today's close = taking on an event exposure that spans the weekend and cannot be adjusted during the week. For 澜起 the key is where within the +14.5%~+32.9% range the ex-non-recurring number lands, not the attributable figure.
2. ⚠️ Peak disclosure density (basis corrected): the 698 companies whose actual disclosure date is marked 8/28 landed last night; 951 companies market-wide have still not disclosed and must all finish before 8/31 (next Monday, the statutory deadline). Because 8/29 is a Saturday and 8/30 a Sunday, whichever of those 951 disclose over the weekend will have their entire share-price reaction pile into next Monday's open — so holding into this Friday's close means taking on an unprecedented-density earnings event exposure spanning the weekend.
3. ⚠️ Systematic risk in the reporting basis (the one method this brief most wants to leave readers with): of last night's three interim reports, two had beautiful H1 cumulative numbers while Q2 alone was deteriorating (德明利 (001309) attributable −20.18% QoQ, 胜宏 ex-non-recurring −7.7% QoQ). More than 1,600 companies still report between today and next Monday, so please do the "H1 minus Q1" step on every one of them before drawing a conclusion — the cumulative basis hides the quarterly inflection.
4. The overseas fork has not healed: if MU keeps weakening tonight while NVDA strengthens, the divergence between A-share memory and compute will widen next Monday.
5. Gap-up-then-dump risk is concentrated in yesterday's limit-up distribution names: 金安国纪 (002636), 东材科技 (601208), 宏昌电子 (603002), 嘉立创 (001232), 长飞光纤 (601869).
6. Jackson Hole: Fed Chair Kevin Warsh delivers his first keynote since taking office tonight (Friday morning US time). The event falls after the A-share close; the impact will show up in next Monday's gap, and it is not a driver during today's session.
7. Testing the ebb in seeds: already trading day 8, with 万向德农 (600371) on 3 consecutive limit-ups. If the highest board stalls at 3 and the limit-up count shrinks from 4 to fewer than 2, treat that as the start of the ebb.
8. The policy shock in grid equipment (recorded in the first draft as an "unresolved event", now established): 思源电气 (002028)'s −10% yesterday was caused by the 8/26 US executive order restricting foreign-made bulk-power system equipment, and is a sector-level rather than a company event. The first draft's verification point — "only a second limit-down would show this is sector-level" — is therefore void. Three things to watch instead today: ① whether 思源's clarification of "no more than 2% US exposure" can stop the decline; ② whether yesterday's counter-trend limit-ups 杭电股份 (Hangzhou Cable, 603618, second board), 华盛昌 (CEM, 002980) and 日丰股份 (Rifeng, 002953) keep attracting money on "low US exposure" — i.e. whether the sector can differentiate into a "no US exposure" side; ③ the DOE's 120-day rulemaking window means policy uncertainty will persist, so the sector's valuation discount will not clear in the short term. Qualifier: there is precedent for such bans being rescinded, the implementing rules are not out yet, and the final scope is undetermined.

---

9. Final Recommendation

9.1 The 5 names most worth watching today

⚠️ There is no "priority deep-dive" today. The first draft listed 深南电路 (002916) as the sole priority deep-dive on the grounds that "the 8/27 evening IR record contained information beyond the interim report body"; QA cross-checking against the report source text confirmed that basis does not hold (the interim report was disclosed 8/26 18:41 and the record restates the body), so the stock has been downgraded to "watch closely". This brief chooses to shorten its certainty rather than swap in another name to fill the top tier.

Note on the structure of the list: all 5 are "watch closely", with no "priority deep-dive". Three trade-offs readers should note: ① none of yesterday's best-looking materials/PCB limit-up names (东材科技 (601208), 宏昌电子 (603002), 嘉立创 (001232), 长飞光纤 (601869)) made a recommendation slot, because the limit-ups came with market-wide net outflows and their own catalysts are mostly old news from 4–12 days ago; ② of the six companies that reported last night, only 金安国纪 (002636) made a recommendation slot — 德明利 (001309), 胜宏科技 (300476) and 中芯国际 (688981) are listed "watch only", and 奥士康 (002913) is a straight Pass; ③ this brief's #1 and #2 are two different stocks: the hardest fundamentals are 金安国纪's, the cleanest register is 生益科技's, and today those two attributes are not on the same name.

9.2 The 3 strongest branches today

Why memory dropped out of the top three (it was #1 in yesterday's recap): three independent pieces of evidence all point to "this is cycle delivery, not a starting point": ① MU was −0.32% overnight and −1.43% after hours while SMH was +3.10% — overseas has priced compute and memory separately; ② 德明利 (001309)'s Q2 single-quarter gross margin was −18.58pct QoQ and attributable net profit −20.18% QoQ, a first-hand negative reading from a domestic leader; ③ 德明利 (001309) / 佰维 / 江波龙 (301308), all three module names, show "PE at an extremely low percentile + PB at a high percentile" simultaneously. Logical strength and where you are in the cycle are two different things.

9.3 Directions not to chase today

1. The five "limit-up + net fund outflow" names from yesterday: 长飞光纤 (601869, −RMB 708 million), 宏昌电子 (603002, −RMB 73 million), 嘉立创 (001232, −RMB 66 million), 东材科技 (601208, −RMB 49 million), and 金安国纪 (002636, −RMB 557 million, which this brief still keeps in a watch slot but explicitly says not to chase on a gap-up). Not because they are high, but because the limit-up and the distribution happened at the same time, and the first four have their own catalysts dating back 4–12 days.
2. The whole PCB manufacturing layer: 胜宏科技 (300476) Q2 single-quarter ex-non-recurring −7.7% QoQ; 沪电股份 (002463) open→close +0.02% yesterday, a second straight day of lagging. This layer is the cost-bearing side of this round of price hikes, doubly confirmed by the companies' own interim-report text and numbers.
3. High-elasticity names whose gains yesterday were concentrated intraday: 联瑞新材 (688300, open→close +17.77%), 晶丰明源 (688368, +18.40%), 赛微电子 (300456, +14.63%), 协和电子 (Xiehe Electronics, 605258, +10.27%). The logic may still hold, but the current move was fully realised yesterday, so a gap-up today is yesterday's buyers' exit window.
4. Names where the gap ate all the room: 大普微 (301666, opened +17.38%, only +2.23% open→close), 万向德农 (600371, opened +9.54%, +0.38% open→close). Both catalysts are out of window and were fully priced yesterday.
5. High-flyer relays: 深中华 A (000017, 6 boards), 楚天龙 (003040, 5 boards, which has itself announced digital-currency revenue is under 5%), 康盛股份 (002418, 3 boards). All three have sealing-order / turnover ratios of 0.022–0.075, far below the 0.15 genuine-seal threshold, and all three had net fund outflows.
6. Shrinking-volume one-tick boards: 海鸥住工 (002084, turnover value RMB 28 million), 捷荣技术 (002855, turnover value RMB 19 million). Sealing-order / turnover ratios of 4.9–11.9 look extremely strong but are actually shrinking-volume one-tick boards with no bid support if the limit opens.
7. Names with US exposure in grid equipment. The sector was −0.01% yesterday with RMB 1.742 billion of net outflow, and the leader 思源电气 (002028) hit limit-down, because on 8/26 the US signed an executive order restricting foreign-made bulk-power system equipment (large power transformers, utility-scale inverters, storage, high-voltage circuit breakers and so on). A-shares already priced this for one day yesterday, so today is Day 2; policy uncertainty will not clear before the DOE issues its 120-day implementing rules. Note the distinction: the order does not apply to local distribution facilities, and US revenue shares differ enormously between companies (思源 says no more than 2%), so it is wrong to treat "grid equipment" as one undifferentiated block — but equally wrong to grab for a rebound before the rules are clear.
8. Banks and dividend names: banks were −1.05% yesterday (the only industry down more than 1%) with RMB 5.392 billion of net outflow, continuing to serve as the ATM under the main theme's siphon effect.

9.4 The final one-sentence call

**The real information today is not in any news item but in the directional fork revealed by the batch of interim reports that landed last night: 金安国纪 (002636) says its CCL volume rose only 10.77% while revenue rose 68.46% (ASP up 52.1%); 深南电路 (002916)'s interim report says upstream electronic cloth and copper foil show "tightening supply and accelerating price increases"; 胜宏科技 (300476) says its own raw materials are precisely CCL, prepreg and copper foil and that "there is a certain lag in passing this downstream"; and the most glaring number comes from 奥士康 (002913), disclosed in the same batch last night and missed entirely by this brief's first draft — a pure PCB manufacturer with revenue +16.45% and ex-non-recurring net profit −57.17%.****But what deserves to be remembered today even more is the boundary of that thesis: ranking the three companies by export exposure coincides exactly with ranking them by supply-chain position, and in 深南电路 (002916)'s own interim report domestic gross margin is +7.36pct while overseas is only +0.73pct — the same company, the same reporting period, a 6.63 percentage point internal fork, which "upstream price hikes" cannot explain but "RMB appreciation compressing export selling prices" can; so "profit is stuck upstream" can only be taken today as a hypothesis awaiting verification, not a proven conclusion — same direction does not establish causation.**As for the trading layer, three things are clear: first, the market already bought 金安国纪 (002636), 东材科技 (601208) and 宏昌电子 (603002) to limit-up yesterday on the "materials price hike" logic, yet all of them showed net outflows on the market-wide inferred basis, while the name that actually saw large buying was the one that failed to seal, 生益科技 (600183, +RMB 1.251 billion) — so whether 生益 can re-seal the day after breaking its limit is the first answer today's open will give; second, the memory leg was absent overseas last night — SMH +3.10% while MU −0.32% — and in that seemingly glorious 德明利 (001309) interim report the H1 numbers had already been made public once on 15 July, buying three one-tick limit-downs at the time, while what is genuinely new is that Q2 single-quarter gross margin collapsed 18.58 percentage points QoQ and RMB 21.45 billion of inventory carries only a 0.194% impairment provision; third, and most easily overlooked — the interim reports of 澜起科技 (688008) and 长鑫科技 (688825) are set for 29 August, which is a Saturday, and 951 companies market-wide have yet to disclose, all of which must land before next Monday, so every position held into today's close carries not today's risk but an earnings exposure that spans the weekend and cannot be adjusted during the week.

---

10. Pre-Publication QA Revision Log

This section discloses the content of this brief that was overturned and rewritten before publication by two rounds — fundamentals checking and independent risk-control QA. They are listed because these revisions changed the recommendation list, and readers have a right to know which conclusions were rewritten after being overturned.

The volume of revision in this brief is unusually large, and item A1 in particular dismantled the core rationale for the first draft's top recommendation slot. Items 1–11 are the fundamentals-check round, items A1–A10 are the risk-control QA round; among them, A1, A3, A4 and A5 are factual errors or major omissions, A2 narrowed the strength of the brief's core thesis, and A6 and A7 are cases where this brief violated the very basis discipline it had declared.

---

⚠️ Risk warning: this list is a pre-market information review and observation only and does not constitute investment advice. A-share volatility is extremely high, and automatically generated content may contain timeliness gaps or supply-chain mapping errors; it must not be used directly as a basis for trading.