A-Share Pre-Market Brief | 2026-09-15 Tuesday

News window: 2026-09-14 (Mon) 15:00 close → 2026-09-15 (Tue) 07:00. Price baseline: 2026-09-14 close (timestamps verified line by line). The overnight U.S. session is the 2026-09-14 U.S. Eastern session (Beijing time 9/14 21:30–9/15 04:00), the entire span occurring after the A-share close.

---

0. One-Sentence Summary for Today

1. The strongest catalyst is a "narrative event," not an order event: Anthropic CEO Dario Amodei published an approximately 3,900-word essay "We Must Pace the Frontier" on 9/12 (Sat), arguing that "we must slow the pace at which AI model capability improves"; OpenAI's Sam Altman, xAI's Elon Musk, and Hugging Face's Clem Delangue endorsed it in turn. The overnight U.S. session accordingly completed an exceptionally clean rotation from "AI hardware → cybersecurity/software."
2. The overnight U.S. divergence is order-of-magnitude in scale: the Philadelphia Semiconductor Index −5.86%; while on the same night Zscaler +16.52%, SentinelOne +14.48%, CrowdStrike +13.85%, Palo Alto +13.09%, software ETF (IGV) +5.04% versus semiconductor ETF (SMH) −4.75%, a single-day spread of about 9.8 percentage points.
   • ⚠️ The optical communications piece cannot be attributed wholesale to Amodei — it must be split into two drivers: the direct cause of Corning (GLW) −13.70% is its own $2 billion ATM equity offering disclosed on 9/14 (Goldman Sachs as sole agent, 1.0% commission), making it the worst S&P 500 single stock that day — this is an equity dilution event; Coherent −12.73%, Lumentum −9.92%, Fabrinet −7.73%, Ciena −8.55% were sympathy declines. Corning simultaneously maintained its Q3 core sales guidance, so its operating picture did not change.
   • What this means for A-shares: a U.S. company diluting via an offering carries no demand-side read-through whatsoever for Chinese optical module makers. Therefore the inference "overnight optical communications crash → A-share optical modules must sell off a second time" does not hold, and this report has downgraded the strength of that conclusion accordingly (see §9③).
   • Two qualifiers: (a) I have reverse-computed and verified the price changes of 52 tickers one by one ((last−prev close)/prev close, all taken from the 9/14 closing reads), with no conflict against any interface field; there is roughly a 1pp difference versus some media conventions, and the reverse-computed close is authoritative throughout. (b) This bid is "narrow" — pure cybersecurity single stocks rose 12%~16%, but cybersecurity ETFs CIBR only +5.99% and HACK +7.88%, with the S&P 500 −0.48%. The buying went only into a handful of pure-play names; it did not broaden into a systematic upward revision of overall security spending.
   • The hardware side is not homogeneous either: Nvidia was only −3.36%, while optical modules fell −9%−13% and semiconductor equipment −6%−8%. Compute chips ≠ optical modules ≠ PCB; the A-share mapping must be layered and cannot be papered over with the single phrase "AI hardware chain."
   • Two qualifiers that must both be stated out loud: (a) I have reverse-computed and verified the price changes of 52 tickers one by one ((last−prev close)/prev close), with no conflict against any interface field; (b) this bid is "narrow" — pure cybersecurity single stocks rose 12%~16%, but cybersecurity ETFs CIBR only +5.99% and HACK +7.88%, while the S&P 500 was −0.48%. The buying went only into a handful of pure-play names; it did not broaden into a systematic upward revision of overall security spending. In other words, offshore money was also buying "narrative," not "orders already visible." This point directly determines how large a position this A-share line deserves.
3. Monday's A-share pricing was "layered," and this is where today's largest gap sits (precision is required here; it cannot be waved through in one sentence):
   • Already priced: optical modules and STAR-market growth names already fell on Monday — STAR 50 −1.62%, 中际旭创 (Zhongji Innolight, 300308) −5.72%, 新易盛 (Eoptolink, 300502) −4.99%, and all gapped down at the open.
   • The other side of "already priced": cybersecurity already rose on Monday — 中新赛克 (Sinovatio) 3 consecutive limit-ups, 天融信 (Topsec)/永信至诚 (Integrity Technology)/博汇科技 (Bohui Technology) limit-up.
   • The PCB leaders have already fallen too: 沪电股份 (WUS Printed Circuit) −1.99%, 深南电路 (Shennan Circuits) −1.90%, 生益科技 (Shengyi Technology) −1.64%, 胜宏科技 (Victory Giant) −1.74%, all four gapped down at the open (open-gap range −4.65%~−2.14%).
   • And the four PCB leaders were "gapped down and absorbed": 沪电 opened −4.65% → closed −1.99% (+2.80% from the open), 深南 −2.14%→−1.90% (+0.24%), 生益 −2.61%→−1.64% (+1.00%), 胜宏 −2.62%→−1.74% (+0.90%). All four closed above their opening prices — this is direct evidence that "the overnight bad news has already been absorbed intraday," the opposite direction from "a second sell-off is still coming today," and it must be stated alongside.
   • What genuinely has not been priced is only that batch of small-cap speculative PCB names: 科翔股份 (Kexiang, 300903) +20.00% (gapped down −1.39%, intraday +21.69%), 满坤科技 (Mankun Technology, 301132) +20.00% (gapped down −2.22%), 中京电子 (Zhongjing Electronics, 002579) +10.01% (gapped down −5.44%, intraday +16.35%), 澳弘电子 (Aohong Electronic, 605058) +10.00% (gapped down −1.89%). Note: 超声电子 (Guangdong Ultrasonic Electronics, 000823) gapped UP +1.12%, not down, a different pattern from the other four in the group. Together they pushed the components sector to +2.84% with RMB 2.754 billion of net main-force inflow.
   • So the accurate statement is: it is not that "A-shares have not priced AI hardware," but rather that "the institutionally priced portion has already fallen and even begun to be absorbed, while the hot-money priced portion is still rising against the tide" — today's risk exposure lands precisely on the latter.
4. Likely direction of flows: cybersecurity/AI security (the National Cybersecurity Awareness Week 9/14–20 is under way, and the same theme surged offshore overnight — resonance on both sides) ＞ innovative drugs/CRO (already started Monday, and carries "non-AI asset" defensive properties) ＞ cash. The AI hardware chain (optical modules, PCB, compute hardware, lab-grown diamond thermal management) carries the greatest absorption risk today.
5. Monday's market state requires clearing up a surface contradiction first: the indices fell broadly (Shanghai Composite −0.07%, Shenzhen Component −0.64%, ChiNext Index −1.10%, STAR 50 −1.62%), but advance/decline was 2,939 up / 2,083 down / 185 flat (Legu snapshot, statistical timestamp 2026-09-14 15:00:00). The two are not contradictory — CSI 1000 +0.55% and CSI 500 −0.12% versus CSI 300 −0.67% gives the answer: Monday was a textbook rotation day of "small caps up broadly, weights and growth sold off." Combined turnover across both exchanges was RMB 1.63 trillion, about RMB 342.7 billion lighter than the prior session. The official limit-up pool had 55 names (verified tc=55, len(pool)=55, no truncation).
   • Industry breakdown of the 55-name limit-up pool (one fact unfavorable to this report must be self-disclosed): components 6｜power 3 + grid equipment 3 = 6｜auto chain (auto parts 3 + passenger vehicles 1 + commercial vehicles 1) = 5｜pharma chain (medical services 2 + TCM 1 + chemical pharma 1 + biologics 1) = 5｜cybersecurity definition (software development 2 + IT services 2 + computer equipment 1) = 5｜general equipment 3｜chemical products 3｜general retail 2｜1 each for the rest.
   • The conclusion has to be honest: cybersecurity was NOT the group with the broadest limit-up count on Monday — components and power/grid equipment were each no fewer. So my listing cybersecurity as Branch 1 rests on "the policy event is still running + overnight offshore resonance in the same theme," not on "the broadest breadth on Monday." Two other groups are not developed in this report: grid equipment with 3 names (including 通达股份 (Tongda Cable) at 2 consecutive limit-ups) and glass/fiberglass (凯盛新能 (Kaisheng New Energy) at 3 consecutive limit-ups, which belongs to photovoltaic glass and is NOT the same chain as the electronic-grade fiberglass cloth/CCL price-hike chain discussed in §7 — do not conflate them). This issue could not verify their drivers and leaves the gap honestly blank.
6. Pre-market assessment: bearish at the index level, extremely differentiated at the structural level. The ChiNext Index/STAR 50 are expected to gap down, with the Shanghai Composite relatively resilient because of its weighting structure; but today's real risk is not in the indices — it is in that batch of high-level small caps that limited up on Monday with no fundamentals to catch them this morning. On the evening of 9/14, 闽东电力 (Fujian Mindong Electric Power), 昀冢科技 (Yunzhong Technology), 风华高科 (Fenghua Advanced Technology), 博汇科技 (Bohui Technology), 众泰汽车 (Zotye Automobile), and 澳弘电子 (Aohong Electronic) and others collectively issued risk warnings or concept denials — a clear signal of regulatory cooling.

---

1. News Overview

⚠️ One item is retracted in this section: the first draft, based on a search summary, listed "星网锐捷 (Fujian Star-net) plans to spin off 锐捷网络 (Ruijie Networks) for a ChiNext listing." Verification shows this is old news — 锐捷网络 (Ruijie Networks) already listed on ChiNext on 2022-11-21 and therefore cannot be a 9/14 announcement. The item has been deleted. This is the classic pattern of a search summary backfilling a years-old event as "today's news," recorded here to avoid a repeat next issue.

---

2. Strongest Positive Branches, Descending

The counter-conclusion this section must also state: today no branch has positive news hard enough to offset the overnight sell-off in the AI hardware chain. Branch 1 is the only direction with "China–offshore resonance + a policy event in progress," but the fundamentals of its A-share vehicles are very weak (see Section 5).

---

3. Overall Single-Stock Positive-Strength Ranking (arranged in groups by branch, NOT in strict total-score descending order)

Price-limit bands: Shanghai/Shenzhen main board ±10%; ChiNext (30x)/STAR (68x) ±20%; Beijing Stock Exchange ±30%. "Monday performance" in the table below is uniformly 2026-09-14 closing data.

⚠️ Two convention statements: (1) This table is arranged in groups by branch, roughly descending by score within a group, but NOT strictly descending across groups — please do not read 中新赛克 (Sinovatio) at No. 4 (41 pts) as stronger than 万邦医药 (Wanbang Pharma) at No. 6 (62 pts). (2) Wherever "PE" is cited without a specific note, it is static PE (based on FY2025 earnings), the default convention of the quote interface; a positive static PE does not mean current-period profitability (电科网安 (CETC Cyberspace Security) is exactly such a case). Anything involving current-period profit or loss uses "latest reporting period net profit attributable to parent / ex-non-recurring."

The list stops at 16 names. The reason, stated honestly: the number of stocks with "new, verifiable, clearly directed" positive news inside today's window is simply insufficient; padding the list to 30 would require importing large amounts of concept mapping without any supporting announcement, which conflicts with this report's core rules. Better short than watered down.

---

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

Special scoring note for this issue: the four names at No. 13–16 were pushed down to 31/18/12/10 mainly because the "risk deduction" item was maxed out — all of them published announcements between 9/11 and 9/14, by the companies themselves, denying the concept or warning of risk. That is not my inference; it is the issuers' written statement.

---

5. Detailed Analysis of Key Stocks (11 names in total; the ①②③… below are this section's own numbering and do NOT correspond to the Section 3 rankings)

1电科网安002268Shenzhen main board ±10% — Watch closely (the recommendation rationale was overturned by QC and rebuilt; fully disclosed here) · CETC Cyberspace Security

• Related news: the 2026 National Cybersecurity Awareness Week 9/14–20 opened in Jinan, and the "AI Safety Governance Framework" 3.0 was released at the 9/14 opening ceremony, issued by the National Information Security Standardization Technical Committee (TC260) (Chinanews 9/14⁵; the 9/1 CAC release was a pre-event notice and did not contain opening-ceremony content).
• ⚠️ One correction that must be made public: in the first draft I ranked it first, with the core rationale that "PE 205x but positive, a scarce item in a broadly loss-making sector." That rationale is wrong:
  • 205 is static PE (based on FY2025 earnings), not TTM, and certainly not current-period;
  • this stock's latest reporting period net profit attributable to parent is −RMB 171 million, and dynamic PE is −36.17;
  • the company already issued an earnings pre-announcement on 2026-07-15, with a projected loss range of −RMB 160 million ~ −RMB 172 million, information that has been public for two months.
  • In other words: by the same yardstick I use to audit others (2026H1 ex-non-recurring / attributable net profit), 电科网安 (CETC Cyberspace Security) sits in the loss-making group alongside 安恒 (DBAPPSecurity) and 中新赛克 (Sinovatio), with no fundamental advantage at all.
• So why does it remain at the top? Because the rationale was swapped to price-volume pattern, not fundamentals: on Monday it gapped up +3.61%, closed +5.56%, and added another +1.88% intraday — it is the only name in the entire cybersecurity group that could still move higher after gapping up. Peers 三六零 (−1.20%), 深信服 (−0.70%), 安恒 (−1.22%) and 奇安信 (−1.99%) all gapped up and faded. In a sentiment-driven branch, this is the single most informative pattern.
• Industry position: a cybersecurity platform within the CETC system, a central-SOE vehicle and a natural recipient of policy-driven procurement. Segment revenue and the AI-security share cannot be confirmed for now; no fabrication.
• Final call: watch closely — but readers must be clear that this is a "pattern + policy vehicle" judgment, not a "fundamentals" judgment.

2三六零601360Shanghai main board ±10% — Watch closely (also requires correction) · 360 Security Technology

• Related news: same Awareness Week; the data-security concept rose against the tide on Monday (Jiemian¹⁴).
• ⚠️ Correction: the first draft said "Monday's gain was only 3.32%, so it has not been hyped." This contradicts the opening price I myself pulled — prior close 8.74 → open 9.14 (gap-up +4.58%) → close 9.03, intraday −1.20%, a textbook gap-up-and-fade. And in Section 8 I define exactly "gap-up magnitude > close magnitude" as "money using the gap-up to distribute," yet I failed to flag it here. This is the one internal self-contradiction that should least have appeared in this report; it has been corrected.
• Fundamentals (convention corrected): latest reporting period net profit attributable to parent +RMB 216 million, positive (on this point it is stronger than 电科网安 (CETC Cyberspace Security)); but the frequently quoted "PE 83" is the static convention, while dynamic PE is 146.5. Long-weak core-business profitability is a fact, not to be glossed over.
• Why it still stays: float of RMB 63.2 billion makes it the only large-cap on this line with "anchor capacity"; if cybersecurity is to become a main theme, there must be a large-cap vehicle to absorb the money.
• Final call: watch closely, with the verification point changed from "turnover value expansion" to the same specification as 安恒 (DBAPPSecurity): "the gap-up magnitude must be smaller than the closing gain."

3安恒信息688023STAR board ±20% — Watch closely (2nd on fundamentals within the branch) · DBAPPSecurity

• Related news: National Cybersecurity Awareness Week 9/14–20.
• Fundamental verification (2026 interim report pulled independently by the sub-agent): H1 revenue RMB 739 million (+0.86%), net profit attributable to parent −RMB 176 million, ex-non-recurring −RMB 190 million — still loss-making, not glossed over. But the quality of the loss narrowing is the most genuine in the sector:
  • gross margin 55.79% → 56.60%, the only YoY increase among the 6 cybersecurity names;
  • R&D expense RMB 209.4 million, not a cent higher than a year earlier; selling expense −5.1%;
  • FY2025 full-year operating cash flow +RMB 353 million (RMB 161 million the year before), collections improving;
  • receivables at only 26% of TTM revenue, the 2nd best in the group.
• AI revenue (quantifiable): the interim report discloses pure AI revenue above RMB 41 million (≈5.5% of H1 revenue), and including indirect AI products, above RMB 105 million in total (≈14.2%). That figure comes from secondary statistics but is internally consistent when reverse-computed against independently pulled revenue (4,100/73,910 = 5.55%).
• The shortfall that must be stated plainly: revenue growth is essentially zero (+0.86%, Q2 standalone −1.79%) — the loss narrowing comes from "saving on expenses" rather than "growing revenue," and that road does not run far. The debt-to-asset ratio of 45.98% is the 2nd highest in the group, with interest-bearing debt of RMB 411 million.
• Technical sentiment: Monday gap-up +10.80%, close +9.44%, intraday −1.22%, already showing signs that the gap-up was sold.
• Final call: watch closely. It does not have the greatest elasticity, but in a sector where 5 of 6 have negative TTM ex-non-recurring profit, it is the one whose bleeding has most genuinely stopped.

4中新赛克002912Shenzhen main board ±10% — Watch only (as a sentiment gauge only) · Sinovatio

• Related news: overall leader of the cybersecurity sector, 3 consecutive limit-ups on Monday.
• Fundamental verification (sub-agent conclusions, materially different from my earlier judgment; the following is the revision):
  • PE 598 is a trap number. TTM net profit attributable to parent is only +RMB 7 million; the high PE is a product of "a denominator approaching zero" and says nothing about earning power; TTM ex-non-recurring over the same period is −RMB 13 million, still a loss on that basis.
  • H1 revenue of RMB 235 million (+3.96%) looks like positive growth, but the mix has completely deteriorated: the "data and network security products" line that genuinely matches this theme was only RMB 4.75 million, 2.02% of revenue, and down 9.0% YoY; the revenue growth was entirely carried by a 126.4% surge in an "other" line carrying just a 23.8% gross margin (the specific composition of that line is not disclosed by the interface, so it cannot be confirmed for now).
  • Blended gross margin 70.02% → 56.99%, a collapse of 13.03 percentage points; net profit attributable to parent −RMB 129 million, losses widened 92.4%; operating cash flow swung from +RMB 89 million to −RMB 142 million, the only one of the 6 to flip from positive to negative.
• Material matter (primary-source announcement): persons acting in concert with controlling shareholder 深创投 (Shenzhen Capital Group) sold 1 million shares via centralized bidding between 2026-09-07 and 09-08 (0.5856% of total share capital), and the sell-down plan pre-disclosed on 5/26 has only been about one-third executed, with the remaining quota still within its validity period (announcement¹⁵, eiTime 2026-09-09 17:41:10).
• Technical sentiment: Monday sealed at the open, with all-day turnover of only RMB 44 million and a 1.06% turnover rate — the 3 limit-ups were sealed by a very small amount of money.
• Final call: watch only. "3 consecutive limit-ups + security business at only 2% of revenue + controlling shareholder selling down + gross margin collapsing 13pct" — those four facts together point to a pure theme gamble. Its value today is as a thermometer, not as a holding.

5奇安信-U688561PassSTAR board ±20% (the most important falsification in this issue) · QI-ANXIN-U

• On the surface, QI-ANXIN's H1 net profit attributable to parent was −RMB 411 million vs −RMB 770 million a year earlier, "losses narrowed 46.6%" — a very presentable turnaround story.
• But ex-non-recurring narrowed only 1.94% (−RMB 755 million vs −RMB 770 million). Where does the RMB 343 million difference come from? The sub-agent pulled the income statement detail: fair-value change gains of +RMB 370 million (versus −RMB 26 million a year earlier), non-cash and non-operating.
• The core business is deteriorating at the same time: revenue RMB 1.497 billion (−14.09%), gross profit RMB 763 million vs RMB 918 million a year earlier (RMB 155 million less earned), while the three expense lines were cut by only RMB 107 million — cost cutting did not outrun gross-profit leakage.
• The balance sheet is the tightest in the group: receivables RMB 4.719 billion = 114% of TTM revenue; total interest-bearing debt RMB 2.706 billion vs cash of RMB 1.532 billion, net interest-bearing debt about RMB 1.17 billion; negative operating cash flow in FY2024 and FY2025 two years running.
• Call: Pass. This is the most valuable finding of this issue's QC — "losses halved," the number most likely to be written into a research note, is accounting makeup.

6永信至诚688244PassSTAR board ±20% (worst operations + most expensive valuation) · Integrity Technology

• The purest AI narrative: AI-related business revenue of RMB 10.6395 million, 19.6% of H1 revenue, the highest share of the 6 — but the absolute amount is barely ten million.
• Operations are the worst of the 6: H1 revenue RMB 54.34 million (−36.28%), Q2 standalone RMB 22.81 million, −58.68% (a cliff); net margin −99.18%, i.e. the loss is roughly equal to total revenue; gross margin 51.51% → 35.08%, a 16.43pct collapse; all three product lines (digital wind tunnel −26.1%, cyber range −46.9%, security protection −44.6%) declined across the board with gross margins collapsing across the board.
• Forward indicators: contract liabilities of only RMB 3 million (the lowest in the group, near zero), meaning there is almost no prepaid backlog and Q3 revenue lacks visibility; receivables at 118% of TTM revenue (the highest in the group).
• Valuation: PS (TTM) 10.13, the most expensive in the group (4.7x 安恒 (DBAPPSecurity)'s 2.14), and that is against revenue of −36%.
• ⚠️ One circulating error that must be actively corrected: online summaries claim its "digital wind tunnel and operations 2026H1 revenue was RMB 42.6652 million, +13.49% YoY" — that figure is in fact 2025H1 data, a case of mistaking last year's interim report for this year's. By the segment data, 2026H1 digital wind tunnel was RMB 31.53 million, −26.1% YoY.
• Call: Pass. See also the notes in the Section 6 Pass list regarding U.S. sanctions and the information-leak rumor (both historical/unverified matters, not used as a basis today).

7万邦医药301520近岸蛋白 (Novoprotein, 688137) — ChiNext / STAR board ±20% — Watch only · Wanbang Pharma

• Related news: CRO/innovative drugs resonated on Monday, with both names hitting 20cm limit-ups. The industry-side logic is a recovery in global R&D demand and accelerating overseas expansion by domestic innovative drugs.
• One point that must be honestly flagged: the 药明康德 (WuXi AppTec) H1 data cited in media attributions (revenue RMB 28.897 billion, +38.93%; full-year guidance raised to RMB 58.5–60.5 billion) is already-disclosed existing information, not a new catalyst after Monday's 9/14 close. Within this window (after 9/14 15:00) I found no new catalyst for innovative drugs/CRO — today this branch is "day two of Monday's move," not a fresh starting point.
• Technical sentiment: 万邦医药 (Wanbang Pharma) turnover rate 38.71%, 近岸蛋白 (Novoprotein) turnover rate 13.03%; the shareholder base is highly unstable.
• Call: watch only. The branch direction is acceptable (non-AI asset, defensive properties), but the levels of these two names are not.

8天融信002212Shenzhen main board ±10% — Watch only (the most real AI monetization, but drowned by the core business and goodwill) · Topsec

• Monday sealed at the open (gap-up +9.96%, intraday 0.00%), turnover value RMB 407 million, float RMB 8.6 billion.
• Its AI revenue has the highest monetization of the sector: the interim report discloses "direct revenue from AI-enabled security business above RMB 80 million," or 14.7% of H1 revenue (8,000/54,540 = 14.67%, reverse-computed and consistent), corroborated by a second independent source. This point is real, no discount.
• But it is drowned by two things:
  1. Core security revenue RMB 410 million vs RMB 719 million a year earlier, −43.0%; total revenue −33.99%, net profit attributable to parent −RMB 273 million, losses widened 321.8%. Low-margin AI-compute cloud doubled from 12.4% to 24.0% of the mix, dragging blended gross margin from 67.41% to 59.35%.
  2. Goodwill of RMB 4.145 billion = 44.7% of the RMB 9.280 billion net assets attributable to parent. Against core revenue at −43%, that goodwill carries impairment pressure (a judgment, not a public fact; whether it is written down depends on the annual report). PB 0.94 looks "cheap, below book," but tangible PB ex-goodwill is 1.70 — the cheapness is an illusion.
• Liquidity is the tightest of the 6: cash and equivalents of only RMB 268 million (the lowest in the group), short-term borrowings RMB 405 million, H1 operating cash flow −RMB 468 million and still deteriorating YoY, sustained by +RMB 298 million of financing cash flow.
• Call: watch only. This is the classic case of "a single support point holds, the whole does not" — you cannot ignore the problems across the entire set of statements just because the AI revenue line is real.

9深信服300454ChiNext ±20% — downgraded from "watch closely" to "watch only" (the most important attribution correction in this issue) · Sangfor Technologies

• Correction note: in the first draft I listed Sangfor as one of the core recommendations in the cybersecurity branch, on the grounds of "positive PE, first-tier fundamentals." Once the sub-agent pulled the revenue breakdown, that classification is wrong, and it is corrected here.
• Its results are real: H1 revenue RMB 3.998 billion (+32.85%, Q2 standalone +35.69% and still accelerating), net profit attributable to parent +RMB 231 million, ex-non-recurring +RMB 179 million, swinging from loss to profit — the only genuinely profitable one among the 6 cybersecurity names. The balance sheet is also the cleanest in the group: zero interest-bearing debt, zero goodwill, contract liabilities RMB 2.011 billion (7x runner-up 安恒 (DBAPPSecurity)), receivables only 5% of TTM revenue (the lowest in the group).
• But the growth engine is not "security" (Eastmoney revenue breakdown, hard data):
  • cloud computing and AI infrastructure business RMB 2.212 billion, 55.34% of revenue, +58.6% YoY;
  • cybersecurity business RMB 1.587 billion, 39.69% of revenue, only +10.6% YoY;
  • of the RMB 989 million total revenue increment, cloud/AI infrastructure contributed RMB 817 million = 82.6%, while the security segment contributed only 15.4%.
• Why this correction matters: Sangfor is substantively closer to an AI compute name, and that is exactly the line that was slaughtered overnight (SOX −5.86%, 工业富联 (Foxconn Industrial Internet) already −3.90% on Monday). Buying it as a "cybersecurity safe haven" is mistaking a risk asset for a defensive asset — exactly the wrong direction.
• Another point not glossed over: H1 operating cash flow −RMB 476 million, twice as bad as −RMB 235 million in the same period last year; it booked RMB 231 million of accounting profit while RMB 476 million of cash flowed out. FY2025 full-year OCF of +RMB 1.342 billion indicates seasonality, but this point needs the Q3 report to verify whether it is seasonality or a deterioration in revenue quality from fronting cash to sell compute hardware.
• Call: watch only. Its fundamentals are the best in the sector, but today it stands downwind of the turn, not upwind.

10润泽科技300442ChiNext ±20% — Watch closely (the only "counter-narrative" name) · Runze Technology

• Related news: 9/14 evening announcement, applying for an additional RMB 50 billion credit line to step up investment in AIDC (Yicai⁷). In-window, and a hard-cash corporate action.
• The contradiction must be spelled out: this announcement points in exactly the opposite direction from today's dominant global narrative ("should AI capex slow down"). The same piece of news would have read as a strong positive on 9/11 and may read as "levering up against the tide" on 9/15.
• Fundamentals: Monday close 60.68, −0.61%, float RMB 99.3 billion, PE 18.54 — the lowest on the entire list.
• Call: watch closely. It is today's only combination of "low valuation + real investment + in-window announcement," but the narrative headwind makes it unsuitable for participation while opening sentiment is at its most chaotic.

---

6. Pass List

---

7. Within-Branch Ranking

Branch One: Cybersecurity / AI Security

In the "fundamental hardness" column below, only 6 names were independently reviewed by the fundamentals-analyst sub-agent pulling the 2026 interim reports: 深信服 (Sangfor), 安恒信息 (DBAPPSecurity), 中新赛克 (Sinovatio), 奇安信 (QI-ANXIN), 天融信 (Topsec), 永信至诚 (Integrity Technology). 5 names were NOT reviewed by the sub-agent: 电科网安 (CETC Cyberspace Security, 002268), 三六零 (360 Security Technology, 601360), 亚信安全 (Asiainfo Security, 688225), 绿盟科技 (NSFOCUS, 300369), 博汇科技 (Bohui Technology, 688004) — flagged in each row. Among them, 电科网安 (CETC Cyberspace Security) and 三六零 (360 Security Technology) happen to be No. 1 and No. 2 on the final list, which is this issue's greatest evidentiary weakness. Additional note: 电科网安 (CETC Cyberspace Security)'s current-period loss (−RMB 171 million) was verified at the QC stage from the public earnings pre-announcement and the quote interface's f105 field, not by sub-agent review; its segment revenue and AI-security share remain unconfirmable for now.

The hardest single stock in this branch: on reviewed conventions it is 深信服 (Sangfor) (the only profitable one in the group), but its growth comes from AI compute and today it is a headwind asset; in the intersection of "has fundamentals AND is aligned with today's wind," only 安恒信息 (DBAPPSecurity, 688023) and the unreviewed 电科网安 (CETC Cyberspace Security, 002268) remain. Highest trading recognizability: 中新赛克 (Sinovatio, 002912), but it is a gauge, not a holding. Who follows: 亚信安全 (Asiainfo Security), 绿盟科技 (NSFOCUS), 任子行 (Surfilter Network, 300311, Monday turnover rate 25.22%). Who is passed: 博汇科技 (Bohui Technology) (the driver is a negotiated share transfer, unrelated to AI security, wrongly attributed into this branch); 奇安信 (QI-ANXIN) (the loss narrowing is accounting makeup); 永信至诚 (Integrity Technology) (Q2 revenue halved yet still carrying a PS of 10x).

⚠️ The structural weakness this branch must admit: of the 6 reviewed cybersecurity names, only 深信服 (Sangfor) had positive 2026H1 ex-non-recurring profit; the other 5 had combined TTM ex-non-recurring losses of about −RMB 1.84 billion. If this line rises today, the driver is the policy event and the offshore mapping, not earnings. Position size must be set on a "theme" basis, not a "fundamentals" basis.

Branch Two: PCB / Copper-Clad Laminate (avoid this branch as a whole; the ranking here only explains "who to look at if you must look")

The causality this branch must make clear: in this round of CCL price increases, the upstream (生益 (Shengyi)) saw gross margin +9.1pp over two years and ex-non-recurring +111%; the downstream (中京 (Zhongjing) −5.1pp into a loss, 奥士康 (Aoshikang) −4.6pp with ex-non-recurring −57%, 满坤 (Mankun) −2.1pp with ex-non-recurring −34%) are the cost bearers. 中京 (Zhongjing) and 奥士康 (Aoshikang) state themselves in their earnings pre-announcements that the main cause of the loss/decline is price increases in "copper, gold, CCL, copper foil, fiberglass cloth, gold salt and other raw materials" and that "price transmission is subject to a lag." Therefore: buying that batch of Monday's limit-up small caps on AI/price-hike logic is the wrong direction. Another hard fact: the leaders did not start falling on 9/14 — 沪电 (WUS) is −20.5% from its year-to-date high, 深南 (Shennan) −19.7%, 生益 (Shengyi) −23.9%, 胜宏 (Victory Giant) −44.1% (−38.4% over 60 days). The leaders have been correcting for nearly two months while the small caps caught up in the last 20 days (+22%~+52%); that is the late-stage diffusion pattern of a main theme, not a fresh new start.

---

8. Opening Verification Signals for Today

Call-Auction Signals

• 中新赛克 (Sinovatio, 002912) is the single most important observation point: 3 consecutive limit-ups + sealed at the open + only a 1.06% turnover rate. If it is again sealed at the open or gaps up and seals today → the cybersecurity line was not disturbed by the overnight bad news and the branch holds; if the seal breaks and it falls quickly → that line has hit its divergence high for the day.
• The cybersecurity "liquidity group" is the second observation point: 奇安信-U (QI-ANXIN-U, 688561), 安恒信息 (DBAPPSecurity, 688023), 深信服 (Sangfor, 300454) all gapped up and faded on Monday. If "gap-up magnitude > closing gain" recurs today, it means money is using the gap-up to distribute, and the bigger the gap-up, the more dangerous.
• The reversal of gap-up-and-fade in the PCB group: on Monday 科翔 (Kexiang) (gapped down −1.39%, intraday +21.69%) and 中京 (Zhongjing) (gapped down −5.44%, intraday +16.35%) were gap-down-and-rally squeezes; 超声电子 (Guangdong Ultrasonic Electronics) gapped UP +1.12% and then added +8.80% intraday, a different pattern from the others — do not lump them together. If today turns into gap-up-and-fade, that is the signal that money has completed its exit.
• "Can the gap-up hold" is the same yardstick for cybersecurity: on Monday only 电科网安 (CETC Cyberspace Security, 002268) gapped up +3.61% and then added another +1.88% intraday; 三六零 (−1.20%), 深信服 (−0.70%), 安恒 (−1.22%) and 奇安信 (−1.99%) all gapped up and faded. Please run the same criterion across the group today: whoever can still move higher after gapping up is the real anchor.

Sector Signals

• Holds: 3 or more cybersecurity names limit up quickly + at least one of 电科网安 (CETC Cyberspace Security)/三六零 (360 Security Technology)/深信服 (Sangfor) rises on expanding volume (the anchor follows) + a second-limit-up tier appears.
• Does not hold: only 中新赛克 (Sinovatio) seals alone while the back row (亚信 (Asiainfo), 绿盟 (NSFOCUS), 任子行 (Surfilter)) all retreats → a one-day wonder confirmed.
• AI hardware absorption watch: the size of the opening gap-down and the absorption in the first 30 minutes after the gap for 中际旭创 (Zhongji Innolight, 300308), 新易盛 (Eoptolink, 300502), 工业富联 (Foxconn Industrial Internet, 601138) and 寒武纪 (Cambricon, 688256). If they gap down 3%~5% then recover quickly → the overnight bad news is digested; if they keep falling on expanding volume after the gap → a second round of selling begins, which will drag the whole ChiNext Index down.

Single-Stock Signals

• Front-row absorption, fast sealing without breaking, large net order inflows, and whether a stronger name has taken the slot.
• Note Monday's seal-breaking: 桂林旅游 (Guilin Tourism, 000978) broke its seal 8 times, 黄河旋风 (Huanghe Whirlwind, 600172) 8 times, 豫能控股 (Yuneng Holding, 001896) 7 times, 澳弘电子 (Aohong Electronic, 605058) 7 times — the quality of Monday's seals was already poor, and the market's ability to absorb is limited.

Calendar for the Next 5 Trading Days (events a news scan cannot see, listed separately)

Risk Signals

• Gap up then dive, a lone limit-up, back-row retreat, the anchor failing to follow, weak indices.
• Ebb of yesterday's strong theme: if 闽东电力 (Fujian Mindong Electric Power, 000993) gaps down or breaks its seal today → the ebb of the power/compute-electricity-coordination line is confirmed; it has already had the concept denied by the company's own announcement and is the most fragile high-level stock in the market.
• Unlocks: 株冶集团 (Zhuzhou Smelter Group, 600961) unlocks about RMB 7.75 billion today, 42.7% of float; 森合高科 (Senhe High-Tech, 920038) unlocks 55.8% of float.
• Macro: the FOMC begins meeting today (decision 9/17 02:00 Beijing time), the 10-year Treasury yield at 4.998% is approaching 5% — highly valued growth stocks lack a long-friendly environment before the decision lands.

---

9. Final Recommendation Conclusions

① The 5 Stocks Most Worth Watching Today

Position discipline (must be stated): 4 of the 5 names on this list are concentrated on the single "cybersecurity" line. That means they are not 5 independent judgments but 4 expressions of 1 judgment — if the cybersecurity line weakens today, all 4 fail together. This is the structural weakness of this issue's list, disclosed honestly; to diversify, the position cap must be set at the branch level, not the single-stock level.

Two corrections after sub-agent review that must be made explicit:

1. 深信服 (Sangfor, 300454) has been removed from this list. The first draft listed it as a core cybersecurity recommendation, but the revenue breakdown shows that 82.6% of its revenue increment comes from "cloud computing and AI infrastructure" (RMB 2.212 billion, 55.34% of revenue), with the cybersecurity segment contributing only 15.4%. It is substantively an AI compute name, and that is exactly the line that was slaughtered overnight — buying it as a defensive holding is the wrong direction.
2. No. 1 and No. 2 on this list (电科网安 (CETC Cyberspace Security), 三六零 (360 Security Technology)) were NOT independently reviewed by a sub-agent; their "positive PE" comes only from a quote snapshot, and segment revenue and the AI-security share cannot be confirmed for now. This is the greatest evidentiary weakness of this issue's list, flagged honestly rather than papered over with "policy beneficiary."

② The 3 Strongest Branches Today

③ Directions Not Recommended for Chasing Today

1. Small-cap PCB / CCL limit-up names (highest-priority avoid) — 超声电子 (Guangdong Ultrasonic Electronics, 000823), 中京电子 (Zhongjing Electronics, 002579), 科翔股份 (Kexiang, 300903), 满坤科技 (Mankun Technology, 301132), 澳弘电子 (Aohong Electronic, 605058), 奥士康 (Aoshikang Technology, 002913). The reasons to avoid must be re-ordered by strength of evidence:
   • (a) Hardest: all 6 had negative 2026H1 ex-non-recurring YoY growth or outright losses, without exception; and several have already issued announcements by the companies themselves denying the concept or warning of risk (超声 (Ultrasonic) denied the Nvidia claim, 中京 (Zhongjing) self-disclosed a PE of 381.87x with compute HDI at <1%, 澳弘 (Aohong) said after the close on 9/14 that it "found no hot-concept situation").
   • (b) Next hardest: they are the cost bearers, not the beneficiaries, of CCL price increases; buying them on AI logic is the wrong direction.
   • (c) Weakest, already downgraded: the first draft listed "the overnight AI hardware chain crash" as the first reason, now downgraded — because this report has itself demonstrated that these names have almost no AI exposure, so the overnight bad news inherently lacks a transmission channel to them. What is actually killing them is valuation and the shareholder base, not Nvidia.
2. 闽东电力 (Fujian Mindong Electric Power, 000993) and the compute-electricity coordination concept — the company has denied the concept in black and white in its 9/14 evening announcement, and interim revenue is −17.65% with net profit −58.68% and a rolling PE of 371.65x. The level of 4 consecutive limit-ups + a denied concept = the highest-risk combination.
3. Optical modules / optical communications — the conclusion has been downgraded per QC, and is explained clearly here: overnight Coherent −12.73%, Lumentum −9.92%, Ciena −8.55%, AAOI −9.11% were indeed brutal, but the leader of that decline, Corning (−13.70%), fell on its own $2 billion ATM equity offering (a dilution event), and it simultaneously maintained its Q3 core sales guidance — a U.S. company issuing equity carries no demand-side read-through for Chinese optical module makers.
   • The first draft's sentence "a second sell-off is highly likely today" is therefore retracted, replaced by: A-share 中际旭创 (Zhongji Innolight, 300308) and 新易盛 (Eoptolink, 300502) already fell 5.72%/4.99% on Monday, and part of the overnight bad news is event-specific and cannot be simply extrapolated; at the same time, all four PCB leaders closed above their opening prices after gapping down on Monday, showing there is absorptive capacity intraday.
   • The executable formulation: do not chase, do not bottom-fish, wait for absorption to be confirmed — watch whether the first 30 minutes after a gap-down can recover the loss, rather than presetting a direction.
4. MLCC / passive components — 昀冢科技 (Yunzhong Technology) announced that "MLCC is not yet applied in AI compute servers" and 风华高科 (Fenghua Advanced Technology, 000636) warned of cost pressure; the same concept was cooled by two companies on the same day, while 风华高科 (Fenghua)'s Monday turnover rate already reached 19.15% on RMB 12.848 billion of turnover value.
5. Today's unlock names — 株冶集团 (Zhuzhou Smelter Group, 600961) unlocks 42.7% of float, 森合高科 (Senhe High-Tech, 920038) 55.8%.

④ Final One-Sentence Judgment

Today is an "anchor swap in the narrative": global money moved its AI dollars from "the ones making the metal" to "the ones guarding the door." But A-share pricing is more complicated than one sentence — the institutionally priced portion (optical modules, the four PCB leaders, STAR growth) already fell on Monday and saw intraday absorption; what truly has not been priced is that batch of PCB small caps in hot money's hands with almost no AI exposure.

So there are only two executable sentences today: screen for the anchor within the cybersecurity line using the yardstick of "can it hold after gapping up" (the only one that held on Monday was 电科网安 (CETC Cyberspace Security)), and resolutely avoid that batch of high-level names that limited up on Monday and had the concept denied by the companies' own announcements.

But the point must be made in full: this line cannot stand on three things: (1) its reason for rising is a policy event and an offshore mapping, not earnings — of the 6 reviewed cybersecurity names only 1 had positive 2026H1 ex-non-recurring profit, and that one (深信服 (Sangfor)) actually grows on AI compute and today stands precisely downwind; (2) cybersecurity was not even the group with the broadest limit-up count on Monday (components at 6 and power + grid at 6 are both no fewer than its 5); (3) the FOMC begins meeting tonight, and the market is pricing the first hike since July 2023.

Therefore: cybersecurity can be participated in, but only at a "theme" position size, not a "fundamentals" position size.

Because this crash did not cancel a single order — it cut valuation, plus one equity offering of Corning's own; and the batch with the most expensive valuations and the worst earnings is precisely the batch that rose the most on Monday.

---

⚠️ Data-retrieval failures and channel switches this issue (not sent to clients)

1. Eastmoney push2 / clist interfaces were persistently rate-limited on this machine: stock_zh_index_spot_em, clist/get (whole-market pagination), ulist.np/get (batch quotes), and industry-sector fund flow (m:90+t:2) repeatedly returned empty responses or RemoteDisconnected.

   • Consequence: the planned "name-by-name paginated advance/decline count across all 5,913 listings" could not be completed (the threaded version failed at page 1). Advance/decline was switched to the Legu stock_market_activity_legu snapshot (2,939 up / 2,083 down / 185 flat / 12 suspended, statistical timestamp 2026-09-14 15:00:00), which totals 5,219 names, fewer than the whole market's 5,913, with the difference not reconciled name by name — this is exactly the "one-sided bias in Eastmoney clist pagination" trap from memory; it was avoided this time by switching sources, but no cross-validation was done.

2. Sina hq.sinajs.cn returned HTTP 403 (Forbidden, 9 bytes).

   • Consequence: batch single-stock quotes were switched to Tencent qt.gtimg.cn, all successful (including field 30 timestamps verified line by line as 20260914 16:xx, confirming post-close settlement values for Monday).

3. Several akshare interfaces were unavailable: stock_sector_fund_flow_rank (JSONDecodeError), stock_dt_pool_em/stock_zb_pool_em (AttributeError, this version lacks those functions), stock_restricted_release_queue_em (returned an empty DataFrame).

   • Consequence: sector fund flow was switched to Tonghuashun stock_fund_flow_industry (the established fallback channel from memory, verified effective this time); the limit-down/seal-break pools could not be obtained from the official pools, so seal-break counts were taken from the "seal-break count" column that comes with stock_zt_pool_em, and the limit-down stock list is missing this issue.

4. The limit-up pool passed the completeness check as required: getTopicZTPool returned tc=55 and len(pool)=55, consistent, no silent truncation. The Legu snapshot's "limit-ups 57 / true limit-ups 52" differs from the official pool's 55 on convention (ST and Beijing Stock Exchange are handled differently); the body of the report uses the official pool's 55 throughout.

5. The Cailian Press telegraph cls.cn/telegraph is JS-loaded dynamically, and WebFetch retrieved only the skeleton with no content; the Tonghuashun "morning must-read" page returned 9/14's old content (verified, not 9/15). The 9/14 evening announcements ultimately used the Yicai "Evening Announcements" roundup as the primary source.

6. The CNBC quote channel was normal, and change_pct for every ticker was reverse-computed and verified with (last-prev)/prev, with no field conflict across 52 tickers (including .SOX −5.86% and US10Y 4.998%).

7. Both sub-agents returned, and both changed conclusions, which is worth recording:

   • PCB group: included block-trade seat de-duplication and announcement eiTime verification; dug out 超声电子 (Guangdong Ultrasonic Electronics)'s 9/11 denial of the Nvidia claim, 中京 (Zhongjing)'s self-disclosed PE of 381.87x, and 科翔 (Kexiang)'s PCB segment gross margin of −1.39%.
   • Cybersecurity group: directly overturned one recommendation slot in the first draft — I had listed 深信服 (Sangfor) in the top three on "positive PE within cybersecurity," when in fact 82.6% of its revenue increment comes from AI compute infrastructure, an attribution mismatch, since downgraded and publicly corrected in the body. It also caught that QI-ANXIN's "losses narrowed 46.6%" was in fact +RMB 370 million of fair-value changes, and that 中新赛克 (Sinovatio)'s security business is only 2.02% of revenue.
   • Still not covered: 电科网安 (CETC Cyberspace Security, 002268) and 三六零 (360 Security Technology, 601360) — and they happen to be No. 1 and No. 2 on the final list. The body flags "not reviewed by sub-agent" at every relevant point; unreviewed conclusions must not be carried over into the next issue.

8. Data that could not be obtained (left blank in the body, not fabricated): segment revenue and AI-security share for 电科网安 (CETC Cyberspace Security)/三六零 (360 Security Technology), new catalysts for the innovative-drug branch after 9/14 15:00, the limit-down stock list, the body text of 中新赛克 (Sinovatio)'s 9/12 unusual-movement announcement, and third-party market-share rankings (IDC/CCID).

9. One proactive correction placed on record: the circulating claim that "永信至诚 (Integrity Technology)'s digital wind tunnel had 2026H1 revenue of RMB 42.6652 million, +13.49%" is 2025H1 data mistakenly transplanted as 2026H1; the actual 2026H1 figure is RMB 31.53 million, −26.1%. The sub-agent spotted it by comparing segment data year by year, and the body now carries the warning. This kind of "year misalignment" belongs to the same family as the "recurring column headlines hide the publication date" trap in memory.

10. risk-auditor QC results and disposition (QC changed 9 items this issue, 3 of which shook recommendation slots):

    • [Fixed · most severe] 电科网安 (CETC Cyberspace Security)'s "PE 205 but positive, one of the sector's few non-loss-makers" is wrong. 205 is static PE (FY2025); the measured Eastmoney f105 latest reporting period attributable net profit is −RMB 171 million with dynamic PE −36.17, and the company already issued a loss pre-announcement on 7/15. The sole fundamental rationale for the top slot was falsified; that rationale has been deleted and rebuilt on the "gap-up holds" price-volume pattern, with the correction made public in the body rather than silently edited. The root cause is that I took the quote interface's default static PE as current-period earning power — the same family as "horizontal comparison must use one yardstick" in memory.
    • [Fixed] I wrote that 三六零 (360 Security Technology) "had not been hyped," yet the opening price I pulled myself was a gap-up of +4.58% with −1.20% intraday, a gap-up-and-fade, and in §8 I define exactly that pattern as a distribution signal. This is an internal self-contradiction, not missing information — matching the memory "self-contradiction beats missing info."
    • [Fixed] The real cause of Corning's −13.70% is its own $2 billion ATM equity offering (Goldman Sachs sole agent), not the Amodei narrative; the company simultaneously maintained Q3 guidance. The drivers have been split, and the conclusion "A-share optical modules will likely sell off a second time today" has been retracted.
    • [Fixed] The FOMC direction was written with the wrong sign of magnitude: the market is pricing a hike to 3.75%–4.00% (the first since July 2023), while the first draft only wrote "lacks a long-friendly environment before the decision lands." Probabilities range from 66% (8/31) to 80%+ across sources, and this issue adopts only the direction, not any single probability number (I could not independently verify the auditor's 93%, so it was not written in).
    • [Fixed] §1 had listed "星网锐捷 (Fujian Star-net) plans to spin off 锐捷网络 (Ruijie Networks) for listing" — old news; 锐捷网络 (Ruijie Networks) listed on 2022-11-21. Deleted, with a trace left.
    • [Fixed] The industry breakdown of the 55-name limit-up pool had never once been written; filling it in exposed a fact unfavorable to this report: cybersecurity (5) is not the broadest group, with components (6) and power + grid equipment (6) both no fewer. Self-disclosed in §0.5, and Branch 1's basis is now explicitly limited to "policy event + offshore mapping."
    • [Fixed] 超声电子 (Guangdong Ultrasonic Electronics) had been written as "gapped down," when in fact it gapped up +1.12%; the self-generated counter-evidence that the four PCB leaders "all closed above their opening prices after gapping down" had not previously been adopted and has now been added.
    • [Fixed] The §3 header "descending by single stock" did not match the actual arrangement (41 pts ranked ahead of 62 pts); it has been changed to "grouped by branch" with a convention statement added; the §5 title "Top10" actually covered 11 names with duplicated numbering, and has been re-ordered.
    • [Not adopted · 1 item] The auditor said "凯盛新能 (Kaisheng New Energy) (3 limit-ups, glass/fiberglass) is precisely the most upstream part of the price-hike chain argued in §7, and the market has already corroborated it." Verification shows the stock belongs to photovoltaic glass, not the same chain as electronic-grade fiberglass cloth/CCL, and writing it as stated would manufacture a false mutual corroboration. The body handles it as "the driver could not be verified, honestly left blank." Skepticism itself must also pass through a primary source.

---

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