A-Share Pre-Market Brief | 2026-09-08 (Tuesday)

Scope and caliber boundaries of this issue (judge the strength of the conclusions accordingly) ① News window = 2026-09-07 15:00 (previous session close) → 2026-09-08 07:10. Market-wide filings inside the window come from Eastmoney np-anotice-stock; 2,000 items pulled, 1,839 in-window items scanned one by one, and the publication timestamp is always taken from eiTime (not notice_date). ② Quotes are on the 2026-09-07 close caliber (Tencent qt.gtimg.cn); for individual stocks we give today's open / high / low / close, not just the close. ③ The limit-up / limit-up-break / consecutive-limit-up pools are Eastmoney's official pools push2ex, with len(pool)==tc verified (93==93). ④ Limit-up determination is bucketed by board price limit: SSE/SZSE main boards ±10%, ChiNext/STAR ±20%, BSE ±30%. Every stock in this issue is tagged with its board. ⑤ Sector moves and fund flows are on the Tonghuashun (THS) caliber (stock_board_industry_summary_ths); its "net inflow" = inflow − outflow, which is not the same yardstick as Eastmoney's "main-force net inflow" and the two must not be subtracted across issues. ⑥ The dragon-tiger (LHB) board is the 9/7 exchange caliber, and this issue did obtain it (generated at 9/8 07:13, past the 18:00 publication time). ⑦ No northbound-flow data (daily net buy stopped being disclosed as of 2024-08-19); this issue gives no northbound figures. ⑧ U.S. markets were closed all day on 9/7 (Monday) for Labor Day, so this window has no overnight U.S. confirmation; tonight's U.S. reopen is the first U.S. verification. ⑨ ⚠️ But "there is no external confirmation at all" is wrong — Hong Kong was inside the window. Hong Kong closes at 16:00, later than the A-share 15:00 close, so for dual-listed A+H names the 15:00–16:00 hour of Hong Kong trading is the only real-time external vote inside this window, and it happened before the MIIT plan and the Goldman Sachs TAM hit. This issue has added that reading (see §1.5), and it materially changes today's gap-up threshold judgment.

⚠️ Data-pull and search log (this section is not sent to clients):

I. Channels that worked

• Quotes/indices via Tencent qt.gtimg.cn (including PE/PB/market cap), getting full OHLC + turnover rate + traded value in one call.
• The limit-up pool via push2ex, with len(pool)==tc verified per memory [[limit-updown-must-use-official-pool]] (93==93), and the consecutive-limit-up structure broken out by lbc.
• Sector quotes / sector fund flows / single-stock fund flows via THS; advance-decline counts via stock_market_activity_legu.
• stock_lhb_detail_em('20260907') successfully returned 70 rows this time — the only difference from the previous issue is the generation time (15:30 vs 07:13), closing the loop on memory [[lhb-publishes-after-1800]] for the fourth time. The pre-market slot is naturally after 18:00, so a pre-market issue can always get the previous day's LHB. This is an information advantage of pre-market over post-close and should be permanently built into the workflow.
• Share-lockup expiries via stock_restricted_release_summary_em + detail_em, with a separate next-5-day calendar table added per memory [[news-window-cannot-see-calendar-events]], which directly caught today's single biggest risk point (Zhongjuxin) this time.

II. Failed channels / data-pull faults this run

• ⚠️ Eastmoney push2 clist primary cluster returned HTTP 502 — the fourth consecutive trading day. Per memory [[ths-fallback-when-em-clist-down]] the whole path fell back to THS and did not degrade into "no sector-level data obtained"; the THS int64 code column swallowing leading zeros was handled with zfill(6), 23/23 all matched, no silent MISSING.
• ⚠️ ak.stock_zh_a_hist('300308') gave RemoteDisconnected; switched to Tencent web.ifzq.gtimg.cn forward-adjusted daily bars and obtained the 60-day series.
• ⚠️ Sina hq.sinajs.cn returned 403 (Referer required); replaced with the Tencent source, no impact on conclusions.

III. Four methodological cross-checks that were actively performed and that changed conclusions

1. ⚠️⚠️ [Most important in this issue] I at one point wrote the wrong judgment that "the CPO line has no overnight catalyst", and only corrected it by digging further. It is true U.S. markets were closed on 9/7; Goldman's coverage of Innolight happened intraday at 09:33 on 9/7 (already absorbed by that day's +10.38%), and GPT-6 Astra is 9/4 old news — none of the three is a new overnight catalyst. But the MIIT "15th Five-Year Plan for the Development of the Information and Communications Industry" was released on the evening of 9/7, and Goldman's optical-module TAM upgrade was published at 21:17 on 9/7 — both are inside the window and both are entirely new. Had I stopped after the first round, I would have dropped the strongest policy catalyst of the day entirely. This is isomorphic to memory [[no-news-found-is-not-no-news]].

2. ⚠️ [The caliber gap between the document's wording and media paraphrase has been separated] Eastmoney's "Selected News on the Evening of September 7" summarized the plan as "deploying 400G-and-above high-speed transmission technology… promoting the downward extension of 100G optical network equipment"; but China Daily's caliber makes clear that among the 13 headline indicators there is no optical-module-specific figure, and the task wording quoted by Jiemian is "strengthen R&D of ultra-high-speed, large-capacity optical transmission, optical access and submarine-cable optical communication technologies, and accelerate R&D and application of new fiber technologies such as hollow-core fiber". This issue writes the "indicator layer" and the "task layer" separately and does not treat media summaries as the original text of the document, avoiding the "paraphrase mistaken for primary source" error of memory [[recurring-column-headlines-hide-date]].

3. ⚠️ [Seal-order quality of yesterday's limit-ups was recomputed name by name, and it contradicts the surface impression that "the sector is very strong"] All 14 limit-ups in Components were first boards, and the seal-order/turnover ratio was generally only 2%–6% (WUS RMB 179 million / RMB 9.614 billion = 1.9%; Suntak RMB 2 million / RMB 1.334 billion = 0.15%), and several only sealed in the afternoon (WUS 13:31, Accelink 13:18, Jianghai 13:00, Guanghe 14:13). The number "14 limit-ups" by itself conceals how poor the seal quality was.

4. ⚠️ [Moore Threads' limit-down was traced to a lockup expiry, and that was directly migrated into today's risk criterion] Five institutional seats net sold RMB 1.308 billion and the STAR-listed stock hit limit-down, because the IPO offline-placement shares unlocked on 9/7 (free float 30.22 million → 56 million shares). Following that thread into today's lockups, I found Zhongjuxin (688549) unlocks RMB 19.78 billion of market value today, equal to 150.7% of its pre-unlock free-float market cap — a bigger ratio than Moore Threads'. This is the only criterion in this issue that "uses yesterday's realized causality to pre-judge today".

5. ⚠️⚠️ [Second self-reversal: the technical position of the whole line was wrong, which changed the No. 1 recommendation] The first draft ranked Zhongtian Technology No. 1 on the pivot "PB 2.90, lowest in the field". Before finalizing I pulled 60 trading days of forward-adjusted data and found the entire optical-communications complex topped out together on 2026-06-22~25, and since then: FiberHome −55.4%, Zhongtian −51.8%, Hengtong −48.8%, Yongding −47.3%, Innolight −36.6%, YOFC −33.2%, WUS −22.5%. That is, yesterday was not "a new main line igniting", it was "a bounce after 2.5 months of decline"; and Zhongtian's low PB is the result of being cut in half, not a value signal (and I found its energy-storage project IRR fell from 9.34% to 6.22%, traditional new-energy revenue −18.65%, and it hit limit-down on 7/13). On that basis the primary ranking variable was switched wholesale from "valuation level" to "relative strength (distance from the 60-day high) + institutional money"; the No. 1 recommendation was changed from Zhongtian Technology to WUS Printed Circuit, and Kinwong Electronic (the only name in the whole complex standing on its 60-day high, −0.0%) was added to the recommendation list. This is the mirror image of the same class of error as memory [[low-pe-high-pb-is-cycle-top]]: before citing a low valuation you must first answer "why is it cheap".

6. ⚠️ [The sentence "the top 7 are all PCB/optical modules" was downgraded in precision] The first draft said that; on checking, No. 7 Jianghai belongs to capacitors (Components), not PCB. It has been changed to an explicit name-by-name list (4 PCB + 2 optical communications + 1 capacitor). A small error, but a textbook case of "sacrificing accuracy for a tidy sentence" — noted.

7. ⚠️ [GigaDevice was kept out of the recommendation list] I considered using it as a "memory leg" to diversify the single variable, but after pulling the data found it is −54.5% from its 60-day high and −4.4% over 20 days — worse than the Zhongtian I had just downgraded. Without that check I would have stuffed in a stock weaker than the one I eliminated, purely to "look diversified".

8. ⚠️⚠️⚠️ [Third and most serious reversal: the entire policy logic of the No. 1 recommendation was wrong] The pivot that pushed Zhongtian Technology to No. 1 (first draft) was "MIIT named submarine cables". After primary-source verification: the document was issued by the information and communications regulator, and "submarine-cable optical communication" means submarine optical cable (transoceanic communications); whereas over 40% of Zhongtian's marine-segment revenue is 500kV EHV power submarine cable (per the interim report's own text), whose policy parent document is the offshore-wind plan. Same words, completely different channel. Moreover, marine accounts for only 12.09% of its revenue (power grid at 42.41% is the largest), and marine backlog even fell from RMB 12.1 billion to RMB 11.3 billion (−6.6%). The names that genuinely match "submarine-cable optical communication" are Hengtong Optic-Electric (which separately discloses marine-communications backlog of about RMB 7.0 billion and issues individual contract-win announcements, e.g. RMB 1.831 billion), and the name with actual numbers on "hollow-core fiber" is YOFC (RMB 159 million tender won at 100%, cumulative delivery of over 10,000 fiber-km). This is a textbook relapse of memory [[right-entity-wrong-channel-and-clock]]: once the name matched, I stopped asking "does this leg actually connect".

9. ⚠️⚠️ ["Zero external confirmation" was wrong — Hong Kong was inside the window, and I drew the window boundary incorrectly] The first draft repeatedly stressed "U.S. markets closed, no external confirmation in this window, tonight is the first vote", and on that basis set a "gap-up >6% = overextended" threshold. But Hong Kong closes at 16:00, later than the A-share 15:00 close, and that hour is inside the window: Innolight 03308.HK +19.58% closing at its high, Cambridge Technology 06166.HK +24.37%, while Tencent was −0.99% (i.e. not a broad rally). This not only overturns "zero confirmation", it simultaneously dissolves the pseudo-contradiction of Cambridge "limit-up yet net outflow", and voids the first draft's gap-up threshold entirely. Memory [[dual-listing-cross-check-beats-thin-volume-prior]] is exactly about this channel, and this time it never even crossed my mind to look.

10. ⚠️⚠️ [A criterion I used 5 times is itself wrong] The first draft used "a limit-up stock had net fund outflow on the day" as a negative/contradiction in 5 places. Yasheng Group was a one-word board, and its "net outflow of RMB 102 million" happens to equal its entire day's turnover of RMB 102 million — a net figure cannot equal total turnover, and this single counterexample is enough to falsify the whole criterion. Root cause: that caliber is an inside/outside-bid split of all trades (inflow + outflow = turnover), and when a limit-up is sealed shut every executed trade is booked as inside-bid. All such statements have been withdrawn, and the corresponding verification point in §8 has been removed. Lesson: before a criterion is used for a conclusion, it must first be tested on a sample where it ought to fail.

11. ⚠️ [A superlative claim was made without a full-sample check] The first draft said Cambridge Technology had "the only double-digit seal-order ratio in the field, best seal quality in the field". After sorting the full sample descending: 8 names had double digits and Cambridge ranked 6th (MeiG Smart 112.0%, Gongjin 26.6%, Huamai 25.9%, Mingpu 20.9%, Ultrasonic Electronics 20.4% all ahead of it, and the top four had zero limit-up breaks); and Cambridge itself broke the board once intraday. Worse, the break-count field zbc was in the very push2ex pool I had already verified, I just never pulled it. Also corrected: of the 93 limit-ups in the field, 57 broke the board at some point intraday. Any "most/only" claim must first be run against a full descending sample with the top three listed — this is now set as a hard rule.

12. ⚠️⚠️ [The limit-up pool's industry mix was only used to count the branch I already cared about, missing two main lines] The first draft expanded Components (14) and Telecom Equipment (8), but said not a word about the third-largest cluster, Auto Parts (7, the robotics/actuator line), or the fourth-largest, General Retail (6) — and that distribution table had in fact already been pulled and printed in the first draft. Now added: the robotics line (catalyst: NVIDIA's roughly $12.9–13.0 billion acquisition of Hugging Face on 9/3) is listed as Branch 7; General Retail's 6 names have no identified catalyst, and are honestly tagged "attribution unknown" with no fabricated reason. This is a verbatim relapse of memory [[ztpool-industry-mix-only-counts-target-branch]].

13. ⚠️ ["No catalyst in the window" was mistaken for "no catalyst"] The first draft judged agriculture as having "no primary catalyst at all". In fact CBOT wheat hit its highest since 2023-02 on 8/28, +54.5% year to date, and corn hit its highest since 2023-07 (USDA August WASDE cut corn yield + Black Sea situation). The root cause is that the scan channel was only the 1,839 Eastmoney filings, and commodity futures / overseas M&A / same-day product launches are inherently not in the filing stream. The same root cause also caused Huawei's Mate XT 2 tri-fold (9/7 launch event) to be missed. A standing check item has been added: beyond the filing stream, commodity prices, overseas M&A and same-day launch events must also be scanned. Memory [[no-news-found-is-not-no-news]] relapsing for the Nth time.

14. ⚠️ [Only the most aggressive sell-side house was cited] The first draft singled out Goldman's HK$3,267 as "opening up upside". On the same day Citi also initiated Innolight's H shares at Buy with a target of only HK$1,524 — less than half of Goldman's. Both are now disclosed side by side. When several major houses appear on the same name in the same window, they must be listed together; keeping only the most aggressive one is not allowed.

IV. What this issue did not obtain / did not do (listed honestly)

• The original PDF of the MIIT plan was not obtained; all plan content here comes from cross-verification of four paraphrasing outlets (Xinhua / China Daily / China News Service / Jiemian), and the complete list of 26 key tasks and 13 special columns was not obtained, so no conclusion is drawn on "whether optical modules get their own column".
• The exact release times of August CPI/PPI and financial data are unconfirmed. The NBS release-calendar page did not yield structured data; the customary practice is mid-September, and this issue treats it as "release time cannot currently be confirmed" and does not write it into the calendar table as a confirmed item.
• The caliber of Innolight's A-share target price of RMB 2,645 could not be closed out: several paraphrases agree (Cailian / Jrj / CnFol), but against the 9/7 close of RMB 898.46 it implies about +194% upside, inconsistent with the usual magnitude of a 12-month target. This issue cites only "Goldman initiates H shares at Buy, target HK$3,267" (the A/H pair of numbers is more internally consistent); the A-share target is a footnote flagged as doubtful and is not a pivot for any recommendation.
• Exclusive attribution for the CPO line's 9/7 rally is still not fully closed out: both the policy and the Goldman TAM were published after the close, so they cannot explain the intraday move; the only intraday attributable item is Goldman's coverage of Innolight (09:33). This partially closes an item carried over from the previous issue, but it remains incomplete.

---

0. One-sentence summary for today

0.0 Yesterday's (9/7) market baseline readings

Any judgment about "strong or weak today" needs a baseline, and this section always provides one. Tencent quotes + THS stock_market_activity_legu + the push2ex official pool.

⚠️ What this table is here to warn about: yesterday was not a broad rally. The SSE was up only +0.07%, the SSE 50 fell, and 1,987 stocks closed lower — a structural move highly concentrated in tech growth (ChiNext +3.41% vs SSE 50 −0.41%, a 3.8pp gap). The "Components sector: 63 up, 0 down" mentioned later is the only zero-decliner sector among all 90 industries market-wide, an extreme outlier that must not be extrapolated into "market sentiment is broadly euphoric".

0.1 One-sentence summary

After yesterday's close a genuine ministry-level document landed: the MIIT "15th Five-Year Plan for the Development of the Information and Communications Industry" — by 2030, cumulative investment in information infrastructure of RMB 3.8 trillion, intelligent computing power of 9800 EFLOPS, advanced storage capacity of 1700 EB, and 1 million 50G-PON ports. This is the only S-grade catalyst in this window, and it landed on the line that was already the strongest yesterday.

A second shoe dropped the same evening: Goldman Sachs sharply raised its global optical-module market forecast, lifting the 2028 size by 115% from its prior forecast to about $148.5 billion (Cailian Press, 9/7 21:17). Policy (demand certainty) + sell-side TAM (upside opened) hitting the same supply chain on the same night — this is today's hardest combination.

But today's difficulty is not "which line to pick", it is "at what price to get on". Yesterday Components had 14 limit-ups and Telecom Equipment 8, all of them first boards (of the 93 limit-ups market-wide, 80 were first boards and only 1 was a 6th board), and seal quality was poor (WUS's seal order was only 1.9% of turnover; several only sealed in the afternoon). This line is on day 2 today, its first test of disagreement, and the catalyst is being added exactly at this moment — a gap-up is close to certain; whether it can be absorbed is the question.

Direction of money: the top 7 net buys on yesterday's LHB all fall inside the computing-hardware complex — 4 PCB names (WUS +RMB 1.743 billion / 4 institutional seats, Kinwong +RMB 1.036 billion / 3 institutions, Fastprint +RMB 593 million / 3 institutions, Shennan Circuits +RMB 550 million / 3 institutions), 2 optical-communications names (Yuanjie +RMB 1.317 billion, Accelink +RMB 1.097 billion / 4 institutions), and 1 capacitor name (Jianghai +RMB 461 million / 4 institutions). Institutions are participating with real money, not just hot money. Driver type: policy + overseas industry (sell-side TAM), dual-driven, not order-driven.

⚠️ But there is one piece of background more important than all of the above, and it must be stated first: this line is not starting to rise, it is bouncing. The entire optical-communications complex topped out together on 2026-06-22~25 and has since fallen for two and a half months: FiberHome is −55.4% from its 60-day high, Zhongtian −51.8%, Hengtong −48.8%, Yongding −47.3%, Innolight −36.6%, YOFC −33.2%, WUS −22.5% (Tencent forward-adjusted daily bars, 60-trading-day window). Yesterday's surge happened inside a sector that has been falling for 2.5 months; it is a bounce within a downtrend, not the ignition of a new main line. This changes two things at once: ① three consecutive months of trapped supply sit overhead, so the natural resistance to a bounce is far greater than for a "new theme"; ② "cheap valuation" here is usually the result of the fall, not undiscovered value — before citing anyone's low PB you must first ask "why did it fall by half".

Pre-market state judgment: a gap-up is likely, but the threshold for "how much of a gap-up counts as overextended" has been rewritten by Hong Kong. Four hard constraints: ① of the 93 limit-ups yesterday, 57 broke the board intraday, and seal orders were generally under 5% of turnover, so absorption is weaker than it looks; ② second-tier optical-module valuations are already at PB 26–67x; ③ the whole line is still in a −22%~−55% hole with heavy overhead supply; ④ U.S. markets were closed Monday, but Hong Kong already voted inside the window — Innolight H shares +19.58% closing at the high, Cambridge Technology H shares +24.37% (§1.5) — so tonight's MRVL/COHR is the second, not the first, external verification.

Primary stock-selection variable: relative strength, not valuation. This issue ultimately gave the recommendation slots to the names with the shallowest drawdown and institutions buying (Kinwong −0.0% from its 60-day high, WUS −22.5%, Shengyi −22.8%), rather than the ones that "fell the most so they're cheapest" — the first draft ranked them exactly the latter way and that was overturned before finalizing; see §5① and §9①.

---

1. News overview

Window: 2026-09-07 15:00 → 2026-09-08 07:10. Impact grades S/A/B/C.

⚠️ Two items that "look like news but are not" are flagged here, so that second-round hype is not treated as a brand-new positive:

• GPT-6 Astra was released on 9/4 (Beijing time), not last night; it is part of the backdrop to the 9/7 rally, but it is no longer incremental for today.
• CXMT's LPDDR6 debut in the Xiaomi 18 Fold was already announced by CXMT on 8/29; last night was only the product's formal launch. Between "preannouncement" and "delivery" the share price has already moved (CXMT +6.70% on 9/7), so it should not be priced today as a brand-new positive.

---

1.5 ⚠️ The external vote inside the window: the Hong Kong close (added before finalizing this issue)

A shares close at 15:00, Hong Kong at 16:00 — that extra hour falls inside this issue's news window. For dual-listed A+H optical-module names, Hong Kong provided a reading that A shares had no time to reflect that day.

This table overturns two conclusions from this issue's first draft, and that must be said explicitly:

1. "There is no external confirmation in this window" is wrong. U.S. markets were indeed closed, but Hong Kong voted inside the window, and voted far harder than A shares — Innolight H shares +19.58% and closing at the day's high, Cambridge H shares +24.37%. Against Tencent at −0.99%, this was not a broad Hong Kong rally but directed bidding by money on the optical-module line specifically.
2. Cambridge Technology's contradiction of "limit-up yet RMB 1.7 billion of net fund outflow" is most likely a pseudo-contradiction (the real reason is in §1.6), and its H shares were +24.37% the same day with no price limit — in a market with no ±10% constraint, the same company was bid to +24%. That says more about true demand intensity than any fund-flow caliber.

⚠️ Direct implications for today's open (the most executable item in this issue):

• Cambridge's A-share +10.00% is a price truncated by the price limit, not a market-clearing price; its H shares reached +24.37% without a limit, so there is roughly 14pp of unreleased spread in the A shares.
• Innolight H shares closed at HK$1,203 ≈ RMB 1,097 (at a 0.912 exchange rate), while the A shares closed at RMB 898.46 — the H shares are at about a 22% premium to the A shares, i.e. the AH relationship has inverted (the norm is an H-share discount).
• Therefore the first draft's threshold of "a gap-up >6% counts as overextension" is too low and has been voided. Given the H-share reference of +19.58%, an A-share gap-up of 5%–8% is a reasonable range for convergence toward the H-share price; what really deserves caution is failing to hold after the gap-up and falling back quickly, not the size of the gap itself.

⚠️ Caliber note: the HKD/RMB conversion is estimated at 0.912, and the rate is my own conversion, not an official mid-price; the AH premium moves daily with the exchange rate and both markets' prices, so it is used here only for direction, not as a precise arbitrage basis. Also: the AH inversion may itself reflect southbound-flow structural factors, so this issue does not treat it on its own as a bullish reason, only as a refutation of the false premise of "zero external confirmation".

---

1.6 ⚠️ A criterion that must be withdrawn: "net fund outflow on a limit-up stock" is a caliber artifact, not a divergence

The first draft used "a limit-up stock had net fund outflow on the day on the THS caliber" as a negative or a contradiction in 5 places (Cambridge −RMB 1.703 billion, Kersen −RMB 251 million, Kinwong −RMB 169 million, Yasheng −RMB 102 million, Longban −RMB 101 million). Before finalizing, checking showed the criterion itself is wrong, and all of it is withdrawn.

Evidence 1 (arithmetic self-proof): on that caliber, "inflow + outflow = the day's turnover". For Cambridge, RMB 921 million + RMB 2.624 billion = RMB 3.545 billion = its turnover — showing this is a split of all trades by buy/sell direction, not a main-force/large-order net figure. Evidence 2 (a decisive counterexample): Yasheng Group was a one-word board, and its "net outflow of RMB 102 million" happens to equal its whole-day turnover of RMB 102 million. A genuine net fund figure cannot equal total turnover. The reason is simple: when a one-word board / limit-up is sealed shut, buy orders all sit unfilled at bid-1, and everything that does execute is a seller hitting bid-1, so it is all booked as "inside bid / outflow". The tighter the seal, the larger the "net outflow".

Conclusion: for limit-up stocks this inside/outside-bid net figure carries no information, and its sign is if anything inverted. All related statements have been deleted from this issue, and the verification point "the three names with money-versus-price divergence" has been removed from §8. Lesson: before a criterion is used for a conclusion, you must find a sample on which it ought to fail and test it — the one-word board is the touchstone here.

---

2. Strongest positive branches, descending

Branch notes

Why Nos. 1 and 2 are written separately: they share the same catalyst (computing power), but the evidence types differ. Optical modules got a valuation-side positive of the "TAM raised 115%" kind; PCB got real institutional money on the LHB (6 of the top 7 net buys are PCB/components, most with 3–4 institutional seats). A valuation-side positive can be fully spent in a day; institutional accumulation usually is not. That is why I rank PCB's "holdability" above optical modules while ranking its "elasticity" below.

Why No. 3 (optical fiber) is placed high but was not given the No. 1 recommendation slot: it is indeed the only sub-branch that was "named directly by the document yet rose only mildly yesterday" (YOFC +5.73%, Hengtong +2.86%, Zhongtian +3.22%, Yongding +4.92%, not one limit-up), and 1 million 50G-PON ports is a hard number written into the 13 headline indicators. But the claim of "the largest catalyst-price gap" has to be discounted after checking the 60-day series: the four fiber names are respectively YOFC −33.2%, Yongding −47.3%, Hengtong −48.8%, Zhongtian −51.8%, and FiberHome −55.4% from their 60-day highs — they "didn't rally" not because they were overlooked, but because the whole line is in a deep correction. That gap may be an opportunity, or it may be the market's ongoing pricing of fiber-industry conditions. This issue's handling: give it branch rank 3 and give YOFC one recommendation slot (it has the shallowest drawdown and strongest momentum on this line), but do not push the whole line as a "low-valuation bargain zone".

⚠️ Appendix 1: full industry distribution of yesterday's limit-up pool (added before finalizing; the first draft looked only at the two industries inside its own narrative)

The first draft expanded only Components (14) and Telecom Equipment (8) and said nothing at all about the third- and fourth-largest clusters, causing two main lines to be missed. This is a recurring failure mode, so this issue lays out all 40 industries and gives an explicit keep/drop conclusion for the top 6.

Lesson: this "limit-up pool industry distribution" table had in fact already been pulled in the first draft, but it was only used to count "how many names my chosen branch has". The correct use is to look at it first and then define branches, not to define branches first and then go back to the table for evidence.

Appendix 2: panorama of the optical-communications complex's technical position (this issue's primary ranking variable)

Tencent forward-adjusted daily bars, 60-trading-day window (2026-06-15 ~ 09-07); "distance from the 60-day high" uses the highest intraday price within the interval.

This table has only one thing to say: the same policy positive landing on Kinwong at −0.0% and landing on FiberHome at −55.4% is not the same business. The former is "a strong stock receiving fresh ammunition"; the latter is "a deeply fallen stock getting a reason to bounce". This issue's recommendation slots go to the former first, not because it is cheaper (quite the opposite — Kinwong's PE of 89.7 is higher than Zhongtian's 30.0), but because in a sector that has just fallen for two and a half months, "not yet beaten to a pulp" is itself the scarcest evidence.

Why foldables were raised from A in the first draft to A+: when the first draft wrote about "two hard flaws" it counted only two foldable events and missed Huawei's Mate XT 2 tri-fold (launched 9/7, from RMB 19,999, on sale 9/12), and therefore underestimated the event density. The reality is three foldable launches in four days: Huawei 9/7 → Xiaomi on sale 9/10 → Apple in the early hours of 9/10 → Huawei on sale 9/12. At that density the characterization of "an old theme" does not hold — what was hyped in April was "trial-production rumors"; this time three device makers are putting products on shelves in the same week.

But two genuine risks are unchanged: ① timing mismatch — Apple's event is at 1 a.m. Beijing time on 9/10, so the outcome is known before the 9/10 open, meaning today and tomorrow are pure expectation games and the third day's open faces "sell the news"; ② on the A-share side, none of the five names confirms Apple (see §6.1), and their 9/7 price ranking is almost exactly the inverse of their Q2 fundamentals ranking. The foldables chain has fallen across the board over the past 5 days (−0.8%−8.1%), showing the market does not believe it right now — which is both an opportunity (no front-running) and a warning (it may simply never ferment). ⚠️ One data correction to the first draft: "−25%−47% from the 60-day high" does not hold for Kersen Technology (its actual figure is about −10%).

---

3. Overall ranking of single-stock positive strength

Sorted by total score descending. Component scores are in Section 4. The board column states the price limit, because ±10%/±20%/±30% directly changes both limit-up determination and the risk scale.

Not on the list but requiring a separate call-out (negative): 中巨芯 Zhongjuxin (688549, STAR ±20%) — unlocks 888 million shares today, RMB 19.78 billion of market value, equal to 150.7% of pre-unlock free-float market cap. Yesterday Moore Threads hit limit-down at −20% within 4 minutes of the open on the same kind of event (free float 30.22 million → 56 million shares), with 5 institutional seats net selling RMB 1.308 billion. Zhongjuxin's unlock ratio is even larger and the company is not yet profitable (PE −3083). Avoid today without exception, and watch for sentiment spillover into the electronic-chemicals sector.

---

4. Single-stock scoring model (100 points total)

Three scoring examples (for cross-checking)

• 中天科技 Zhongtian Technology, 62 pts (first draft 78 → intermediate draft 70 → final 62, a total cut of 16 points over two rounds) = authority 9 (the first draft gave 14 on the grounds that "the ministry document names submarine cables directly"; after verification, the document's "submarine-cable optical communication" means submarine optical cable, whereas Zhongtian's marine business is mainly power submarine cable — a channel mismatch, so that authority does not apply to it, −5) + directness 12 (the first draft gave 17 for "core business is the beneficiary"; the actually relevant revenue is only about 26%, which on this table's caliber falls in the "supply confirmed but customer unnamed = 11–15" bucket, −5) + elasticity 9 + fundamentals 11 (first draft 13 → intermediate 8 → final 11: the low PB is a product of the fall, and its PE/PB sit in the 81.8%/88.6% percentiles of its own last 3 years, so it is not cheap, but its Q2 ex-non-recurring +59.31%, cash flow turning positive and lowest debt ratio are real credits) + expectation gap 8 + durability 9 + trading attributes 6 + risk deduction 2 (marine backlog −6.6%) = 62

  This example deliberately preserves all three rounds of scores and the reason for each cut, so that the whole falsification process can be audited. The two cuts correspond to two classes of error: the first was "citing a low valuation without asking why it is cheap"; the second was "getting the transmission channel wrong". The latter is more insidious — because the two characters for "submarine cable" appear both in the document and in the company's own statements, and match perfectly on the surface.

• 中际旭创 Innolight, 74 pts = authority 12 (Goldman) + directness 19 (global leader) + elasticity 12 + fundamentals 13 + expectation gap 3 (already +10.38% yesterday) + durability 9 + trading attributes 9 (turnover RMB 37.6 billion, best liquidity) − risk 3 (PB 26.6 + trapped supply from being −35% off the 60-day high) = 74
• 源杰科技 Yuanjie Semiconductor, 38 pts (⚠️ the first draft wrote 48, an addition error: 9+15+8+8+0+6+7 = 53, less 15 gives 38) = authority 9 + directness 15 + elasticity 8 + fundamentals 8 + expectation gap 0 (+16.10% yesterday) + durability 6 + trading attributes 7, risk deduction 15 (PB 67.45, the most expensive in the field + already +16% in a day + STAR ±20% two-way volatility) = 38

  Caliber clarification: the risk item is recorded in this table as a "deduction", and the text always writes "deduct N points" rather than "− risk −N", to avoid ambiguity from a double negative.

---

5. Detailed analysis of the top 10 names

Reading note: the order in which this section is laid out does not indicate rank (the No. 1 recommendation was changed before finalizing; see §9①). Each heading states the name's rank in the §3 overall table directly — go by that label.

中天科技 Zhongtian Technology (600522) | §3 overall rank No. 4 | SSE main ±10% | close 32.72, +3.22%

• ⚠️⚠️⚠️ The most serious self-reversal in this issue: the first draft treated it as the primary beneficiary of "MIIT naming submarine cables", and that used the wrong transmission channel. The MIIT document was issued by the information and communications regulator, and its "submarine-cable optical communication" means submarine optical cable (transoceanic communications). Zhongtian's marine segment is mainly submarine power cable — its 2026 interim report describes delivering ±500kV DC and 500kV AC submarine cables for Yangjiang Sanshan Island, and explicitly states that "500kV ultra-high-voltage-class submarine cable revenue accounts for over 40% of marine-segment revenue". The policy parent document for submarine power cable is the "15th Five-Year Plan for Renewable Energy Development" (offshore wind), not this ICT plan. They are both called "submarine cable", but the benefit channels are entirely separate.
• Three verifiable falsification points:
  1. Share of revenue: the marine series is only 12.09% of Zhongtian's revenue (FY2025), behind power-grid construction at 42.41% and optical communications at 14.04% — Zhongtian is essentially a "grid + optical communications" company, with marine as the third-largest segment. Moreover, the 2026H1 segment information is marked "not applicable" — the company simply did not disclose the H1 submarine-cable share, so any circulating H1 submarine-cable figure is not on the company's caliber.
  2. Orders are falling: on the same primary-data caliber, energy-network backlog went from RMB 30.8 billion (3/31) to RMB 33.0 billion (7/31, +7.1%), but within it the marine series went from RMB 12.1 billion to RMB 11.3 billion (−6.6%). The total pot is growing while marine shrinks; the increment comes from grid/new energy, not marine.
  3. The valuation is not cheap: PE 30.00 is in the 81.8% percentile of the past 3 years, and PB 2.90 in the 88.6% percentile. "PB 2.90, lowest in the field" holds only in cross-section; placed back into its own history, this is the most expensive tenth-or-so of the past three years.
• But this must be said at the same time: its earnings themselves are genuinely good. 2026H1 revenue RMB 30.939 billion (+31.10%), net profit attributable to parent RMB 2.387 billion (+52.29%), ex-non-recurring RMB 2.212 billion (+50.72%); Q2 net profit attributable to parent +56.21%, ex-non-recurring +59.31%, gross margin 16.48% (+1.41pp); operating cash flow turned positive from −RMB 1.856 billion a year earlier to +RMB 148 million; debt ratio 40.82%, the lowest of the three; and it achieved +52% growth while absorbing an RMB 262 million FX-loss swing (finance expense −RMB 143 million → +RMB 119 million). Both receivables and inventory grew slower than revenue, so working capital was not eaten up by the revenue growth.
• Branch stage: not one fiber name hit limit-up yesterday (Zhongtian +3.22%, Hengtong +2.86%, YOFC +5.73%, Yongding +4.92%), so there is no limit-up ladder and no anchor name.
• Fundamentals check: PE (TTM) 30.00, PB 2.90, total market cap RMB 111.672 billion (Tencent qt.gtimg.cn, 9/7 close caliber).
• ⚠️⚠️ A cross-check that overturned my first-draft conclusion (must read): I originally treated "PB 2.90, lowest in the field" as a reason to recommend it — that was wrong. After going through the forward-adjusted daily bars day by day: Zhongtian is −51.8% from its 60-day high (peak 67.84,2026-06-25) and −36.0% cumulatively over 60 days, and it hit limit-down on 2026-07-13. Its low PB is the result of the share price being halved, not undiscovered value. Further company-level negatives found: traditional new-energy business revenue −18.65% year on year; the revised new-type energy-storage project's IRR is 6.22%, below the original project's 9.34%; and Q1 operating net outflow widened (Sohu Finance¹⁶). ⚠️ The above is on a media caliber, not the original text of company filings; this issue tags it "pending primary verification".
• But these negatives are concentrated in the storage/new-energy segment, while the policy positive points to submarine-cable optical communication — different businesses inside the same company. The key variable determining this stock's quality is "what share of revenue is submarine cable + optical communications, and at what gross margin", and this issue did not obtain it. It is the biggest unclosed item in this recommendation.
• Industry standing: one of the first tier in optical fiber & cable and submarine cable (alongside YOFC and Hengtong).
• Technical sentiment: SSE main board ±10%. Yesterday +3.22%, turnover RMB 4.10 billion, turnover rate 3.70%, a mild volume increase with no limit-up; −0.6% over 20 days, the weakest 20-day performance among the five fiber names (YOFC +16.8%, Hengtong +7.3%, Yongding +3.5%, FiberHome +1.8%).
• Final judgment: downgraded from the first draft's "priority deep-dive" to "watch closely". Reason: the policy direction is real, but the pivot of "cheapest valuation" does not survive verification (it is a product of the fall), and it has the weakest 20-day momentum among the five fiber names. It is an odds stock, not a quality stock — if the verified submarine-cable share turns out to be low, it should be removed from the recommendation list outright.

沪电股份 WUS Printed Circuit (002463) | §3 overall rank No. 1 | SZSE main ±10% | close 122.65, limit-up +10.00% (open 113.61 / high 122.65 / low 113.50)

• Related news: computing infrastructure (the MIIT plan's intelligent computing power of 9800EFLOPS); on the 9/7 LHB, net buy of +RMB 1.743 billion, the largest market-wide, including 4 institutional seats, with the net buy equal to 18.13% of the day's total turnover.
• Positive logic: direct. A primary supplier of high-layer-count PCB for AI servers, and one of the fastest links for computing investment to reach A-share financial statements.
• Branch stage: day 1 of the bounce. All 14 limit-ups in the Components sector were first boards; the sector was +6.83% yesterday with 63 up and 0 down — verified, this is the only one of the 90 THS industry sectors with zero decliners. But there is no name on a 2nd board or higher, i.e. this line still has no ladder.
• ⚠️ Relative strength (the main reason this issue ranks it No. 1): WUS is −22.5% from its 60-day high (peak 158.20 @ 2026-07-01), the fourth-shallowest drawdown in the whole optical-communications complex (ahead of it: Kinwong −0.0%, CXMT −5.4%, Cambridge −21.3%), far better than the fiber line's −33%~−55%. The same policy landing on a name that has not been beaten to a pulp and on a name that has halved are not the same thing.
• Fundamentals check (closed out): 2026H1 revenue RMB 13.689 billion (+61.2%), net profit attributable to parent RMB 2.923 billion (+73.7%), ex-non-recurring RMB 2.795 billion (+70.0%); Q2 revenue RMB 7.475 billion (+67.8%), net profit attributable to parent RMB 1.681 billion (+82.7%), gross margin 41.28% — it is the only one of the three verified names whose Q2 year-on-year growth accelerated versus Q1 (revenue +53.9% → +67.8%), with volume and price moving together. PCB is 94.92% of the business, exports 81.58%.
• Two pieces of positive evidence: ① the most diversified customer base (top five 53.32%, largest customer 14.26%), materially better than Innolight's 76.0%; ② the most solid primary capacity evidence — on 2026-02-13 it announced a new high-end PCB project with RMB 3.3 billion of investment and on 2026-03-07 a Kunshan project with RMB 5.5 billion, about RMB 8.8 billion combined, with construction in progress rising from RMB 2.236 billion to RMB 3.571 billion (+60%) and H1 purchases of long-term assets of RMB 3.602 billion (26.3% of revenue).
• ⚠️ Three negatives that must be stated together: ① the tightest balance sheet — debt ratio 52.20% (highest of the three), long-term borrowings from RMB 707 million to RMB 3.984 billion (+463%), OCF/net profit attributable to parent down from 1.01 for full-year 2025 to 0.39; ② PB 13.12 sits in the 92.6 percentile over three years, and this is the only "genuine, undistorted high reading" among the three (Innolight's PB percentile is distorted by the H-share raise and unusable); ③ a tense trap — the interim report's own text says the 800G switch product has "begun sampling" and 448Gbps is "in pre-research"; these are future products, not current revenue, and treating "1.6T switch PCB" as already-realized revenue would be a misreading.
• Industry standing: ⚠️ needs a discount. In the interim report the company describes itself only as "one of the important brands in the industry" and does not claim to be "first" or "the leader". This issue will not upgrade that for it — the earlier phrase "domestic leader in AI server PCB" is a market convention, not company disclosure.
• Technical sentiment: ⚠️this is the most important technical warning in this issue — WUS did hit limit-up, but its seal order was only RMB 179 million, 1.9% of the day's RMB 9.614 billion turnover, and it first sealed the board at 13:31 (afternoon). By contrast, Cambridge Technology's seal order was 17.8% of turnover and it sealed at 09:30. The word "limit-up" here conceals the fact that the seal was extremely weak, and the probability of being knocked open after a gap-up today is materially higher than for names with good seal quality.
• Final judgment: watch closely, do not chase a gap-up. Real institutional money (4 seats) and an extremely weak seal are two signals pointing in opposite directions; the reasonable reading is that institutions are accumulating but are in no hurry to push the board. This pattern suits buying a pullback rather than chasing the auction.

长飞光纤 YOFC (601869) | §3 overall rank No. 3 | SSE main ±10% | close 400.51, +5.73%

• Related news: same as ①, the plan calls to "accelerate R&D and application of new fiber technologies such as hollow-core fiber".
• Positive logic: direct. Public reports say it has achieved hollow-core fiber attenuation of 0.04dB/km and has built out the supply chain (Sina¹⁷), but ⚠️that statement comes from media and self-media; this issue did not obtain company-filing-level primary confirmation, so it is counted as an "industry-trend positive" rather than a "direct order positive".
• Branch stage: day 1 of the launch phase, same as ①.
• Fundamentals check: PE 96.32, PB 20.26, market cap RMB 331.584 billion. Its valuation is already clearly above Zhongtian's (PB 2.90) and Hengtong's (PB 4.68), i.e. hollow-core fiber expectations are partly reflected in YOFC rather than across the whole sector.
• Technical sentiment: +5.73%, turnover RMB 4.91 billion, turnover rate 3.07%, the largest gain among the four fiber names.
• Final judgment: watch closely. The purest logic, but PB 20.3 means the "hollow-core fiber" story is already not cheap in YOFC; if playing the fiber branch, Zhongtian offers better odds than YOFC and YOFC offers better purity than Zhongtian — a genuine trade-off, and this issue does not pretend it does not exist.

中际旭创 Innolight (300308) | §3 overall rank No. 2 | ChiNext ±20% | close 898.46, +10.38% (open 838.61 / high 902.26 / low 836.13)

• Related news: ① Goldman raised its 2028 optical-module TAM to $148.5 billion, +115% versus its prior forecast (Cailian Press, 9/7 21:17); ② Goldman initiated coverage of the H shares at "Buy" at 09:33 on 9/7 with a target of HK$3,267, with 2026/27 net profit forecasts 25%/42% above consensus and a 26–28 revenue CAGR forecast of 78%.
• Positive logic: direct, the global optical-module leader. But the timing must be distinguished: ② happened intraday on 9/7 and was already absorbed by that day's +10.38%; only ① is incremental overnight.
• Branch stage: a continuation within a bounce, not a breakout to new highs. ⚠️ Innolight's current price is still −36.6% from its 60-day high (1416.88,2026-06-22), and +3.9% over 20 days (Tencent forward-adjusted daily bars, 60-trading-day window). Yesterday's surge is a bounce after two and a half months of deep decline, not a new high. That means both that this is not buying at a top and that three consecutive months of trapped supply sit overhead.
• Fundamentals check (closed out): 2026H1 revenue RMB 41.778 billion (+182.5%), net profit attributable to parent RMB 13.651 billion (+241.7%), ex-non-recurring RMB 13.092 billion (+229.3%); Q2 revenue RMB 22.281 billion (+174.6%), net profit attributable to parent RMB 7.917 billion (+228.2%), gross margin 46.42%, and quarterly revenue has now grown sequentially for 10 consecutive quarters. The core business is unambiguous: optical transceiver modules and components at RMB 41.331 billion = 98.93%, and the interim report says "1.6T silicon-photonics modules have entered the volume-ramp stage". Debt ratio 36.08%.
• ⚠️⚠️ PB 26.55 is a distorted value, and this issue draws no conclusion from it. The company's H shares listed on 2026-07-30 (3308.HK, issue price HK$980, net proceeds of about HK$52.891 billion, the greenshoe fully exercised on 8/27, total share count 1.115 billion → 1.178 billion shares). But the latest balance sheet stops at 6/30 — the market cap (numerator) already includes the H shares, while net assets (denominator) do not include the H-share proceeds. On a pro-forma restatement, PB is about 11.5x, not 26.55x. The judgment "Innolight's PB is in the 87th percentile over three years" does not hold until the Q3 report is out.
• Conversely, the PE caliber gives a positive signal: PE (TTM) 51.74 sits in only the 3.8% percentile of the past year — its PE was higher than today's during 96% of the past twelve months, i.e. earnings growth outran the share price over the past year. This is entirely consistent with "−36.6% from the 60-day high while H1 net profit attributable to parent rose +241.7%". 26E forward PE 32.9x and PEG 0.54 are the lowest among the three verified names in this issue.
• ⚠️ But two blemishes must be stated alongside: ① broken cash conversion — H1 operating cash flow of RMB 1.800 billion / net profit attributable to parent of RMB 13.651 billion = 0.13 (1.01 for full-year 2025), and Q2 operating cash flow was −RMB 1.568 billion. In fairness, inventory and receivable turnover days actually improved (184→159 days, 78→65 days), and the swelling absolute amounts mainly reflect working capital tied up by revenue nearly tripling; but the single quarter turning negative is still a fact. ② double concentration — overseas revenue is 94.8% and the top five customers 75.98% (2025 annual report caliber), and the largest supplier is 35.76%, an upstream single-point risk that is often overlooked.
• Industry standing: the company states it has been "first in global optical-interconnect solution revenue for five consecutive years" (citing CIC). ⚠️ This is the company citing a third party; this issue did not independently verify the CIC original, so it is accepted as "company disclosure" and not upgraded to an independent fact.
• Technical sentiment: ChiNext ±20%, so +10.38% is not a limit-up (that requires +20%). It closed at 898.46 against a high of 902.26, only 0.4% below the high — excellent late-session absorption.
• Final judgment: upgraded from the first draft's "watch closely" to "priority deep-dive", and raised to No. 1 in the overall table. The direct reason for the upgrade is that fundamentals verification overturned two of the first draft's assumptions: ① "PB 26.6 is too expensive" — that value is distorted because the H-share proceeds are not in the denominator; pro forma it is about 11.5x; ② "the expectation gap has already been eaten" — its PE is in only the 3.8% percentile of the past year, 26E forward PE is 32.9x, PEG is 0.54, and consensus requires it to "decelerate" in H2 (gap −32.4pp) rather than accelerate, giving it the thickest cushion of the three. It simultaneously has: the strongest already-delivered growth (Q2 revenue +174.6%), the best liquidity (RMB 37.6 billion), and the most direct catalyst (the TAM upgrade is precisely about 1.6T). What remains unresolved is cash flow and customer concentration; those two decide whether it is an overweight or a standard position, not the direction.
• ⚠️ On "Goldman's A-share target price of RMB 2,645": several paraphrases agree, but against yesterday's close of RMB 898.46 it implies +194% upside, inconsistent with the usual magnitude of a 12-month target, so this issue flags it as doubtful and does not use it as a recommendation pivot (see the internal notes).

生益科技 Shengyi Technology (600183) | §3 overall rank No. 5 | SSE main ±10% | close 148.15, +7.75%

• Related news: computing infrastructure; copper-clad laminate (CCL) is the upstream material for AI server PCB.
• Positive logic: direct. High-frequency high-speed CCL sees both volume and unit price rise in 800G/1.6T and AI motherboards.
• Branch stage: in step with PCB, day 1 of the launch phase.
• Fundamentals check: PE 69.28, PB 19.45, market cap RMB 359.857 billion; day's turnover RMB 12.169 billion, net inflow RMB 1.614 billion.
• Industry standing: the domestic CCL leader.
• Technical sentiment: +7.75% and no limit-up (the main board requires +10%), i.e. it has not entered the limit-up ladder, a slightly better position than sector peers that already limited up.
• Final judgment: watch closely. It occupies the "rallied but did not seal" position on the PCB chain, avoiding the poor-seal-quality problem; but PB 19.45 requires sustained earnings verification.

长鑫科技 CXMT (688825) | §3 overall rank No. 6 | STAR ±20% | close 58.47, +6.70%

• Related news: ① the plan's "advanced storage capacity reaching 1700EB"; ② the Xiaomi 18 Fold debuts CXMT LPDDR6 globally (CXMT announced this on 8/29; the product landed on the evening of 9/7); ③ at the 9/7 earnings call, on "cooperation with Apple": "we are exploring cooperation opportunities with quality global customers in an open attitude" and "we already have the capability to compete with mainstream international makers", with no direct confirmation (21jingji⁹).
• Positive logic: storage entering national indicators = medium-term demand certainty; the LPDDR6 debut = product-capability validation.
• ⚠️ Three reasons for the de-weighting that must be spelled out:
  1. The LPDDR6 debut is 8/29 old news; last night was only its delivery, not incremental, and the share price was already +6.70% yesterday;
  2. "Apple cooperation" is a rumor plus an open-ended response and constitutes no usable evidence. By this issue's iron rules that is "unverified", and it must never be written as "entered Apple's supply chain";
  3. Market cap of RMB 3.97 trillion, the largest in the A-share market; a single day of +6.70% has already consumed a great deal of money (net inflow RMB 3.498 billion), and the amount of money needed for further upside does not match its elasticity.
• Fundamentals check: PE 48.52, PB 19.74, day's turnover RMB 20.493 billion, turnover rate 7.86%.
• Final judgment: watch only. The logic is real, the position is poor, the elasticity is weak, and the most eye-catching item (Apple) is exactly the one that does not hold. It is better used as a sentiment weathervane for the memory branch than as a holding.

亨通光电 Hengtong Optic-Electric (600487) | §3 overall rank No. 7 | SSE main ±10% | close 64.00, +2.86%

• Related news: same as ①, the plan names submarine-cable optical communication.
• Positive logic: direct, dual main businesses of submarine cable + optical fiber.
• Fundamentals check: PE 37.70, PB 4.68, market cap RMB 157.849 billion; day's turnover RMB 8.93 billion, turnover rate 5.73%.
• Technical sentiment: +2.86%, the smallest gain among the four fiber names, but turnover of RMB 8.93 billion exceeded Zhongtian's RMB 4.10 billion — rising volume without a price move is disagreement, not neglect.
• Final judgment: watch closely. A large expectation gap (smallest gain) and a moderate valuation (PB 4.68). The volume-up-price-flat pattern needs today's open to confirm its direction and should not be characterized in advance.

深南电路 Shennan Circuits (002916) | §3 overall rank No. 8 | SZSE main ±10% | close 364.98, limit-up

• Related news: computing carrier; LHB net buy +RMB 550 million, 3 institutional seats.
• Technical sentiment: sealed at 10:14, seal order RMB 204 million / turnover RMB 3.314 billion = 6.2%, above average seal quality among the 14 Components limit-ups (better than WUS's 1.9%). Turnover rate of only 1.39% — holders are reluctant to sell.
• Fundamentals check: PE 59.66, PB 13.79.
• Final judgment: watch closely. It is the one name on the PCB chain that combines an early seal, low turnover and institutional participation — better quality than WUS, but more expensive and less elastic.

景旺电子 Kinwong Electronic (603228) | §3 overall rank No. 9 | SSE main ±10% | close 105.97, limit-up +10.00% (prior close 96.34)

• Related news: LHB net buy of +RMB 1.036 billion, 3 institutional seats.
• ⚠️ The first draft wrote here that "LHB net buy of RMB 1.036 billion vs THS net outflow of RMB 169 million — opposite directions" and downgraded it a notch on that basis; that contradiction has been withdrawn. The reason is in §1.6: the inside/outside-bid net figure for a limit-up stock is a caliber artifact (when the limit-up is sealed shut, every executed trade is booked as inside-bid), and it cannot be compared side by side with LHB seat net figures. Kinwong's institutional net buy of RMB 1.036 billion (3 institutional seats) is valid evidence; that "net outflow" is not.
• ⚠️ But it does have one real, unresolved weakness: it broke the board 6 times intraday (first seal 09:58, final seal 10:32), the most erratic sealing process among the PCB big four (WUS broke 7 times, Shennan 3, Fastprint 0). "Standing on a 60-day new high" and "breaking the board 6 times in a day" are two facts that hold simultaneously, and readers must weigh them themselves.
• ⚠️ Technical sentiment (the sole reason this issue added it to the recommendation list): Kinwong is the only name in the entire optical-communications complex standing on its 60-day high — −0.0% from the peak (peak 106.01 @ 2026-08-13, prior close 105.97), +10.4% over 20 days. While the rest of the complex is still in a −22%~−55% hole, only it has fully recovered the June decline. It sealed at 09:58, seal order RMB 177 million / turnover RMB 5.269 billion = 3.4%, turnover rate 5.15%.
• Fundamentals check: PE (TTM) 89.69, PB 7.73, total market cap RMB 106.107 billion (Tencent caliber, 9/7 close). PE 89.7 is higher than WUS (46.6) and Shennan (59.7) on the PCB chain, and lower than Fastprint (288.4).
• Final judgment: watch closely. Its institutional buy amount ranks third on the PCB chain, but the fund calibers conflict, so it is handled one notch down.

剑桥科技 Cambridge Technology (603083) | §3 overall rank No. 10 | SSE main ±10% | close 207.21, limit-up (open 206.00 / high 207.21 / low 202.34)

• Related news: the Goldman TAM upgrade; the company's 1.6T optical modules have completed customer validation, with a mass-production shipment cadence corresponding to 2026 (THS F10¹⁸), and its JDM model locks in core customers.
• Technical sentiment: first sealed at 09:30, seal order RMB 631 million / turnover RMB 3.545 billion = 17.8%, but it broke the board once intraday and only sealed for good at 09:33; that seal-to-turnover ratio ranks 6th among the 22 components + telecom-equipment limit-ups (see the correction below). Its real strength is trend structure: 60-day +13.2%, 20-day +40.1%, and only −21.3% from the 60-day high — all three are first on the optical-module chain, and it is the only name with a positive 60-day cumulative return.
• ⚠️ The first draft wrote here that "net outflow of RMB 1.703 billion constitutes serious counter-evidence" — that criterion has been withdrawn in full, for the reasons in §1.6 (the inside/outside-bid net figure for limit-up stocks is an artifact; inflow RMB 921 million + outflow RMB 2.624 billion = RMB 3.545 billion, exactly equal to turnover, i.e. it is merely a directional split of all trades). The external evidence that replaces it is: Cambridge's H shares (06166.HK) were +24.37% the same day with no price limit — being bid to +24% in a market with no ±10% constraint says more about demand intensity than any inside/outside-bid caliber. It is indeed absent from the 9/7 LHB net-buy list, and that point still stands.
• ⚠️ A factual error in the first draft must also be corrected here: the first draft said it "sealed in the first minute at 09:30, seal order 17.8% of turnover, the only double-digit reading in the field and the best seal quality in the field". After checking the zbc/lbt fields: it broke the board once intraday and only sealed for good at 09:33; and its 17.8% seal ratio ranks 6th among the 22 components + telecom-equipment limit-ups, behind MeiG Smart 112.0%, Gongjin Electronics 26.6%, Huamai Technology 25.9%, Mingpu Optomagnetic 20.9% and Ultrasonic Electronics 20.4%, of which the top four had zero board breaks all day, so their seal quality is better than Cambridge's. There were 8 double-digit names, not 1. The superlatives "best in the field / the only one" are void.
• Fundamentals check: PE (TTM) 162.24, PB 8.15, market cap RMB 76.305 billion. PE 162x requires the 1.6T ramp to be delivered before it can be digested.
• Final judgment: watch closely, and it is the most contested name in this issue. Its price structure and seal quality are among the best in the field, while its fund caliber is among the worst. This issue does not force a reconciliation of that contradiction — today's action is a direct test of which of the two signals is more real.

---

6. Pass list

6.1 Foldable supply chain: primary-evidence grading (verification closed out)

Before finalizing, the original text of filings / annual reports / IR platforms was checked for 5 names and graded into four tiers. The single most important conclusion: among these five, the number of confirmed "Apple foldable suppliers" is currently zero — not one of the five has ever given an affirmative confirmation.

⚠️ Three "cracks between narrative and fact" that must be called out:

1. The market label "Jingyan Technology = the foldable hinge MIM leader" is something the company itself has never said. Its 2025 annual report contains 0 instances of "leader"; it calls itself "one of the leading enterprises" in global MIM, and its risk disclosures state that the industry is "relatively unconcentrated, with a fairly dispersed market share". More importantly, the scale is the other way around: on 2025 MIM revenue, NBTM's RMB 2.540 billion > Jingyan's RMB 1.643 billion, NBTM being 55% larger. By revenue, the "MIM scale leader" crown belongs on NBTM's head.
2. "Jingyan's PE −549 = it is loss-making" is a misreading. Full-year 2025 net profit attributable to parent of RMB 51.318 million was a profit; the negative PE is purely a TTM window effect (two loss quarters, 25Q4 at −RMB 99.41 million and 26Q1 at −RMB 49.72 million, entered the rolling window, summing to −RMB 15.47 million across four quarters). But that is no reason to turn bullish either: H1 impairment provisions of RMB 98.4087 million are 7x its H1 net profit attributable to parent of RMB 13.97 million; operating cash flow of −RMB 368 million worsened further from −RMB 217 million a year earlier; and it received a warning letter on 2026-04-22 with the actual controller planning to reduce holdings.
3. ⚠️ The 9/7 price ranking is almost exactly the inverse of the Q2 fundamentals ranking. That day's moves: Kersen +9.98% > Jingyan +5.01% > Everwin +4.20% > NBTM +3.81% > Triumph +3.07%; whereas on Q2 ex-non-recurring year on year: Triumph +18.42% (the only positive of the five) > Jingyan −41.2% > NBTM −51.5% > Kersen and Everwin, both loss-making for the quarter. The two biggest gainers were one Q2 loss-maker and one with ex-non-recurring at −41%; the only name positive on all three Q2 metrics, Triumph, rose the least.

Resulting handling:

• 东睦股份 NBTM New Materials (600114) — the strongest evidence of the five (the only quantified disclosure) plus the best quality (operating cash flow +RMB 709 million = 3.54x net profit attributable to parent, debt ratio 44.82% and the only one falling, the only interim dividend). The headline net-profit decline of −23.16% must be unpacked: excluding share-based payments it is only −1.35% year on year. ⚠️ But PE 42.10 is "new share count × old profit" (the share issuance for asset purchase completed 8/22, share count 631.4 million → 667.5 million shares, and the target's profit is not yet in the TTM denominator), so it is not a comparable caliber. → Watch closely, the only name on this chain worth following.
• 凯盛科技 Triumph Science & Technology (600552) — UTG is a real business and phase II is already in production, and it is the only one positive on all three Q2 metrics. ⚠️ But two hard warnings: ① the company declines to confirm Apple (its IR answer was "please follow our filings"); ② for its +82.3% surge in June–July, the company's own attribution is TGV rather than UTG, and it explicitly writes that "TGV technology is still in the early R&D stage, has not achieved any commercial mass production, and has not generated any operating revenue". Treating Triumph as a pure UTG name would misjudge the source of its share-price driver. The UTG revenue share was not obtained. → Watch only.
• 精研科技 Jingyan Technology (300709) — as above; the label diverges from the facts, and impairments and cash flow are hard flaws. → Watch only.
• 科森科技 Kersen Technology (603626) and 长盈精密 Everwin Precision (300115) — moved to Pass, reasons in the table above.

---

7. Intra-branch ranking

Branch 1: optical modules / CPO

Hardest name: Innolight (standing + liquidity). Strongest trend: Cambridge Technology (20-day +40.1%, drawdown of only −21.3%, sealed at 09:30 with a 17.8% seal ratio — first in this branch on all four) — and its H shares were +24.37% the same day (no price limit), the strongest external confirmation in this branch. ⚠️ But one first-draft error needs correcting: it broke the board once intraday, and its 17.8% seal ratio ranks 6th, not 1st, within this complex (ahead of it: MeiG Smart 112.0%, Gongjin Electronics 26.6%, Huamai Technology 25.9%, Mingpu Optomagnetic 20.9%, Ultrasonic Electronics 20.4%, of which the top four had zero board breaks all day). Who follows: Taichenguang, Shijia Photons, Taclink (all high-PB, high-turnover catch-up positions). Who is Passed: Yuanjie Semiconductor (right logic, but PB 67 leaves no margin for error).

Branch 2: PCB / copper-clad laminate

Hardest name: WUS Printed Circuit (the largest institutional buy in the field). Strongest relative strength: Kinwong Electronic (the only name in the entire optical-communications complex standing on its 60-day high). Best-quality seal: Shennan Circuits (early seal, low turnover, 6.2% seal ratio). Best position: Shengyi Technology (rallied without sealing, so it carries no board-break risk). Who is Passed: Suntak Technology (a seal order of virtually zero).

PCB is the shallowest-drawdown link in the whole complex (WUS −22.5%, Shengyi −22.8%, Shennan −24.7%, Kinwong −0.0%), while optical fiber is the deepest (−33%~−55%). The same policy catalyst landing on two completely different technical positions is the core reason this issue ranks PCB ahead of optical fiber, and it has nothing to do with "who is cheaper".

Branch 3: optical fiber & cable / hollow-core fiber

⚠️ After primary-source verification, the recommendation logic for this branch has been rewritten wholesale, and that must be stated clearly:

The inference "MIIT names submarine cables → Zhongtian Technology benefits" is wrong, and it is wrong on the transmission channel. This document was issued by the information and communications regulator, and its "submarine-cable optical communication" means submarine optical cable (transoceanic communications); whereas Zhongtian's marine segment is mainly submarine power cable (interim report text: "500kV ultra-high-voltage-class submarine cable revenue accounts for over 40% of marine-segment revenue"), whose policy parent document is the offshore-wind / renewable-energy plan, not this ICT plan. Both are called "submarine cable", but the benefit chains are entirely different.

Re-sorted on primary evidence, the two legs of this branch belong to different names:

• The "submarine-cable optical communication" leg → 亨通光电 Hengtong Optic-Electric (600487). It is the only company that separately discloses marine-communications backlog of about RMB 7.0 billion; it owns the PEACE transoceanic submarine cable system (backlog > $300 million), an Indonesian submarine optical cable base, and full integration capability for submarine branching units and repeaters; and it issues individual contract-win announcements (the 2026-08-27 "Announcement on Winning a Marine Energy Project" at RMB 1.831 billion, listing projects one by one and noting it equals 2.74% of 2025 revenue). Its verifiability is materially better than Zhongtian's (Zhongtian does not issue individual win announcements; its submarine-cable orders appear only as aggregate figures in periodic reports).
• The "hollow-core fiber" leg → 长飞光纤 YOFC (601869). The only hollow-core player with a verifiable amount and shipment volume: on 2026-08-13 it won 100% of China Telecom's Guangdong hollow-core cable and G.654.E framework procurement at RMB 159 million (incl. tax), and it disclosed cumulative hollow-core fiber delivery of over 10,000 fiber-km across 13 commercial and pilot projects. ⚠️ That RMB 159 million is a combined "hollow-core + G.654.E" package, and the pure hollow-core amount is not broken out.
• Zhongtian is not without evidence on hollow-core (its 2025 annual report records delivering hollow-core cable for Sichuan Telecom's quantum communications network, cooperating with Huawei on O-band anti-resonant hollow-core fiber for data centers, and winning pilots with China Mobile Chongqing and a Hong Kong operator), but nowhere in any of its documents does it disclose hollow-core revenue, order value or shipment volume — it has the technology and no revenue evidence.

⚠️ Zhongtian Technology's "backlog" also gives a contrary signal: on the same primary-data caliber, energy-network backlog went from RMB 30.8 billion (3/31) to RMB 33.0 billion (7/31, +7.1%), while within it the marine series went from RMB 12.1 billion to RMB 11.3 billion (−6.6%). The total pot is growing while marine shrinks; the increment comes from grid and new energy. This directly contradicts the "primary submarine-cable beneficiary" narrative, and this issue discloses it head-on rather than avoiding it.

So can Zhongtian still be watched? Yes, but with a different set of logic. Its H1 revenue was +31.10%, net profit attributable to parent +52.29%, Q2 ex-non-recurring +59.31%, operating cash flow turned positive from −RMB 1.856 billion to +RMB 148 million, debt ratio 40.82%, lowest of the three, and it did all of that while absorbing an RMB 262 million FX-loss swing. What should support it is this earnings logic, not the "MIIT names submarine cables" policy logic — using the wrong logic to pick the right stock gets you slapped at the verification point, because this logic's falsifiable points (submarine-cable orders, hollow-core amounts) are exactly its two weakest items.

Another cycle risk that cannot be ignored: Zhongtian's own interim report says the price of conventional G.652.D fiber has "risen more than 400% cumulatively since the start of 2025". YOFC's H1 net profit attributable to parent of +888.88% and gross margin of 53.36% are mainly driven by this fiber price-up cycle, not company-specific alpha; its PB of 20.26 embeds an assumption that prices persist, making it the most exposed of the three to a cyclical drawdown.

Overall character of this branch: not one limit-up, no ladder, no anchor, and the entire line is still in a −33%~−55% hole — either it starts today with the policy, or it keeps grinding lower; "steady" is not one of the options.

---

8. Verification signals at today's open

Auction signals (09:15–09:25)

• ⚠️ Gap-up size for first-tier optical modules (threshold reset per Hong Kong): the first draft's "gap-up >6% = overextension" is void — it rested on the false premise of "zero external confirmation". Given the reference of Innolight H +19.58% and Cambridge H +24.37%, an A-share gap-up of 5%–8% is a reasonable convergence range toward the H shares and does not constitute overextension. New criterion: watch whether it holds after the gap, not how big the gap is. ① Not breaking the opening price within 30 minutes of the gap-up = convergence confirmed; ② a fast fade below yesterday's close after the gap-up = the Hong Kong jump was not accepted by A shares, and the whole line is downgraded; ③ a gap-up of <1% or a gap-down = policy + TAM + Hong Kong, three positives, all rejected, which is the strongest bearish signal.
• Cambridge Technology is the cleanest sample for this criterion: its A shares were truncated at +10.00% while its H shares were +24.37% the same day, implying roughly 14pp of unreleased spread. If it cannot gap up and strengthen today, the "truncated by the price limit" explanation does not hold.
• Whether first-tier PCB (WUS / Shennan / Kinwong) opens one-word limit-up: WUS's seal order was only 1.9% of turnover yesterday. If a one-word board today comes with a seal order above 10% of turnover, yesterday's weak seal has been repaired; if it is knocked open at the open, that verifies the "institutions accumulating but not pushing the board" reading.
• Auction volume on the fiber line (Zhongtian / Hengtong / YOFC): this line had no limit-up and no ladder yesterday, so auction volume is its only launch evidence today. If Zhongtian gaps up >3% with expanding volume, treat it as the policy line broadening; if it opens flat on shrinking volume, money only accepts optical modules and not fiber, and this issue's No. 1 recommendation is thereby falsified.

Sector signals

• 3 or more fast limit-ups: if the Components / Telecom Equipment sectors produce 3 or more limit-ups before 10:00, treat the branch as continuing; ⚠️ note the normalization — ChiNext/STAR require +20% and the BSE +30% to count as a limit-up, and the ±10% intuition must not be applied.
• Anchor volume: if Innolight (turnover RMB 37.6 billion) and WUS (RMB 9.6 billion) hold comparable volume today and close green, the main line holds; rising on shrinking volume is instead a precursor to disagreement.
• Whether a second-board ladder appears: all 22 components + telecom-equipment limit-ups yesterday were first boards. Whether a 2nd board emerges today is the only marker of this line moving from "ignition" to "trend". If fewer than 3 of the 22 reach a 2nd board, yesterday was the sentiment peak.
• Overall performance of yesterday's limit-ups: the average move today of the 93 limit-ups (80 first boards) is a direct read on market-wide sentiment.

Single-stock signals

• Effectiveness test of the top-5 absorption / seal-quality criterion: the degree of divergence today between the four highest seal-to-turnover names yesterday (MeiG Smart 112.0%, Gongjin Electronics 26.6%, Huamai Technology 25.9%, Mingpu Optomagnetic 20.9%, all with zero board breaks) and the lowest few (Suntak Technology 0.15%, WUS Printed Circuit 1.9% with 7 board breaks) directly tests whether the "seal order/turnover ratio + break count" criterion works. ⚠️ Note that MeiG Smart's 112% seal ratio results from being a one-word board (turnover of only RMB 140 million, turnover rate 1.7%), so liquidity is extremely poor and it is untradable, serving only as a criterion sample.
• The three names with money-versus-price divergence — this verification point has been deleted, for the reasons in §1.6: the inside/outside-bid net figure for limit-up stocks is a caliber artifact and constitutes no divergence. It is replaced by the genuinely falsifiable verification point below.
• ✅ Replacement verification point: A/H spread convergence. Cambridge A shares +10.00% (truncated by the limit) vs its H shares +24.37%; Innolight A shares RMB 898.46 vs H shares of about RMB 1,097 (an AH inversion, H at about a 22% premium). If the A shares converge toward the H shares today (gapping up and holding), yesterday's A-share move really was suppressed by the price limit; if the A shares gap up and immediately fade with the spread widening instead, then the Hong Kong jump was Hong Kong's own money behavior and cannot be extrapolated to A shares. This can be judged intraday today.
• Re-checking seal quality (using the zbc break count, not just first-seal time): of yesterday's 93 limit-ups, 57 broke the board intraday. WUS broke 7 times, Kinwong 6, Shennan 3, Cambridge 1, Fastprint 0. Today, watch whether these names' break counts fall — the break count reflects absorption better than the seal time.
• Large-order net inflow: if any single name on the fiber line shows a net inflow of >RMB 1.0 billion, policy money is starting to spill over.

Risk signals

• ⚠️ 中巨芯 Zhongjuxin's (688549) opening action is the first must-watch today — its unlock equals 150.7% of free-float market cap, more extreme than yesterday's limit-down name Moore Threads (+85%). If it opens at limit-down, watch for sentiment spillover into electronic chemicals and unprofitable STAR-listed names.
• ⚠️ Gap up and dive: yesterday had 80 first boards with generally poor seal quality, so if the index gaps up and turns red within half an hour, yesterday's limit-ups will break their boards en masse. On 9/7 there were 19 board breaks (with len(pool)==tc verified on the official pool), a break rate of 19/(93+19) = 17.0% — that figure is healthy in itself (sentiment leaning strong), but it was achieved on the premise that seal orders were generally under 5% of turnover, meaning yesterday was held up by sustained buying rather than seal thickness; once buying weakens today, the break rate will rise non-linearly.
• ⚠️ 2 of the board-break names are in electronic chemicals — stacking with today's Zhongjuxin unlock, that sector is one of today's most fragile sentiment positions.
• ⚠️ A lone limit-up / anchors not following: if only Innolight is strong on optical modules while the second tier all fades, this is money huddling rather than a sector move.
• ⚠️ Tonight's U.S. reopen is this line's real referee: 9/7 was closed for Labor Day, so this line still has no overnight overseas confirmation. If MRVL/COHR/GLW do not follow tonight, A-share optical modules lose external support tomorrow — which means today's position size should not be set as though the trend were already established.
• ⚠️ Moore Threads' unlock selling pressure is not yet digested; if domestic GPU keeps weakening, it will drag on overall sentiment for the computing main line.

---

9. Final recommendation conclusions

① The 5 stocks most worth watching today

⚠️ A defect that must be acknowledged first: this list has only one variable. All 5 are driven by the same catalyst (computing power / optical communications), so they are not 5 independent judgments but 5 expressions of the same judgment. I considered inserting a memory or foldables name to "make it look diversified", but: GigaDevice is −54.5% from its 60-day high and −4.4% over 20 days (worse than Zhongtian), CXMT's catalyst is 8/29 old news and it is already at RMB 3.97 trillion of market cap, and no primary evidence has been obtained on the foldables chain's supply relationships — forcibly inserting a stock with no catalyst to fake diversification is more dangerous than admitting concentration. The correct handling is: control total exposure as "one position", not "five positions", and accept that "if U.S. optical communications do not follow tonight, these five will be wrong together".

Also: this list's primary ranking variable has been changed from the first draft's "valuation level" to "relative strength (distance from the 60-day high) + institutional money". The reason is in §5①: in a sector that has just fallen 33%–55%, a low PB only tells you it fell harder.

The two names moved out of the first draft's recommendation list, and why:

• 中天科技 Zhongtian Technology (600522) — No. 1 in the first draft. After checking the 60-day series it is −51.8% from the peak and −0.6% over 20 days, the weakest on the fiber line; "PB 2.90, the lowest" is the result of being halved, not a value signal, and there are company-level negatives such as the downward revision of the energy-storage project IRR. Reclassified as "watch closely"; no recommendation slot until the submarine-cable revenue share is verified.
• 亨通光电 Hengtong Optic-Electric (600487) — No. 4 in the first draft. Volume up, price flat (turnover RMB 8.93 billion for only +2.86%) leaves the direction undecided, and it is −48.8% from the peak, a drawdown second only to Zhongtian and FiberHome. Reclassified as "watch closely", awaiting today's directional choice.

② The 3 strongest branches today

③ Directions not recommended for chasing today

1. Second-tier optical-communications names up more than +14% yesterday (Yuanjie +16.10%, Taichenguang +14.50%, Taclink +10.32%) — the logic is right, but PB is already at 28–67x, the expectation gap is zero, and STAR/ChiNext ±20% two-way volatility amplifies drawdowns. Right direction does not mean right price.
2. 中巨芯 Zhongjuxin (688549) and recently unlocking STAR names — today's unlock equals 150.7% of free-float market cap, and yesterday Moore Threads hit limit-down at −20% within 4 minutes on the same kind of event with 5 institutional seats net selling RMB 1.308 billion. This is the only negative criterion in this issue with empirical support from yesterday.
3. Pure AI-concept names on long limit-up streaks such as 龙版传媒 Longban Media (605577) — the company's own filing admits the AI video business has generated no operating revenue; as of 9/4 it had 5 consecutive boards with a cumulative +61.14%, and 9/7 was its 6th limit-up (caliber now aligned with §6), with repeated risk warnings issued. It is also the highest consecutive-board name in the field, while its publishing industry has only 3 names in the limit-up pool — the highest board count sitting in an industry with no main-line support is itself a characteristic of a sentiment top.
4. High-level consecutive boards in agriculture with no news driver (Yasheng Group, 3rd board) — zero primary catalysts inside the window, and turnover of only RMB 102 million with extremely poor liquidity.
5. The 6G concept (ZTE, Shenglu Telecommunication) — the plan's wording is "launch 6G commercial use at an appropriate time", the most distant delivery horizon; yesterday ZTE was only +0.64% and Shenglu −1.22%, with money completely rejecting it. Appearing in a policy document ≠ tradable now.
6. Chasing the foldables chain — the events are certain but no primary evidence has been obtained on A-share supply relationships, and Apple's event is in the early hours of 9/10, so today and tomorrow are pure expectation games and the third day faces "sell the news".

④ Final one-sentence judgment

Today's catalyst is "real" (the MIIT five-year plan + Goldman's 115% TAM upgrade, both inside the window and both pointing at the same supply chain), and it has already been voted on once — Hong Kong gave Innolight +19.58% and Cambridge Technology +24.37% in the 9/7 close, and that hour is inside this issue's window and was evidence I missed entirely in the first draft. But the position is "deep" (the whole optical-communications complex topped on 6/22 and has fallen 22%–55%; yesterday was a bounce, not ignition), and the structure is "brittle" (of 93 limit-ups, 80 were first boards, 57 broke the board intraday, and seal orders were generally under 5% of turnover).

So today's operating logic is: follow the direction, but select stocks by "relative strength" rather than "who is cheap" (in a sector that has just halved, the cheapest name is usually the one that fell hardest); the size of the gap-up is no longer a no-go zone in itself (Hong Kong has given a +19.58% reference), and what really needs watching is whether it can hold after the gap.

One last line for the reader as this issue's self-assessment: this draft overturned itself three times before finalizing — the No. 1 recommendation changed hands (Zhongtian Technology → Innolight), a fund criterion used 5 times was withdrawn in full, and the premise of "no external confirmation" was falsified by Hong Kong. None of the three was dirty data; all three were reasoning errors. So please treat this issue's conclusions as "hypotheses with explicit falsification conditions", not as conclusions; every verification point in §8 is a gallows I built for myself, and every one of them can be judged right or wrong intraday today.

---

To be closed out after finalizing (must be resolved in the next issue)

1. Primary supply relationships for the four foldables names (Jingyan 300709 / NBTM 600114 / Triumph 600552 / Kersen 603626): they must be graded on the four tiers of "① explicitly disclosed direct orders ② already in mass supply with the customer unnamed ③ has the product but chain entry unproven ④ pure concept". Because that verification was not closed out, this issue excluded all four from the recommendation list and did not score them on the concept.
2. Quarterly interim-report data for this issue's Top 10 (Q2 revenue / ex-non-recurring / gross margin / operating cash flow) was not closed out name by name; the places marked "not obtained" in §5 need to be filled in. The highest-priority item among them is: Zhongtian Technology's "share of revenue and gross margin of the submarine cable + optical communications businesses" — the policy names submarine cable, while the negatives found (energy-storage IRR revision, traditional new-energy revenue −18.65%) sit in a different business; that share figure decides whether it is a "direct policy beneficiary" or a "diversified company being dragged down", and it is the sole criterion for whether it can return to a recommendation slot. 2b. The three company-level negatives on Zhongtian Technology (energy-storage project IRR 6.22% vs 9.34%, traditional new-energy revenue −18.65%, widening Q1 operating net outflow) currently exist only on a media caliber and must be verified back against company filings / the interim report original text. This issue tagged them "pending primary verification", but it used them to cut the score, so this item must be closed out.
3. The original PDF of the MIIT plan and the list of 26 key tasks / 13 special columns were not obtained; it needs confirming whether optical modules get their own column.
4. The caliber of Goldman's A-share target price of RMB 2,645 for Innolight was not closed out (against yesterday's close of RMB 898.46 it implies about +194%); the original research note must be checked to confirm whether it is a staged target or a paraphrase error.
5. The 9/7 board-break pool was not separately pulled — added before finalizing: 19 board breaks, a 17.0% break rate, now written into §8.
6. The exact release times of August CPI/PPI and financial data are unconfirmed; the next issue needs structured data from the NBS release-calendar page.
7. The catalyst for the 6 general-retail limit-ups (including Baida Group's 3rd board, one-word) was not identified — this issue tagged it "attribution unknown" and fabricated no reason. The next issue must fill it in.
8. The full text of the plan's 26 key tasks / 13 special columns was not obtained. Known paraphrases mention "focus on building 1.6Tb/s optical transmission systems, 200G-PON optical access systems, and hollow-core optical fiber cables" and "accelerate R&D of 100G-and-above high-speed inter-satellite and satellite-ground optical communication equipment". If accurate, then: ① the policy evidence optical modules received is more direct than this issue wrote; ② this issue's use of 50G-PON as the pivot for the fiber branch is weak (the task layer is already talking about 200G-PON); ③ the whole satellite / satellite-ground laser communication branch is absent from this issue. These three points require a re-assessment of branch ranking once the original text is obtained.
9. Zhongtian Technology's 2026H1 marine-segment revenue and gross margin (the interim report's segment information is marked "not applicable", so the company did not disclose it); the reason for the marine backlog decline from RMB 12.1 billion to RMB 11.3 billion (whether revenue recognition accelerated or new order intake slowed — the two readings mean opposite things); and any amount-type data on Zhongtian's hollow-core fiber (currently zero figures). The company will hold its interim results briefing in September, and the Q&A transcript is the only possible official outlet.
10. T&S Communications' 2026Q3 quarterly revenue is the sole adjudication point for its "consensus vs delivered" gap: ≥RMB 2.5 billion = the hockey stick delivers; RMB 1.5–2.0 billion = consensus needs a large downward revision; <RMB 1.5 billion (failing to break the 2025Q2 peak of RMB 1.511 billion) = a 5th consecutive quarter without a new high. Also watch whether inventory can come back down from RMB 929 million.
11. Innolight's Q3 operating cash flow and post-H-share-raise net assets attributable to parent — the former settles earnings quality (whether Q2's −RMB 1.568 billion is an inflection point or growth-phase working-capital use), the latter settles the valuation caliber (what multiple PB 26.55 should be rewritten to). Until then, any judgment based on "Innolight's PB is in the 87th percentile over three years" does not hold.

---

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