A-Share Pre-Market Brief | 2026-08-24 Monday

Scan window: 2026-08-21 (Friday) 15:00 close → 2026-08-24 (Monday) 08:20, covering the full weekend. Quote basis: East Money push2delay endpoint, 2026-08-21 closing prices, local read moment 2026-08-24 07:0x–08:0x (Beijing time). US equities are 2026-08-21 closes (Beijing time 8/22 04:00, timestamps verified).

⚠️ One basis disclosure for this issue, please read first: the "P/E" field returned by the quote endpoint is a forward-annualised PE (half-year profit × 2), not TTM. For three key names the gap is enormous — 诺德股份 (Nuode) is 90.55 on the annualised basis but −151.31 on TTM (it actually lost money over the past 12 months); 中际旭创 (Innolight) is 40.40 annualised but 53.93 TTM; 星网锐捷 (Star-net) is 33.84 annualised but 44.29 TTM. Every valuation judgement below is labelled with its basis; PE figures in the master table that were not individually recomputed are all on the annualised basis, useful only for rough cross-sectional ranking, not as a valuation conclusion.

---

0. Today in One Sentence

The strongest catalysts are three pieces of primary material that landed only over the weekend: 中际旭创 (Innolight)'s 8/22 interim report (H1 net profit attributable to parent RMB 13.651 billion, +241.70%) plus its Sunday-evening 20:00 institutional call on 8/23; 长存控股 (Changcun Holdings / YMTC)'s IPO filing accepted after hours at 23:10 on 8/21; and Alibaba's 8/23 placement of HKD 80 billion with 100% of net proceeds going into AI infrastructure. Separately, the State Council executive meeting on 8/21 at 19:40 heard a report on the construction of next-generation communications networks.

The strongest branch is not "next-generation communications networks" — it is "compute materials (CCL/PCB)." The evidence comes from 中际旭创 (Innolight) management's own words to roughly 150 institutions on Sunday evening: "in H1 the supply of core materials such as optical chips, electrical chips and PCB remained fairly tight," "some materials did see price increases," "we expect suppliers' capacity to be released only gradually from H2 this year into H1 next year." This is a downstream leader's third-party endorsement of upstream pricing power — and upstream, 生益科技 (Shengyi Technology)'s H1 CCL gross margin has already risen 5.47pp year on year to 29.16%, with operating cash flow up 53.44% year on year to RMB 2.983 billion, covering 91% of net profit attributable to parent — the only name among the twenty-plus stocks covered here whose cash flow is improving (中际旭创 (Innolight) OCF/net profit is just 0.13, 星网锐捷 (Star-net) −RMB 3.567 billion, 华海清科 (Hwatsing) −81.44%, 诺德股份 (Nuode) −RMB 1.094 billion). More importantly: PCB/CCL is the only link in this cycle that feeds off both overseas CSP and domestic cloud capex — 中际旭创 (Innolight) draws about 95% of revenue from overseas, Alibaba's HKD 80 billion flows mainly to domestic servers and switches, and CCL supplies both sides.

Direction of flows: from optical-module assemblies already worth RMB 1.05 trillion in market cap, out toward upstream materials whose valuation and chart position still have room. On Friday the optical-communications-module concept saw main-force net inflow of +RMB 10.216 billion and CPO +RMB 10.034 billion, with 中际旭创 (Innolight) (+RMB 3.595 billion) and 新易盛 (Eoptolink) (+RMB 2.529 billion) taking the top two single-stock inflow slots; components saw +RMB 4.490 billion and printed circuit boards +RMB 4.118 billion, while the PCB group posted zero limit-ups on Friday — sentiment has not yet peaked.

Pre-market read: strong, but today's main thing to guard against is "buying the right branch for the wrong reason." Three points must be spelled out:

1. The State Council meeting has no causal link to Friday's telecom-equipment limit-up wave. 星网锐捷 (Star-net) sealed its limit at 09:44:30 on 8/21, roughly 10 hours before the State Council meeting (19:40). The meeting can only act on today and later, and the official readout mentions no investment size, no timetable, and names none of 5G-A / 6G / 10G optical networks.
2. 中际旭创 (Innolight)'s "margin jump" barely exists on a sequential basis. Q1 gross margin 46.06% → Q2 46.42%, only +0.36pp QoQ; of the +6.12pp QoQ rise in net margin, about 5.2pp comes from fair-value changes (H1 total RMB 537.12 million versus −RMB 0.8244 million in the prior-year period — almost entirely new this period), a lower effective tax rate, a lower minority-interest share, and R&D expense down 21.2% QoQ. And Q2 standalone operating cash flow was −RMB 1.568 billion (all of H1's RMB 1.800 billion came from Q1).
3. NVIDIA's FY2027Q2 results are released after the US close on 8/26 (Eastern), i.e. the early hours of 8/27 Beijing time, so the A-share reaction day is Thursday 8/27, not Wednesday.

---

1. News Overview

This Week's Calendar (verified; drives position pacing)

One "News Item" Discarded Entirely

Jiemian News's "Cailianshe Morning News Digest for August 23" contains "Xiaomi Q2 revenue RMB 45.236 billion," "P2P compliance inspections to be completed by end-December 2018," and "Shanghai market daily turnover below RMB 100 billion" — it is a 2018 article from a recurring column of the same name. Recurring-column headlines carry no year and are extremely easy to swallow wholesale as same-day news; this issue discards that source entirely.

⚠️ Data-collection failures and limitations, not for clients:

• East Money push2.eastmoney.com / 82.push2.eastmoney.com returned 502 and RemoteDisconnected; everything was rerouted through push2delay.eastmoney.com (cross-checked as consistent against the SSE Composite at 3905.20 and the 8/21 recap).
• The limit-down pool / failed-limit pool push2ex endpoint returned rc=206, data=None, so this issue could not re-pull the 8/21 limit-down (13) and failed-limit (18) counts; the 8/21 recap figures are carried over without re-verification. The limit-up pool showed tc=54 and 54 were actually pulled (validation OK), 1 fewer than the 55 recorded in the recap, presumably an after-hours adjustment.
• datacenter.eastmoney.com's RPT_F10_FINANCE_MAINFINADATA returned an empty array for every test code; the workaround was to download each interim-report PDF and parse it with pypdf (the environment has no pdftotext/poppler, and the Read tool's paginated rendering was also unavailable).
• The quote endpoint's f9 field is annualised PE, not TTM. This was caught by a sub-agent during verification; the first draft of this issue had written an incorrect valuation judgement on that basis, and before finalising, a basis disclosure was added across the whole table and the three key names were recomputed. This is the closest this run came to an error.
• Search engines are badly lagged on 8/22–8/24 indexing (repeatedly returning "today is August 23" as well as the 2018 same-name column article); weekend information came mainly from the np-anotice-stock filing feed and the original PDFs.
• The composition of 拓荆科技 (Piotech)'s H1 total profit +1601.84% could not be broken down item by item; this issue lists the raw figure, flags the low base, and draws no conclusion about its profit quality.

---

2. Strongest Positive Branches, Descending

Three Points That Need Spelling Out

① Friday's 6 telecom-equipment limit-ups were not driven by the State Council meeting. 星网锐捷 (Star-net) first sealed the limit at 09:44:30 on 8/21 and finally re-sealed at 10:44:45 (it re-sealed after opening intraday), while the State Council readout was published at 19:40 — about 10 hours later, so causation is physically impossible. The composition of the day's 6 limit-ups is also mixed: 通鼎互联 (Tongding Interconnection) was a second consecutive limit-up driven by its own 8/20 interim report, while 东信和平 (Eastcompeace) and 楚天龙 (Chutian Dragon) belong to the smart-card / digital-currency line and have nothing to do with AI compute. The State Council meeting can only act on today and later, and its readout contains no numbers. This issue therefore downgrades "next-generation communications networks" to B and gives it no recommendation slot.

② Why the optical-module branch was cut from S to A (the most important self-correction in this issue). The first draft, following media framing, wrote "Q2 net margin rose from 29.4% to 35.7% on 1.6T mix upgrade." After checking the interim report and the call transcript, this must be corrected:

Gross margin moved only 0.36pp sequentially. If a sharp rise in the 1.6T share were really the driver, gross margin would have had to move — and it did not. Of the +6.12pp net-margin gain, only about 0.9pp is attributable to the core business (gross margin 0.36 + expense ratio 0.56); the rest comes from: fair-value gains (on the interim-report basis, H1 totalled RMB 537.12 million versus −RMB 0.8244 million in the prior-year period; the corresponding non-recurring line, "gains/losses from entrusting others with investment or asset management," is RMB 567 million) (+2.41pp), the effective tax rate falling 16.18%→13.18% (+1.31pp), the minority-interest share falling 9.22%→5.94% (+1.24pp), and R&D expense of RMB 508 million in Q2 versus RMB 645 million in Q1, −21.2% QoQ (+1.03pp). The year-on-year mix upgrade is real (the optical transceiver module segment's gross margin was 46.59%, +6.63pp YoY), but on a sequential basis the upgrade has already plateaued. One more item must be placed alongside it: Q2 standalone operating cash flow of −RMB 1.568 billion, with OCF/net profit collapsing from 1.01 for full-year 2025 to 0.13 in H1; trading financial assets rose from RMB 10 million at the start of the period to RMB 2.529 billion at period end, and the interim report does not disclose their composition.

③ 中际旭创 (Innolight) draws about 95% of revenue from overseas (H1 overseas revenue RMB 39.615 billion), so the State Council meeting and the Alibaba placement are sentiment beta for it, not an order channel. This correction matters: the two domestic weekend catalysts genuinely map to the domestic server/switch chain (紫光股份 (Unisplendour), 锐捷网络 (Ruijie Networks), 浪潮信息 (Inspur)) and to CCL — not to 中际旭创 (Innolight). CCL ranks first precisely because it supplies both sides.

On Item 7: A Popular Branch This Issue Rejects

The industry data is real (an HVLP4 shortfall of 1,500 tonnes in 2026), but the available A-share names cannot carry it:

• 诺德股份 (Nuode) (600110, SSE main board) hit the limit-up at +10.04% on Friday. In its 2026-06-18 "Share Trading Risk Warning Announcement" (Lin 2026-037) the company states verbatim: "the company's RTF-series copper foil, HVLP-series copper foil, carrier copper foil and other high-end products have only completed small-batch sample validation… mass production has not yet been achieved," and describes its main business as "electrolytic copper foil for lithium-ion batteries."
• 德福科技 (Defu Technology) (301511, ChiNext)'s interim ex-non-recurring +1,916.90% looks striking, but the segment data shows: electronic-circuit copper foil (including HVLP/RTF) revenue of RMB 1.308 billion is only 14.2% of the total, at a 4.85% gross margin — below lithium-battery copper foil's 9.47%; the full interim report mentions "NVIDIA" 0 times and "compute power" 0 times.

Conclusion: the industry trend is real, the beneficiary names are not.

Coverage Self-Check (executing "Fix 3" set by the 8/21 recap)

Friday's top 15 industries by main-force net inflow, accounted for one by one: non-ferrous metals (+RMB 8.829 billion) → branch 6; electronics (+RMB 8.553 billion) → branch 2; telecom equipment (+RMB 8.532 billion) → branch 5; telecom network equipment and devices (+RMB 7.852 billion), telecom (+RMB 7.813 billion) → branches 3/5; components (+RMB 4.490 billion), printed circuit boards (+RMB 4.118 billion) → branch 1; industrial metals (+RMB 3.450 billion), precious metals (+RMB 3.423 billion), copper (+RMB 2.782 billion), silver (+RMB 2.045 billion) → branch 6; power equipment (+RMB 2.039 billion) → led by 诺德股份 (Nuode), handled under branch 7.

Three that were not folded into a branch and must be accounted for:

---

3. Master Table of Single-Stock Catalyst Strength (descending by stock)

Limit-up/limit-down is judged by each board's own cap: SSE/SZSE main boards ±10%, ChiNext/STAR Market ±20%, BSE ±30%. The STAR Market and BSE additionally require an RMB 500,000 investor threshold. Unless otherwise noted, PE figures in this table are on the annualised basis (half-year profit × 2), for rough ranking only.

A cycle-position warning on the memory-module group. 江波龙 (Longsys) (301308, ChiNext) PE 7.73/PB 7.35, 佰维存储 (Biwin Storage) (688525, STAR Market) PE 9.72/PB 13.36, 德明利 (Demingli) (001309, SZSE main board) PE 7.04/PB 14.22 — extremely low PE stacked on extremely high PB is the classic pattern of a cyclical's earnings peak (PE is low because the denominator is peak-cycle profit). The Changcun prospectus discloses a 2026Q1 net margin as high as 70.9%, which precisely indicates NAND prices are at an extreme high. The Changcun IPO is positive for equipment and materials (capex), not for module makers (prices) — the two point in opposite directions.

The same pattern appears in 中际旭创 (Innolight): PE (TTM) 53.93 is at the 50.5th percentile since 2018, while PB 27.67 is at the 95.2nd percentile. This is one of the quantitative grounds for cutting the optical-module branch from S to A in this issue.

---

4. Single-Stock Scoring Model (100 points total)

Three grounds for raising 中际旭创 (Innolight)'s risk deduction from the draft's −10 to −14: ① Q2 standalone operating cash flow of −RMB 1.568 billion; ② PB at the 95.2nd percentile since 2018 and above NVIDIA's 26.61; ③ about 95% of revenue comes from overseas, so the weekend's two domestic catalysts (the State Council meeting, Alibaba) are mismatched with its revenue structure.

One ownership fact that must be clarified (corrected after a page-by-page review of the filing feed before finalising): the market has been circulating that "the controlling shareholder cashed out over RMB 5.3 billion in the past year." The selling did happen, but it is already over — the filing sequence is 2025-08-01 executive share-reduction pre-disclosure → 2025-09-26 controlling-shareholder reduction pre-disclosure → 2025-11-25 executive reduction plan expiry → 2026-01-16 controlling-shareholder reduction plan completed → 2026-07-29 chairman proposes a RMB 4–8 billion buyback. After 2026-01-16 there has been no new reduction-plan filing whatsoever. It therefore does not constitute current overhang, and this issue does not count it in the risk deduction.

---

5. Detailed Analysis of the Top 10

① 生益科技 (Shengyi Technology) (600183, SSE main board ±10%) — the strongest evidence in this issue

• Related news: 2026-08-23 20:00–21:00, 中际旭创 (Innolight)'s call with about 150 institutions (original record¹): "in H1 the supply of core materials such as optical chips, electrical chips and PCB remained fairly tight"; "some materials did see price increases, but the increases on products where our supply volume is large were not big"; "suppliers of PCB, optical chips and other materials are also actively expanding, and we expect supplier capacity to be released gradually from H2 this year into H1 next year."
• Positive logic: direct, but it must be said clearly that it is also on the receiving end of cost pressure. 生益科技 (Shengyi Technology) sits midstream in the chain "copper foil/electronic cloth → copper-clad laminate (CCL) → printed circuit board (PCB) → optical modules/AI servers": upstream price rises are a cost to it, while downstream shortage gives it pricing power. The net effect shows up in gross margin, and gross margin gives the answer — CCL and prepreg gross margin of 29.16%, +5.47pp YoY, i.e. price pass-through outran cost inflation.
• The company's own words (2026 interim report, "Management Discussion and Analysis," full PDF¹³): "faced with sustained sharp price increases in upstream raw materials, the sales team, on the objective basis of market demand exceeding supply and tight capacity along the industry chain, drove product pricing-strategy adjustments while balancing customer-relationship maintenance against rising cost pressure"; the same document also says that in the industry "leading manufacturers' high-end capacity is oversubscribed, with order schedules filled to year-end."
• The real pressure on the cost side (not hidden): from September 2025 to May 2026, the LME copper price rose from about USD 10,000/tonne to over USD 13,000/tonne; mainstream 7628-grade electronic cloth rose from RMB 4.15/metre to nearly RMB 7/metre. Meanwhile the average CCL price rose from RMB 108.16 per square metre to RMB 117.04 per square metre (about +8.2%) — clearly a smaller increase than raw materials. So the +5.47pp gross-margin gain comes more from "a marked rise in the high-end product share and continuous mix optimisation" (the company's own words) than from price pass-through alone. This must be made clear: if the high-end share stops rising, price increases alone may not keep covering costs.
• Why it deserves the top slot more than optical-module assemblers: 中际旭创 (Innolight) draws about 95% of revenue from overseas, while Alibaba's HKD 80 billion goes mainly to domestic servers and switches — these two capex streams belong to two different customer systems, and CCL supplies both. It is the only link in this cycle feeding off both domestic and overseas capex.
• Stage of the industry branch: mid-fermentation. On Friday the PCB concept saw main-force +RMB 6.248 billion and components +RMB 4.490 billion, but the PCB group had zero limit-ups on Friday — sentiment has not yet peaked.
• Fundamentals verification (all four items taken from the original interim report, not media paraphrase): H1 revenue RMB 19.0256 billion (+50.05%), total profit RMB 4.2212 billion (+129.33%), net profit attributable to parent RMB 3.2870 billion (+130.42%), ex-non-recurring RMB 2.9077 billion (+110.99%); weighted ROE 18.17% (+8.82pp), ex-non-recurring ROE 16.08% (+7.04pp).
• ⚠️ This is the most differentiating piece of evidence in the issue — its cash flow is the only one improving in the whole report: net operating cash flow of RMB 2.9832 billion, +53.44% YoY, equal to 91% of net profit attributable to parent. Against the other stocks covered here: 中际旭创 (Innolight) OCF/net profit is just 0.13 (Q2 standalone −RMB 1.568 billion), 星网锐捷 (Star-net) −RMB 3.567 billion, 华海清科 (Hwatsing) −81.44%, 拓荆科技 (Piotech) −92.63%, 德福科技 (Defu) −RMB 613 million, 诺德股份 (Nuode) −RMB 1.094 billion. On a chain where nearly every company is trading cash for growth, it is one of the few actually collecting its profits.
• Segments: CCL and prepreg revenue RMB 12.357 billion (+47.74%), gross margin 29.16%, +5.47pp YoY; printed circuit boards RMB 5.519 billion. M9 ultra-high-grade CCL has passed NVIDIA Rubin and GB200 full-platform certification and is shipping in volume (source¹⁴, media basis, not verified word for word in the interim report, so discount accordingly when citing). CCL is about 65% of revenue, an extremely high match to the main business.
• Industry position: the world's second-largest CCL maker, a core holding.
• Technical sentiment: Friday +5.16%, closing at RMB 132.80, main-force +RMB 1.011 billion (5th in the whole market), turnover only 2.76%, free-float market cap RMB 320.3 billion. No limit-up, no locked-limit board, no one-day-wonder signature.
• Final judgement: priority deep-dive. All three conditions met, and it is the only name whose overnight incremental driver comes from a third party rather than from self-description. Its weakness is that its own interim was already disclosed on 8/15, so there is no new earnings catalyst this week.

② 沪电股份 (Wus Printed Circuit) (002463, SZSE main board ±10%)

• The same material-shortage chain. H1 pre-announced net profit attributable to parent RMB 2.83–3.0 billion (+68.17%~78.28%), ex-non-recurring RMB 2.73–2.88 billion (+66.08%~75.20%); Q1 revenue RMB 4.038 billion (+56.25%); the Thai subsidiary turned profitable in 2026Q2 standalone, indicating the capacity ramp is complete, a positive contribution to H2 gross margin.
• Expectation gap is on the low side: the pre-announcement was already released on 7/13 and is old information; the only new element is downstream corroboration.
• Friday +6.03%, closing at RMB 121.21, main-force +RMB 785 million, turnover 2.6%.
• Risk: the 8/26 (Wednesday) interim report falls on the same day as NVIDIA's results.
• Conclusion: watch closely.

③ 北方华创 (Naura) (002371, SZSE main board ±10%) — the largest expectation gap in this issue

• Related news: 长存控股 (Changcun Holdings)'s IPO filing was accepted after hours at 23:10 on 8/21, raising RMB 33 billion, of which RMB 20.8 billion goes to the "Yangtze Memory volume-production line technology upgrade project" and RMB 12.2 billion to R&D. The prospectus discloses for the first time: 2026Q1 revenue RMB 47.042 billion, net profit attributable to parent RMB 33.379 billion; full-year 2025 RMB 63.185 billion / RMB 14.211 billion; 2024 RMB 45.203 billion / RMB 6.771 billion. Major shareholders are 湖北长晟 (Hubei Changsheng) 26.54%, 芯飞科技 (Xinfei Technology) 25.35%, the National IC Fund Phase I 11.97% and Phase II 11.38%, with no actual controller and no controlling shareholder.
• Positive logic: direct — the RMB 20.8 billion allocated to the volume-production line technology upgrade is in essence an equipment procurement budget. Note: the claim that 北方华创 (Naura) is Changcun's main equipment supplier comes from self-media compilations; company filings contain no corresponding order amount, and this issue cites no specific order figure.
• Expectation gap: the highest in this issue. The news only became public at 23:10, while on Friday the memory-chip concept rose just +0.54% with main-force net inflow of RMB 39 million, advanced packaging −RMB 517 million and photoresist −RMB 422 million, and the stock itself was up only +0.26% on 0.8% turnover — completely unpriced.
• The counter-evidence that must be placed alongside it (this is why it can only be "watch closely"): every equipment name that has already reported interims is decelerating — 华海清科 (Hwatsing) revenue +35.58% but net profit attributable to parent only +11.44%, operating cash flow −81.44%; 盛美上海 (ACM Research Shanghai) revenue only +13.87%; 拓荆科技 (Piotech) operating cash flow −92.63%. "Changcun expansion" is a 2027 story, while equipment names' current filings are decelerating. 北方华创 (Naura)'s own interim is not out until 8/26, so it is currently a blind box.
• Friday +0.26%, closing at RMB 714.43, turnover 0.8%, free float RMB 518.0 billion. Clean position, no chase-high pressure.
• Conclusion: watch closely. What you are buying is the customer capex expectation, not its own current filings — this distinction is essential, otherwise the 8/26 interim could falsify it outright.

④ 中际旭创 (Innolight) (300308, ChiNext ±20%) — both sides must be on the table

Bull side (all verified hard data):

• H1 revenue RMB 41.778 billion (+182.49%), net profit attributable to parent RMB 13.651 billion (+241.70%), ex-non-recurring RMB 13.092 billion (+229.32%), weighted ROE 37.62%.
• Q2 gross margin of 46.42% is the company's highest quarterly gross margin ever; the optical transceiver module segment's gross margin was 46.59%, +6.63pp YoY — the year-on-year mix upgrade is real.
• Management's own words: "major customers have already given 2027 guidance and orders," "in the past the optical-module industry typically signed 3-month orders; now many customers have given us 2027 orders" — order visibility extending from 3 months to more than a year is the most substantive change in this cycle.
• Non-recurring items of RMB 559 million are only 4.1% of net profit attributable to parent, and ex-non-recurring profit is still +229.32%, so the claim that "profit was propped up by investment gains" does not hold.
• Construction in progress rose from RMB 1.422 billion at the start of the year to RMB 3.707 billion (+160.7%), and fixed assets from RMB 7.081 billion to RMB 9.823 billion (+38.7%); the July H-share issue raised USD 7 billion.
• PE (TTM) fell overnight from about 73.8 to 53.93 on the results, at the 1.6th percentile of the past year; at RMB 943 the share price is still 14.1% below the MA60 (1,098.23) and 31.8% below the 6/22 year-to-date closing high of 1,382.33.
• The chairman proposed a RMB 4–8 billion buyback on 7/28.

Bear side (equally verified hard data):

• Q2 gross margin rose only +0.36pp QoQ (46.06%→46.42%). If "a rising 1.6T share" were the main driver, gross margin would have had to move — and it did not; of the +6.12pp QoQ net-margin gain, only about 0.9pp is attributable to the core business.
• Q2 standalone operating cash flow of −RMB 1.568 billion (Q1 was +RMB 3.368 billion); OCF/net profit collapsed from 1.01 for full-year 2025 to 0.13 in H1. In Q2, receivables rose RMB 5.465 billion and inventory RMB 4.154 billion, while payables rose only RMB 1.679 billion.
• Trading financial assets rose from RMB 10.02 million at the start of the period to RMB 2.5291 billion at period end, with H1 fair-value gains of RMB 537.12 million (prior-year period −RMB 0.8244 million), and the interim report does not disclose their composition. That gain is 3.9% of H1 net profit attributable to parent and is the largest single non-recurring item.
• R&D expense of RMB 508 million in Q2 < RMB 645 million in Q1 (−21.2%); this item alone contributes +1.03pp to net margin.
• PB 27.67 is at the 95.2nd percentile since 2018 and above NVIDIA's 26.61; yet NVIDIA's gross margin is 74.15% versus 中际旭创 (Innolight)'s 46.42%.
• Contract liabilities are only RMB 20.70 million, below RMB 26.33 million at the start of the year; the interim report contains no quantified backlog disclosure.
• Upstream 天孚通信 (TFC Optical Communication) (300394, ChiNext) posted H1 revenue growth of only +15.1% YoY; competitor 新易盛 (Eoptolink)'s implied Q2 net profit attributable to parent is +51.8%~+87.8% QoQ, faster than 中际旭创 (Innolight)'s +38.1%.
• Revenue is almost entirely overseas — on the interim report's own wording: for optical transceiver modules the "main revenue source region: overseas," with sales volume of 16.85 million units and sales revenue of RMB 39.599 billion, i.e. 94.8% of H1 total revenue of RMB 41.778 billion; the company states its "products are mainly aimed at North America, Europe and other countries or regions." So the weekend's State Council meeting (domestic communications networks) and the Alibaba placement (domestic AI infrastructure) are sentiment beta for it, not an order channel.
• Ownership: the controlling shareholder did sell (pre-disclosed 2025-09-26), but that plan was announced as completed on 2026-01-16, with no new reduction plan since; on 2026-07-29 the chairman instead proposed a RMB 4–8 billion buyback. This issue therefore does not list share reduction as a current overhang — see the clarification below §4.

Core disagreement: is that 6.12pp jump in Q2 net margin an operating trend — "1.6T mix upgrade + scale effects" — that can carry into H2, or a patchwork of "fair value 2.41pp + tax rate 1.31pp + minority interest 1.24pp + R&D 1.03pp"? The +0.36pp QoQ gross margin is a fact both sides accept.

Technical sentiment and final judgement: Friday +4.29%, closing at RMB 943.00, main-force +RMB 3.595 billion (first in the whole market, about 1.4× the second), turnover 2.59%, free-float market cap RMB 1.047 trillion. Conclusion: watch closely, with strict conditions attached — only consider participating if the auction gaps up by no more than +3%; abandon it for the day above +5%. The basis: since 2020 the stock has gapped up ≥+5% on 18 occasions, with a median open-to-close of −1.73%, a 61% rate of closing red-to-black, and an average intraday low of −4.40% relative to the open.

⑤ 深南电路 (Shennan Circuits) (002916, SZSE main board ±10%) and ⑥ 东山精密 (Dongshan Precision) (002384, SZSE main board ±10%) — two catch-up slots in branch 1

• Both are catch-up slots within branch 1. On Friday 深南电路 (Shennan Circuits) rose +3.84% but turnover was only 1.21% (the coldest in the PCB group); 东山精密 (Dongshan Precision) rose only +1.73% while main-force inflow was +RMB 703 million (11th in the whole market) — money came in but the price did not move.
• Shared weakness: no company-level news of any kind within the window; what you are buying is purely branch diffusion. If branch 1 fails, these two are the first to be dropped.
• Conclusion: watch closely.

⑦ 中微公司 (AMEC) (688012, STAR Market ±20%)

• The other beneficiary of Changcun's RMB 20.8 billion equipment allocation; Friday −0.82%, clean position.
• The same two-sidedness as 北方华创 (Naura): the catalyst is fresh and unpriced, but peers' current filings are decelerating.
• Note that the STAR Market's ±20% cap means volatility is twice that of the main board, and there is an RMB 500,000 / 2-year threshold, so not everyone can buy it.
• Conclusion: watch closely.

⑧ 紫光股份 (Unisplendour) (000938, SZSE main board ±10%)

• One of the best-matched names to Alibaba's HKD 80 billion of domestic AI capex: it holds 87.98% of H3C, has a 38.2% share of the enterprise-network switch market, and its 51.2T CPO silicon-photonics switch is already shipping in volume (media basis).
• But two things must be kept apart: the 8/24 filing that "the placement application was accepted by the SZSE" is not new information — the placement plan's second amendment was disclosed on 8/18 and the application materials were filed on 8/21, so acceptance is only a procedural node; and the placement itself dilutes equity.
• Friday +2.19%, closing at RMB 36.44, turnover 5.33%.
• Conclusion: watch closely.

⑨ 士兰微 (Silan Micro) (600460, SSE main board ±10%)

• Friday limit-up, main-force +RMB 1.148 billion (3rd in the whole market), turnover value RMB 2.70 billion, turnover rate 4.63%. Drivers are the start-up of the 8-inch SiC project and the automotive-grade IGBT/SiC ramp.
• Stated honestly: this is an independent line unrelated to this issue's main branches, and there is no new news within the window (8/21 after hours to now). The only reason it is in the table is the size of the money flow, which "Fix 3" requires be accounted for — not because a new catalyst for today was found.
• Conclusion: watch closely.

⑩ 星网锐捷 (Star-net) (002396, SZSE main board ±10%) — the biggest downgrade in this issue

The first draft listed it as "watch closely" on the market's attribution; after verification it must be cut to watch only, on four grounds:

1. The State Council meeting has no causal link to its 8/21 limit-up. It first sealed at 09:44:30 and finally sealed at 10:44:45, while the State Council readout came out at 19:40, about 10 hours later.
2. The interim report is not a first disclosure. A pre-announcement was already issued on 7/3 (net profit attributable to parent RMB 310–430 million, +46.27%~102.9%), and the pre-announcement text already spelled out all three items: "sharp growth in the data-centre switch business," "an increase in the fair value of the industry fund," and "the 德明通讯 (Deming Communication) equity transfer." The interim's actual figures land inside the pre-announced range, and non-recurring items of RMB 98.5 million are even slightly below the pre-announced floor of RMB 100 million. Over the five trading days from 8/14 to 8/21 the share price rose only +1.53%.
3. The basis problem with "net profit +75%." H1 net profit attributable to parent was RMB 372.71 million (+75.86%), but ex-non-recurring was only RMB 274.16 million (+45.91%); cross-checking on a pure operating basis of "gross profit − period expenses" gives +46.3%, which agrees. The largest single item in the difference is the RMB 81.45 million fair-value change on the industry fund. As for the widely circulated "RMB 477 million recovered from disposing of non-core assets" — RMB 477 million is the consideration for 65% of 德明通讯 (Deming Communication), and only RMB 20.6725 million of it entered H1 profit, and that transaction was already closed and announced on 4/25. The tax break was no help either: the effective tax rate rose from 7.48% to 10.13%, a drag on profit.
4. Negative new information disclosed by the financials for the first time: operating cash flow −RMB 3.5665 billion (−645.60%), gearing up from 42.37% to 49.98%, short-term borrowings from RMB 1.68 billion to RMB 4.88 billion, and cash from RMB 4.22 billion to RMB 1.80 billion. None of these four items appeared in the 7/3 pre-announcement; they were disclosed for the first time in the 8/15 interim, while the market's reaction focused on the positives.

Two structural facts to add: 83.6% of its attributable profit comes from 锐捷网络 (Ruijie Networks) (301165, ChiNext — not the STAR Market), in which it holds 44.88%; and on the 8/21 Dragon-Tiger List, the four "institutional dedicated" seats were combined net sellers of RMB 33.7 million, while two hot-money seats (Guotai Haitong Beijing Zhichunlu, Kaiyuan Securities Xi'an Xidajie) were combined net buyers of RMB 292.6 million, 72.7% of total net buying. Shenzhen-Connect net buying of RMB 112.74 million is confirmed.

Conclusion: watch only. The branch fits and its free float of just RMB 25.1 billion gives good elasticity, but the attribution chain for the limit-up fails both the timing test and the accounting-basis test.

---

6. Pass List

---

7. Intra-Branch Ranking

Branch 1: Compute Materials (CCL/PCB)

• Hardest name: 生益科技 (Shengyi Technology) — the only one with double verification of "an overnight incremental driver from a third party + its own gross margin already up 5.47pp."
• Who is just following: 深南电路 (Shennan Circuits), 东山精密 (Dongshan Precision) (no company-level news). Who is a Pass: 胜宏科技 (Victory Giant).

Branch 2: Semiconductor Equipment (Changcun IPO)

• The core contradiction in this branch: the catalyst is in 2027, the financials are decelerating. Of the 4 that have disclosed interims, 3 are decelerating or have deteriorating cash flow. Buying this line means buying expectations rather than current earnings, and position size must be set as for an "expectation trade."

Branch 3: AI Optical Interconnect (Optical Modules)

• Key structural read: money is concentrating into the leaders, not diffusing broadly. On Friday both leaders rose while all three second-tier names fell — this is "pre-climax divergence," not "ignition." The odds on chasing the second tier are clearly worse than on the leaders.
• 天孚通信 (TFC Optical Communication)'s H1 revenue growth of only +15.1% is an important industry cross-check: if 1.6T were really ramping explosively, upstream optical devices could not be growing at only 15%.

Branch 4: Domestic AI Infrastructure (Alibaba's HKD 80 billion)

• Note one relationship that is easy to get backwards: 锐捷网络 (Ruijie Networks) (301165) has higher earnings quality than its parent 星网锐捷 (Star-net) (002396) — the parent holds only 44.88% of it, and 83.6% of the parent's attributable profit comes from it. Buying the parent means buying a holding platform at roughly a 60% discount, while the platform's own non-Ruijie business saw H1 revenue fall −15.2% YoY. The discount could close either through the parent rising or through the subsidiary falling.

---

8. Today's Opening Verification Signals

Auction Signals (09:15–09:25)

1. 生益科技 (Shengyi Technology) (600183): a gap-up of +2%~+5% is healthiest. If it gaps up more than +7%, treat the Sunday-evening call as fully cashed in at once and downgrade it to watch only for the day.
2. 北方华创 (Naura) (002371): this is the largest expectation gap in the issue. A gap-up of +2% or more on volume → the Changcun capex logic is accepted; a flat or lower open → the market judges that "a 2027 story + an 8/26 blind-box interim" is not worth front-running, and the branch is void for the day.
3. 中际旭创 (Innolight) (300308): only consider it on a gap-up within +3%; abandon it for the day above +5% (historical statistics: across the 18 gap-ups of ≥+5%, median open-to-close was −1.73% and 61% closed red-to-black). A flat or lower open would mean the market treats Q2 standalone cash flow of −RMB 1.568 billion as the primary contradiction, and the entire branch should be downgraded immediately.
4. 新易盛 (Eoptolink) (300502): a large gap-up (>+6%) is a risk signal instead — the interim is tomorrow, and money front-running today has to face realisation pressure tomorrow.

Sector Signals (09:30–10:30)

• Confirmation signal: 3 or more fast limit-ups in the PCB / CCL direction, with 生益科技 (Shengyi Technology) rising on volume without breaking its limit → branch 1 is confirmed.
• Core-holding signal: whether 中际旭创 (Innolight)'s main-force net inflow ranks in the whole market's top three for a second consecutive day (it was first on 8/21). This is more reliable than the price change.
• Ladder signal: whether 2 or more of Friday's 6 telecom-equipment limit-ups produce a second board. If only 通鼎互联 (Tongding Interconnection) (already 2 boards, turnover 30.46%) carries the baton alone, branch 5 is ebbing for the day.
• Diffusion vs contraction: whether second-tier optical modules (博创科技 (Broadex), 仕佳光子 (Shijia Photons), 德科立 (Taclink)) turn from down to up. If they keep falling while the leaders rise, money is still contracting — do not touch the second tier.
• Whether the memory chain starts to price it: whether 北方华创 (Naura) and 中微公司 (AMEC) see main-force net inflow turn from near zero on Friday to +RMB 300 million or more — that is the minimum threshold for branch 2.

Risk Signals

• Gap-up then dive: if 中际旭创 (Innolight) gaps up more than +5% and turns negative within half an hour, avoid the entire AI compute chain for the day — when an RMB 1.05 trillion core holding weakens, no single stock is spared.
• Lone limit-up: if telecom equipment is left with only 通鼎互联 (Tongding Interconnection) at the limit and nothing new, that is ebbing, not fermenting.
• Core holding not following: if 紫金矿业 (Zijin Mining) (turnover only 1.46% Friday) falls on volume, money is starting to leave the precious-metals branch and the risk in 湖南白银 (Hunan Silver) (turnover already 18.29%) is magnified.
• Yesterday's strong theme ebbing: the silver sector rose +8.68% on Friday; if it opens lower today and 湖南白银 (Hunan Silver) breaks its limit, treat that as confirmation of the precious-metals climax.
• This week's overhang: 8/25 新易盛 (Eoptolink) interim → 8/26 北方华创 (Naura) and 沪电股份 (Wus Printed Circuit) interims + NVIDIA results (after the US close) → 8/27 the A-share reaction day to NVIDIA. Today's and tomorrow's positions should not be built on the assumption that they can be held safely into Friday.

---

9. Final Recommendations

① The 5 Stocks Most Worth Watching Today

Screening rule (carrying over "Fix 2" set by the 8/21 recap): recommendation slots are reserved for names that have all three of an overnight incremental driver + already-delivered financials + verifiable business purity; anything characterised as a "sentiment trade" is excluded from the Top 5.

Why 新易盛 (Eoptolink) (300502) is not in a recommendation slot: the fundamentals fully qualify (incentive target of 2026 revenue ≥ RMB 50 billion versus RMB 24.842 billion in 2025), but two conditions fail — ① the incentive draft was already public on the evening of 8/20 and Friday's +6.76% priced it, so it is not an "overnight incremental driver"; ② the interim is tomorrow (8/25), so buying on Monday means holding a T+1 earnings event, and the pre-announced range of RMB 7–8 billion spans 14%.

Why 星网锐捷 (Star-net) (002396) was cut from a draft recommendation candidate to watch only: ① it sealed the limit at 09:44:30 on 8/21, 10 hours before the State Council meeting, so the attribution chain fails; ② the 7/3 pre-announcement already disclosed every driver and the interim adds nothing; ③ ex-non-recurring growth is only +45.91%, not +75.86%, and the "RMB 477 million asset disposal" actually contributed only RMB 20.6725 million of profit; ④ operating cash flow of −RMB 3.567 billion.

② The 3 Strongest Branches Today

③ Directions Not Worth Chasing Today

1. High-end HVLP copper foil (诺德股份 (Nuode), 德福科技 (Defu Technology)) — the industry trend is real, the beneficiary names are not. 诺德股份 (Nuode) itself wrote in its 6/18 filing that HVLP "has not yet achieved mass production," it is actually loss-making on a TTM basis, two CSRC cases are unresolved and the 2025 annual report carried a qualified opinion; 德福科技 (Defu Technology)'s AI-related revenue is only 14.2% of the total at a 4.85% gross margin, below its own lithium-battery foil business.
2. Yangtze Memory shadow plays (养元饮品 (Yangyuan Beverage), 市北高新 (Shibei Hi-Tech), 大众交通 (Dazhong Transportation), 东方明珠 (Oriental Pearl), 国脉文化 (Guomai Culture)) — 养元饮品 (Yangyuan Beverage)'s stake is carried under FVOCI, so neither fair-value changes nor disposal gains will ever enter the income statement, and the chain "Changcun lists → Yangyuan's EPS explodes" fails on accounting grounds alone; moreover its 8/22 interim, the day after acceptance, still carried the stake at cost of RMB 1.6 billion with no revaluation, and its core business posted 2026Q2 standalone net profit attributable to parent of −54.5%. The other four are multi-layer nested remote mappings, and three of them are loss-making. Better to buy equipment names than shadow plays.
3. Chasing precious metals (湖南白银 (Hunan Silver), 盛达资源 (Shengda Resources), 白银有色 (Baiyin Nonferrous)) — the gold rally is real, but on Friday the A-share silver sector rose +8.68% in a single day and 湖南白银 (Hunan Silver)'s turnover hit 18.29%, already climax territory. The gold price's +1.86% happened in Friday's New York session, while A-shares had already priced in a larger move on Friday.
4. Pure-sentiment telecom-equipment tickets (通鼎互联 (Tongding Interconnection), 楚天龙 (Chutian Dragon), 瑞斯康达 (Raisecom), 东信和平 (Eastcompeace)) — the State Council meeting has no causal link to Friday's limit-up wave (a 10-hour timing gap), and the readout contains no investment size or timetable at all. 通鼎互联 (Tongding Interconnection)'s turnover was 30.46%, 楚天龙 (Chutian Dragon)'s PE is negative, 瑞斯康达 (Raisecom)'s turnover value was only RMB 260 million, and 东信和平 (Eastcompeace) belongs to the smart-card line, which does not fit this cycle's theme.
5. Second-tier optical modules (博创科技 (Broadex), 仕佳光子 (Shijia Photons), 德科立 (Taclink), 太辰光 (T&S Communications)) — on Friday they all fell on a day the branch rallied broadly (−3.23% / −3.66% / −4.57%), and 太辰光 (T&S Communications)'s turnover already reached 18.32%. Money is contracting toward the leaders.

④ Final One-Sentence Judgement

Today's main line is not a "new theme" but the three pieces of primary material that landed over the weekend; the single most information-dense sentence among them is what 中际旭创 (Innolight) management said out loud on Sunday evening — "supply of core materials such as PCB remained fairly tight" — and it redirects the reasonable destination of money away from an optical-module leader with 95% of revenue overseas, a PB already above NVIDIA's and negative Q2 standalone operating cash flow, and toward upstream materials whose valuation and chart position still have room and which supply both domestic and overseas compute capex.

---

Self-discipline record for this issue (for recap comparison):

1. Three judgements from the first draft were overturned before finalising: ① the optical-module branch cut from S to A (Q2 gross margin only +0.36pp QoQ, standalone OCF −RMB 1.568 billion); ② 星网锐捷 (Star-net) cut from recommendation candidate to watch only (it sealed the limit 10 hours before the State Council meeting); ③ a basis disclosure added across the whole PE table (the endpoint's f9 is annualised PE, not TTM; 诺德股份 (Nuode)'s TTM is actually −151.31). Item 3 is the closest this run came to publishing a wrong conclusion.
2. Discarded 1 high-risk source (the Jiemian News "Cailianshe Morning News Digest for August 23," actually a 2018 article from a recurring column of the same name).
3. Issued a Pass against the tape on 3 market-favourite stocks (诺德股份 (Nuode), 德福科技 (Defu Technology), 养元饮品 (Yangyuan Beverage)), with every ground taken from the companies' own filings, segment data and accounting measurement attributes rather than third-party commentary.
4. Deliberately downgraded 1 of the hottest branches (precious metals), on the grounds that A-shares had already priced it ahead.
5. Explicitly acknowledged 1 coverage gap (consumer electronics / 歌尔股份 (GoerTek), +RMB 1.988 billion of inflow but no reliable 2026 news found within the window).
6. Explicitly flagged 1 piece of unfinished verification (拓荆科技 (Piotech)'s profit composition), and therefore drew no conclusion about its profit quality.
7. Traced 4 "expectation gaps" back to their first disclosure date (新易盛 (Eoptolink)'s incentive plan on the evening of 8/20, 沪电股份 (Wus)'s pre-announcement 7/13, 紫光股份 (Unisplendour)'s placement plan 8/18, 星网锐捷 (Star-net)'s pre-announcement 7/3), and downgraded their incremental character accordingly.
8. Corrected 1 factual error: NVIDIA reports after the US close on 8/26, so the A-share reaction day is 8/27 (Thursday), not Wednesday.
9. One more erroneous risk item was self-detected and withdrawn before finalising: the media-sourced claim that "中际旭创 (Innolight)'s controlling shareholder cashed out over RMB 5.3 billion between 2025-05-28 and 2026-07-30" was confirmed, after a page-by-page review of the filing feed, to be wrong — the reduction plan was announced as completed on 2026-01-16, with no new plan since, and on 7/29 the chairman instead proposed a buyback. The original wording turned a closed matter into a current overhang, i.e. the direction was reversed; it has been corrected in all three places in the text and removed from the risk deduction.
10. Added 1 piece of primary data absent from media paraphrase that determines the quality of the top recommendation slot: from the original text of 生益科技 (Shengyi Technology)'s interim report, net operating cash flow of RMB 2.983 billion, +53.44% YoY, covering 91% of net profit attributable to parent — the only cash-flow improvement among the twenty-plus stocks here. Its cost-side exposure was added at the same time (copper up over 30% across the range, electronic cloth up about 69%, while the average CCL price rose only about 8.2%), making clear that the +5.47pp gross-margin gain comes more from high-end product mix than from price increases alone — avoiding writing a midstream link up one-sidedly as a pure price-rise beneficiary.
11. 诺德股份 (Nuode)'s core evidence was checked word for word back against the primary filing: announcement number Lin 2026-037, dated 18 June 2026, with the original text "have only completed small-batch sample validation… mass production has not yet been achieved" and "the main business is electrolytic copper foil for lithium-ion batteries," and stating 2025 net profit attributable to parent of −RMB 298.51 million. The citation is accurate.

⚠️ Risk disclaimer: this list is pre-market information compilation 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 industry-chain mapping errors; it must not be used directly as a basis for trading.