A-Share Pre-Market Brief | 2026-08-07 Friday

Data conventions disclosed up front (please read this section first)

1. News time window: 2026-08-06 (Thursday) 15:00 close → 2026-08-07 07:20. Anything outside the window is listed separately and down-weighted in Section 1.2 "Old-News Screening", and is not counted as a new catalyst for today. There are 3 S-grade items falling inside today's window in this brief, of which the only bullish one is Goldman Sachs raising its AI-server PCB/CCL forecasts (the other two S-grade items are bearish: the post-earnings rout in US storage names, and rising Fed rate-hike expectations); the core facts behind the two hottest branches — electronic specialty gases and indium phosphide — all fall outside the window, and each is tagged as such.
2. Market-data convention: throughout this brief, A-share closing prices, percentage moves, turnover rates, and 5-/20-/60-day and year-to-date gains are all as of the 2026-08-06 close, taken from the Tencent Finance single-stock snapshot (stock_zh_a_spot_tx, 5,538 rows). The limit-up pool / broken-limit-up pool / limit-down pool / consecutive-limit-up ladder come from East Money (stock_zt_pool_em etc.). Industry percentage moves and industry net inflows come from the THS (Tonghuashun) industry taxonomy (90 industries). Industry net inflow and single-stock net main-force inflow are two different conventions and must not be added together.
3. Daily price limits differ by board, so every stock in this brief is tagged with its board: SH/SZ main board ±10%, ChiNext (30/301) ±20%, STAR Market (688) ±20%, BSE (codes starting with 8/4) ±30%. On ChiNext / STAR Market, +12% is not a limit-up — do not read a ±20% stock with ±10% intuition. The STAR Market and the BSE additionally have an RMB 500,000 investor-eligibility threshold.
4. Overnight overseas convention: "overnight" in this brief means the US session of 2026-08-06 (Thursday), which closed at 04:00 Beijing time on 2026-08-07. Index and single-stock quotes were measured live from the CNBC quote endpoint (timestamps 2026-08-06 16:00–17:23 ET) and cross-checked item by item against Cailianpress's "Overnight Global Headlines You Need to Know on Friday" (08-07 06:00), with full agreement.
5. Dragon-Tiger List: the 2026-08-06 Dragon-Tiger List has been obtained (East Money convention). "Net buying / net selling" in this brief always means the Dragon-Tiger List convention, which is not the same as whole-day main-force flow; the two must not be mixed.
6. ⚠️ One falsification that must come first (otherwise the whole first branch of this brief will be read wrong): the widely circulated claim that "Japan's Kanto Denka and Central Glass permanently halted tungsten hexafluoride production from 2026-07-01, jointly withdrawing 2,200 tonnes/year and cutting 25% of global high-end capacity" — after checking, this brief judges it "unverified, and directly contradicted by both companies' own public statements", detailed in Section 1.2. This brief does not use that claim as the basis for any recommendation.
7. Items not obtained / left blank in this brief:
   • Single-day northbound net inflow: cannot be provided. Since August 2024 the exchanges have discontinued real-time and post-close disclosure of single-day northbound net buying; this is a disclosure gap, and this brief will not substitute any proxy data for it.
   • The aggregated morning wraps for August 7 (Stock Sea Navigator / Must-Read Before the Open / the front pages of the four securities newspapers): as of press time (07:20) they had not been published or not yet indexed, so this brief does not cite them. This morning's news instead comes from screening the raw Cailianpress / East Money newsflash stream item by item (08-06 15:00 → 08-07 07:08, 200 items in total).
   • The complete August 7 list of new-share subscriptions / lockup expiries: not obtained, so this brief does not include one. The only verified trading notices are halts and resumptions (see Section 1.3).
   • CSSC Specialty Gases' WF6 revenue by product, in amount and share: the company explicitly ticked "breakdown of operating revenue: not applicable" in its interim report, i.e. it actively chose not to disclose. Every "WF6 revenue of RMB X hundred million / share of Y%" in circulation is an outside estimate; this brief neither cites them nor derives its own.
   • The latest WF6 spot quote (July–August): the most recent confirmable data point is 2026-06-11, nearly two months old; this brief treats it as "price momentum stalled" and tags it accordingly.
   • The first-hand publication page for the CCI thermal coal index: behind a paywall; RMB 839/tonne is the cross-check of two independent secondary sources and has not been traced back to the publisher's original text.

⚠️ Data-sourcing and implementation notes for this run (internal only):

• Today's most important lesson: I "independently re-verified" a rumour, and very nearly made it the pivot of the first branch. Before drafting, the main agent ran a dedicated independent search on "Kanto Denka + Central Glass permanently halting WF6 from 7/1, 2,200 tonnes, 25% of global supply", found a large number of Chinese sources corroborating each other, and therefore judged it "confirmed". The sub-agent bypassed Chinese secondary sources and went straight to the two Japanese companies' own websites, and came back with the exact opposite conclusion: ① Central Glass's official IR release of 2026-06-25, "On the supply situation of specialty gases for semiconductors" (《半導体向け特殊ガスの供給状況について》), states verbatim: "お客様からの発注数量に見合う原材料を確保しながら、供給を継続しております" (we are continuing supply while securing raw materials commensurate with customer order volumes) — published 6 days before the alleged "halt on 7/1". ② Kanto Denka at its 2026-06-08 results briefing: "China has tightened dual-use export controls on tungsten, but Kanto Denka is still able to procure raw material… there will be no situation in which we cannot produce"; and to Nikkei on 6/25: "苦労しながら生産に影響が出ないぐらいは調達できている" (with difficulty, we are still procuring enough that production is unaffected). Corroborating evidence: the FY2027/3 guidance the company issued on 2026-05-15 is operating profit +50.9% to JPY 10 billion — a company planning to permanently shut its flagship product line two months later does not guide like that. ③ Reconstructing the propagation chain: The Elec (South Korea, anonymous sources) originally wrote "Japanese suppliers are preparing to cut output in H2, supply disruption possible" → English media rewrote it as "permanent halt from July" → Chinese social media escalated further to "last shipment 6/30, permanent halt 7/1, never to restart". Each retelling upgrades the wording by one level. → A process assertion that must be hardened: for any supply-side fact of the form "some overseas company halts / withdraws / cuts off supply", mutual corroboration among Chinese sources does not constitute verification — they share a single origin. It must be traced to that company's own IR page, earnings release, or a regulatory disclosure in its home country. Had we not checked today, the entire recommendation logic of the first branch would have rested on a rumour the parties themselves deny. This is the closest this brief came to blowing up.
• Second falsification: CSSC Specialty Gases' interim report was not disclosed on 8/6, it was disclosed on 2026-07-18, and the market's first reaction was a 20cm limit-down. The sub-agent cross-checked all 20 announcements from 6/1–8/7 against both the Shanghai Stock Exchange and CNINFO: the company made zero announcements on 8/6. The interim report was disclosed on 7/18 (Saturday); on the next trading day, 7/20, it closed at RMB 200.80, exactly −20.00% from RMB 251.00 on 7/17 (limit-down), and on 7/21 it probed as low as 161.37 intraday. The main reason for the limit-down was operating cash flow of −RMB 241 million, receivables +89.29%, inventory +95.00%. → East Money's 8/6 18:18 piece "Tungsten hexafluoride is hot again" bundles the stale 7/18 financials with the unverified Japanese rumour to produce a post-close attribution for the 8/6 sector move. This is the third appearance of the same failure class as the memory note on "AI limit-up attribution pieces stitching in old announcements", and this time it stitched in both "stale financials" and "a fake rumour". The first draft's line that "the interim report is today's substantive catalyst" was a factual error and has been rewritten throughout.
• The first draft had ranked electronic specialty gases / WF6 as the first branch (S-grade); it has now been cut to third and re-judged. The re-judgement rests on three things holding simultaneously: the supply-side narrative is unverified; the WF6 price has been consolidating at a high level for two months since June (Longzhong convention: only +2.1% month on month in June), while upstream wolframite concentrate halved from RMB 1.05 million/tonne on 3/27 to RMB 400,000 in May, only recovering to RMB 527,500 on 6/10; and the 8/6 limit-up structure is itself a topping signal — CSSC Specialty Gases, the name with the most solid fundamentals, rose only 7.57%, while Zhongjuxin (+19.99%), Do-Fluoride (+10%) and Heyuan Gas (+10%), whose WF6 revenue is zero or near zero, all locked limit-up.
• Another order-of-magnitude error, tenfold (direct translation from an English summary): the first draft wrote Yunnan Germanium's indium phosphide contract as "RMB 5.70 billion – 8.55 billion"; the actual figure is RMB 570 million – 855 million (VAT inclusive). The sub-agent falsified it by back-solving RMB 1.066 billion × 53.48% = RMB 570 million. → Process assertion: whenever a Chinese amount is taken from an English summary at the "billion/million" scale, it must be validated by back-solving from "the percentage of some known base".
• The overnight-lag trap appeared again today, in exactly the same shape as the one recorded on 8/6. Sina's "August 6 close: US stocks mixed, Dow hits another record" (Dow 54349.12, +0.49%) was published at 08-06 04:08 and reports the US close of 8/5 (Wednesday) — an overnight session A-shares had already digested on 8/6. Today's real overnight is the US session of 8/6 (Thursday), with the Dow at 53885.10, −0.85%. The test is the CNBC previous_day_closing chain check: the previous close attached to the 8/6 close is exactly 54349.12, and all three major indices line up. The −0.85%/−0.18%/−0.06% given in Cailianpress's overnight wrap (06:00) match the CNBC measurements exactly, forming a second cross-check.
• Search engines failed systematically at the day boundary again: repeated queries for August 7 content replied directly with "today is August 6, 2026; August 7 has not arrived yet". All of the early-morning news was obtained by pulling the raw newsflash stream with ak.stock_info_global_em() (200 items, covering 08-06 15:00 → 08-07 07:08). Suggest hardening "early-morning news goes through the newsflash stream, not search" into the process.
• Two data channels failed: ① Cailianpress nodeapi/updateTelegraphList returns HTTP 404 (the endpoint has changed); ② this machine's akshare 1.18.79 has no stock_telegraph_cls function. stock_info_global_em works and gives the best quality. Also: in the sub-agent's environment, the akshare/East Money endpoints were blocked by an upstream proxy with 502, so it switched to Tencent quotes plus exchange/CNINFO endpoints — the main agent and the sub-agent do not have the same set of working channels, which is worth remembering.
• Of the two notes on work/spot_20260806.csv, one was my own misjudgement: ① the code column is empty for the entire file (100% mismatch when matching by code) is true, and matching by the name column instead works fine. ② I at one point judged that "the zsz / zljlr columns are zero for the entire file" — that was wrong, it was my own unit error: in that file zsz/ltsz are in RMB 100 million and zljlr/turnover are in RMB 10,000, and I divided by 1e8 as if they were in yuan, so everything displayed as 0.00. The risk-control agent caught this, and the cost was high: it directly caused the first draft to substitute "THS industry net inflow" for single-stock fund flow throughout, which in turn produced the core claim "PCB has the worst chip structure" — a claim the single-stock data refutes (see the next note). → Process assertion: when a field displays as 0, verify the unit assumption before concluding "the data is corrupt". Mistaking usable data for corrupt is more dangerous than missing data — it quietly makes you switch to a substitute measured on a different convention.
• The single biggest judgement error in the first draft was using an industry reading in place of single-stock fund flow to assess chip quality. The first draft turned "the components industry saw net outflow of RMB 3.085 billion on 8/6 (THS industry convention, the largest outflow in the market)" into "the strongest catalyst has landed on the sector with the worst chip structure, and that is today's central contradiction", and let it carry Section 0, 5.4, 8.2 and 9① — four places. The measured East Money single-stock data says exactly the opposite: JCET Group +RMB 2.297 billion (largest in the market), T&S Communications +RMB 1.197 billion, Kinwong Electronic +RMB 1.060 billion, Founder Technology +RMB 975 million, Victory Giant Technology +RMB 890 million; and even yesterday's broken-limit-up names — Jiangnan New Material (+RMB 190 million), Huazheng New Material (+RMB 245 million), Litong Electronic (+RMB 74 million) — were all net inflows. The truth is "severe divergence within the industry, money concentrating into the leaders", not "bad chip structure". → Yesterday's recap 2026-08-06-recap.md:283 already recorded the same lesson: "'the industry rose' does not mean 'the industry was bought'; when picking names in these two industries tomorrow, you must go down to single-stock fund flow and not substitute an industry reading". The first version of this brief violated the rule it had itself written down yesterday, and the root cause was that the unit error above made the single-stock convention "look unusable". Rewritten throughout.
• Second consecutive day of missing the same line: semiconductor packaging and testing. Yesterday's recap :125 recorded "missed the two strongest lines of the day: coal and semiconductor packaging/testing", and proposed the improvement "add a step pre-market: take the TOP20 stocks by prior-session net main-force inflow whose 20-day return is negative, as a separate watch pool". This brief added coal, but missed packaging/testing again: JCET Group (600584) hit limit-up on 8/6 with net main-force inflow of RMB 2.297 billion, the largest in the market, and a 20-day return of −26.7%, and appeared 0 times in the first draft. It has now been added to Section 8.2 as a "flow-driven candidate". Suggest hardening this step into a data-fetch script rather than relying on memory.
• One trap of the form "a concept round-up piece lists the wrong company": several "indium phosphide supply-chain analyses" list Grinm Advanced Materials (600206) as a core name with a "high-purity indium → InP single crystal → substrate closed loop", but the company announced on 2026-06-30 that it "currently has no indium phosphide-related products or technology, and investors are urged to make decisions with caution". It has been removed from the indium phosphide branch. Lesson: the constituent list in a supply-chain round-up must be checked back against company announcements, especially while that stock is going up.
• The coal sub-agent falsified three numbers currently in circulation: "effective thermal coal supply down 1.9% year on year to 3.806 billion tonnes" returned zero hits on dedicated search; the "capacity-reduction window period" has no document from any competent authority; and "H1 average Qinhuangdao price of RMB 767/tonne, +13.2%" appears only in aggregated summaries. All three have been deleted from this brief and replaced with verifiable National Bureau of Statistics data (June raw coal −9.7%, H1 −1.7%). It also ran a consistency test: Liaoning Energy's pre-announcement says in black and white that "coal output and selling prices fell year on year", causing its first-ever loss, yet it hit limit-up on the same day as Beijing Haohua Energy, whose +66% pre-announcement was driven by rising selling prices.
• The tungsten sub-agent pulled the first-hand Federal Register text (91 FR 50701): the BIS restriction takes effect on 2026-08-27, not in September, and is a one-year temporary rule under the DPA plus a 100% domestic-allocation order; grepping the full text for "China" returns 0 occurrences; the China Tungsten Industry Association's own assessment is "limited direct impact on mainland China". Meanwhile the Ganzhou Tungsten Association's August long-term contract guidance price, published 8/5, fell another 8% month on month, and APT fell 15.28% in July alone — the share price is up 30% in 3 days while the commodity is down 60% in 4 months.
• The Ellington Electronics sub-agent falsified the AI narrative in a single stroke using same-period peer comparison: in 2026Q1, WUS Printed Circuit was +62.90%, Shennan Circuits +73.01%, Shengyi Electronics +122.16%, while Ellington's revenue was −1.13% and net profit −67.60%. Add an IRR of only 13.01% on the fundraising project, a customer list that is entirely automotive (Valeo / BYD / Leapmotor), and the fact that none of WUS, Shennan or Shengyi appears among the 20 benchmark companies the company itself selected — "this is the company's own classification of its own tier".
• One convention conflict that must be self-flagged: tungsten concentrate / APT are falling (July −15.28%, August long-term contracts down another 8%), while WF6 is consolidating at highs (6N at RMB 2.2–3.0 million/tonne in June). This is not a contradiction — the upstream move is a pullback after quotas were loosened, while WF6 is an independent tightness at the processing stage. But precisely for that reason, "tungsten prices rising" and "WF6 prices rising" must be written separately; conflating them yields the wrong conclusion that "the whole tungsten chain is booming" (today the opposite is true: tungsten is a Pass, WF6 is downgraded).

---

0. Today in One Sentence

The strongest catalyst is the only S-grade item that falls inside today's window: Goldman Sachs sharply raised its AI-server PCB and CCL forecasts (08-06 20:45) — the 2027 global AI-server PCB market was revised up 38% to USD 37.5 billion, and 2028 to USD 84 billion; CCL for 2027 was revised up 18% to USD 22.1 billion, and 2028 to USD 48 billion; the 2026–2028 CAGRs are 148% and 161% respectively. The increment does not come only from NVIDIA GPU servers — cloud vendors' in-house ASIC servers are the more important source.

And that catalyst has landed on a sector where "the industry reading looks ugly while single-stock money is concentrating into the leaders" — the single most easily misread thing today. On 8/6 the components industry rose 2.97%, with net outflow of RMB 3.085 billion on the THS industry convention (the largest outflow in the market); Jiangnan New Material, Litong Electronic and Huazheng New Material all broke their limit-ups together, and Kexiang and Baoding Technology both issued clarification announcements that evening. But switch to the East Money single-stock convention and the conclusion reverses completely: JCET Group +RMB 2.297 billion (largest in the market), T&S Communications +RMB 1.197 billion, Kinwong Electronic +RMB 1.060 billion, Founder Technology +RMB 975 million, Victory Giant Technology +RMB 890 million, Shennan Circuits +RMB 255 million, WUS Printed Circuit +RMB 160 million — and even yesterday's broken-limit-up names, Jiangnan New Material (+RMB 190 million) and Huazheng New Material (+RMB 245 million), were net inflows.

These two numbers do not contradict each other; they are two conventions describing different sides of the same thing: the industry as a whole is being sold while leading single stocks are being bought heavily = money is rotating out of small and mid caps into the leaders. So today's question for PCB/CCL is not "can bad chips absorb good news", but "will this research note accelerate the process of money concentrating into the leaders". The test is in the single-stock data, not the industry reading.

The most solid branch is [indium phosphide / optical chips], because its earnings are already on the income statement and it rose the least yesterday. Yuanjie Semiconductor (688498) pre-announced H1 net profit attributable to parent up 1196.91%–1304.98% with revenue +339% to +364%, yet rose only 2.70% on 8/6; Yunnan Tin (000960) pre-announced H1 net profit attributable to parent up 38.43%–47.85% and ex-non-recurring up 44.23%–51.91% (ex-non-recurring better than headline), yet rose only 1.68% on 8/6. On the industry side: 6-inch high-end indium phosphide substrates have gone from USD 1,400/wafer at the start of 2025 to over USD 5,000/wafer (+250%), with a global supply-demand gap of >70%, domestic self-sufficiency below 5%, and over 90% of substrate capacity concentrated in three overseas companies.

⚠️ Two popular claims that must be demolished first today (otherwise you will trade off a false premise)

First: "Kanto Denka + Central Glass permanently halted tungsten hexafluoride production on July 1, withdrawing 2,200 tonnes / 25% of global capacity" — unverified, and denied by both companies themselves. Central Glass's official IR release of 2026-06-25 explicitly says it is "continuing supply while securing raw materials commensurate with customer order volumes"; Kanto Denka says it is "still able to procure raw material at present, and there will be no situation in which we cannot produce", and its FY2027/3 guidance is operating profit +50.9%. The original source is South Korea's The Elec reporting "preparing to cut output in H2", escalated through two rounds of retelling into "permanent halt". The only things that actually hold are: China's dual-use export controls on tungsten are real, and overseas raw material is genuinely tight — which is a different matter from "Japan permanently exiting 25% of supply".

Second: "CSSC Specialty Gases' interim report was a new catalyst after the close on August 6" — wrong. The disclosure date was 2026-07-18, and the market first priced it with a 20cm limit-down (−20.00%) on July 20, mainly on operating cash flow of −RMB 241 million, receivables +89.29% and inventory +95.00%. The company made zero announcements on 8/6.

Today you have to look at two fund-flow conventions at once; looking at only one gives you the opposite conclusion. On 8/6, on the THS industry convention, among the top-gaining industries only electronic chemicals (+RMB 1.811 billion), coal (+RMB 1.433 billion) and precious metals (+RMB 348 million) saw net inflows, while components was −RMB 3.085 billion, minor metals −RMB 2.612 billion and semiconductors −RMB 2.221 billion. But the top net inflows on the East Money single-stock convention are: JCET Group +RMB 2.297 billion (largest in the market; semiconductor packaging and testing; limit-up on 8/6, 20-day −26.7%), T&S Communications +RMB 1.197 billion, Kinwong Electronic +RMB 1.060 billion, Founder Technology +RMB 975 million, Victory Giant Technology +RMB 890 million, Do-Fluoride +RMB 736 million, Grinm Advanced Materials +RMB 586 million (this stock has announced it "currently has no indium phosphide-related products or technology" and does not belong to the indium phosphide branch, see Section 6), Advanced Fiber Resources +RMB 571 million.

The same "components/semiconductor" industry shows a large net outflow on the industry convention and large net inflows in leading single stocks — this is not the data fighting itself, it is money concentrating. Yesterday's recap already set the rule on the same phenomenon: "the industry rose" does not mean "the industry was bought"; stock selection must go down to single-stock fund flow and must not substitute an industry reading. Accordingly, this brief has switched the chip-structure judgement in every branch to the single-stock convention. The Dragon-Tiger List is a third convention, and it can point the opposite way from the East Money main-force convention: the 8/6 Dragon-Tiger List shows Xiamen Tungsten net sold RMB 1.139 billion and China Tungsten & Hightech net sold RMB 762 million, while on the same day the East Money main-force convention shows these two at net inflows of +RMB 83 million and +RMB 302 million respectively. The Dragon-Tiger List only counts the top five seats; the two cannot substitute for each other, still less be added together (see Section 2.2).

There was also a rare "wave of clarifications" after the close. Six high-flying stocks — Tongyu Communication, Olai New Material, Bojay Electronics, Baoding Technology, Kexiang and Jiangxi Tungsten Equipment — collectively clarified and denied rumours, and Aile Home (10 consecutive limit-ups) issued a serious-abnormal-volatility risk warning — five of them were at limit-up yesterday. Jiangxi Tungsten Equipment's announcement states outright that "the company's share price may fall rapidly at any time in the future". A post-close wave of clarifications is the most literal form of divergence at highs, and how this batch of names is absorbed at the open today is the market's first sentiment checkpoint.

Driver types: overseas investment-bank forecast upgrades (PCB/CCL) + interim earnings delivery + commodity price rises (compute metals / copper-cobalt) running in parallel, with macro as the headwind. The three major US indices closed lower together overnight (Dow −0.85%, S&P −0.18%, Nasdaq −0.06%), with the storage chain routed after earnings: Western Digital −13%, SanDisk and Kioxia ADRs −6%, SK Hynix −5% (during the Asia-Pacific session both SK Hynix and Kioxia fell more than 10%, and South Korea's KOSPI closed down 4.58%). The rates side needs more caution — CME shows a 55% probability of a 25bp Fed hike in September, Bank of America expects three consecutive hikes starting in September, and Chair Warsh has been reported as "ready to hike". US July non-farm payrolls are also due tonight at 20:30.

Pre-market state assessment: a day of structural divergence, neither a broad rally nor a broad selloff. The Shanghai Composite was +0.57% yesterday while the ChiNext Index was −0.55%, two-market turnover was RMB 2,528.781 billion, 79 stocks hit limit-up and the broken-limit-up rate was 20.2% (29.5% the prior day — sentiment is warming but height is lacking: 57 of the 79 were first-day limit-ups). Today will most likely continue "earnings rise, pure concepts die", with the indices calm and single stocks split between fire and ice. FTSE China A50 futures closed the night session +0.03% at 14938, so foreign pricing is close to neutral.

---

1. News Overview

1.1 New in-window items (2026-08-06 15:00 → 08-07 07:20)

Impact grade definitions: S = changes an industry trend or produces direct order elasticity; A = clearly bullish (or bearish) for a branch with multiple stocks affected; B = bullish but with a long chain that needs verification; C = a sentiment spark with weak persistence.

1.2 Old-news screening and factual corrections (the section of this brief you should read first)

The "core catalysts" behind today's two hottest branches both fail to hold up, or are not new.

⚠️ Addendum: the earnings pre-announcements underpinning the top of this brief's recommendation list are likewise all outside the time window, and must be down-weighted too. The first draft applied this rule only to CSSC Specialty Gases and not to the names at the top of the list, which is a double standard; it is now made consistent:

Note on convention consistency: the "expectation gap" column in this brief is not scored by "whether the news is new", but by "whether the price has fully reflected the news since it was disclosed". This is what the first draft failed to spell out, and why CSSC Specialty Gases (expectation gap 2 pts) and Yuanjie Semiconductor (9 pts) looked as if they were judged on different criteria — in fact both use the same standard: CSSC Specialty Gases was priced with a limit-down after its 7/18 disclosure and has since rebounded to +666% year to date, while Yuanjie Semiconductor was still up only 2.70% on 8/6 after its 7/21 disclosure. ⚠️ One thing must be flagged at the same time: Yuanjie Semiconductor is already +31.1% over the last 5 sessions; "not yet priced" is only relative to its +1197% earnings growth, and does not mean the share price has not moved.

1.3 Confirmed events and trading notices for today (verified)

• Lingda Group (300125, ChiNext ±20%): halted for one day on 8/6; resumes today with its delisting-risk warning and other risk warnings removed, with the short name changing from *ST Lingda to "Lingda Group". The only confirmed "warning removed and resuming trading" name today; its main business is solar cells (Jinzhai Jiayue), and no substantive improvement in fundamentals is evident.
• Unitree Robotics online roadshow 14:00–17:00 (today), subscription on 8/10.
• Others still halted: Biolight (300246), *ST Fazhan (000838), Zhaori Technology (300333), Landun Photoelectron (300862, resumption expected 8/10), ZWZ Bearing B (200706).
• US July non-farm payrolls and unemployment rate tonight at 20:30 — against a backdrop of rising rate-hike expectations this is the week's biggest external variable, and there may be risk-avoidance selling into the close.
• China's July trade balance and July foreign exchange reserves are released today (exact time to be determined).

1.4 Overnight and overseas (convention: US close of 2026-08-06 Thursday, 08-07 04:00 Beijing time)

How to read this (important): the indices closed lower, but the Philadelphia Semiconductor Index bucked the trend at +0.33%, SMH +0.31%, AMD +1.50%, TSMC +1.01% — overnight was "guidance killing storage", not "semiconductors collapsing across the board". Western Digital's FY26Q4 revenue was USD 3.747 billion (+44%) with non-GAAP EPS of USD 3.56 (+109%); SanDisk's revenue was USD 8.965 billion (+372%, a growth rate 8x that of same-chain Western Digital's +44%, an order-of-magnitude outlier that this brief tags as to be verified). Both beat on results; what got killed was guidance (SanDisk's 2027Q1 guidance midpoint of USD 10.55 billion came in below expectations, and Western Digital's next-quarter USD 4.1 billion missed). Their year-to-date gains were about +460% and +200% respectively (as of 8/5) — this is "profit-taking at highs", not "the cycle being falsified". The A-share storage chain has already fallen 40%–55% within 20 trading days (see Section 8.2), so the direction of pressure today is clear, but the downside need not be proportional.

---

2. Strongest Bullish Branches, in Descending Order

2.0 Sector distribution of yesterday's limit-ups (used to check whether the table above missed a direction)

On 8/6 there were 79 limit-ups (East Money limit-up pool convention; another convention counts 83), 20 broken limit-ups and 1 limit-down. The leading industry distribution:

The only gap that needs a dedicated explanation is "communications equipment" — it had the most limit-ups in the market yesterday (8), yet this brief does not list it as a bullish branch, and that is deliberate:

• Xintian Technology (300615, ChiNext) posted 4 consecutive limit-ups and is +107.4% over 5 sessions — the public attribution is a change of control plus a high-premium stake transfer (the new controlling shareholder has a BUPT/Huawei background), which is an event-driven shell-asset story, not industry orders.
• Tongyu Communication (002792, main board) is +42.30% over four sessions — and clarified that evening that there is "no R&D cooperation of any form with NVIDIA".
• The rest — Wuhan Fingu (002194), Hengbao (002104), Chutian Dragon (003040), Guangha Communication (300711), Topscomm (603421) and Huilv Ecology (001267) — show no new in-window industry catalyst.

Assessment: communications equipment had the most limit-ups yesterday but the weakest industrial support — it is driven by a change of control and one rumour that has already been clarified. This brief therefore does not list it as a bullish branch, and puts two of its main names on the Pass list. But leading on count means it was yesterday's sentiment battleground, and if it gaps down as a group today it will drag on the market-wide limit-up count and money-making effect, so it should be watched as a sentiment indicator in Section 8.2.

2.1 Why coal is cut to No. 7 — a consistency test

Yesterday the coal mining and processing industry was +4.88%, the best in the market, with 7 limit-ups. But cross-checking the limit-up list against the disclosed 2026 interim earnings pre-announcements reverses the conclusion entirely:

Addendum (to avoid selective evidence): the complete list of the 7 coal limit-ups on 8/6 is Beijing Haohua Energy, Dayou Energy, Pingdingshan Tianan Coal (601666), Huaibei Mining (600985), Yankuang Energy, Liaoning Energy and Lu'an Environmental Energy (601699). This brief did not obtain the H1 pre-announcements of the three not listed above — Pingdingshan Tianan Coal, Huaibei Mining and Lu'an Environmental Energy — so no judgement is made on them. The table above can only prove that "the limit-up list contains names with severe earnings divergence", not that "all the limit-ups diverge".

But even so, the conclusion stands: if a sector can take a company whose falling selling prices caused its first-ever loss and a company whose rising selling prices produced a +66% pre-announcement, and give both a limit-up on the same day, then in that day's pricing mechanism beta and sentiment carry far more weight than fundamentals.

Three supporting pieces of evidence:

1. The move and the price elasticity differ by two orders of magnitude. Beijing Haohua's 2026 planned output is 17.8 million tonnes; +RMB 5/tonne, even if fully realised, annualises to roughly RMB 89 million of extra pre-tax profit, about RMB 66.75 million after tax = 0.36% of market cap, which cannot explain a 10% limit-up. +RMB 5 against RMB 839 is itself only +0.60%.
2. It is neither a new high nor an oversold bounce. RMB 839/tonne is below the RMB 840 of end-May 2026, and the last three months have oscillated in an RMB 820–860 box. On the long-contract-inclusive convention, BSPI printed 701 on 8/1, −RMB 2 month on month (falling), and CCTD Qinhuangdao printed 725 on 7/31, unchanged. Meanwhile 7 of the limit-up names sit in the 82%–98% percentile of their 12-month price range (Beijing Haohua 97%, Huaibei 98%).
3. The sealing structure is extremely weak. Of the 7 limit-ups, 6 sealed within the 17 minutes between 13:52 and 14:09, with sealing capital generally only RMB 1–4 million. A collective late-session ramp plus tiny sealing orders = heavy absorption pressure the next day.

The medium-term logic is still worth tracking (June raw coal −9.7%, H1 −1.7%, and Emergency Management Ministry Order No. 21 effective 7/1 is a genuine institutional constraint), but if an inflection point exists, it happened at RMB 609 in June 2025, not at RMB 839 on August 5, 2026. Also: CCTD itself forecast on 7/27 that "the thermal coal market in August may be on the weak side".

2.2 Why tungsten goes straight to Pass — the commodity and the share price are running in opposite directions

The core judgement is built on the commodity side, not the fund-flow side. August long-term contract prices down another 8% month on month, July APT −15.28%, and quotas +3.45% year on year (loosening) — these three are verifiable, point the same way, and diverge completely from a 30% share-price gain in 3 days, which is sufficient grounds for a Pass.

⚠️ The fund-flow evidence must be downgraded to "a single convention": the nearly RMB 1.9 billion of Dragon-Tiger List net selling only counts the top five seats, while on the same day the East Money main-force convention shows these two with a combined net inflow of about +RMB 385 million, the opposite sign. This brief therefore does not treat "distribution" as an established conclusion — it is one reading under the Dragon-Tiger List convention, not a verdict on market-wide flows.

2.3 Why electronic specialty gases / WF6 goes from S to A and from branch No. 1 to No. 3 (the largest re-judgement in this brief)

The demand logic is still hard: WF6 is the only commercialised precursor for tungsten fill in 3D NAND word lines and contact holes, stack counts are iterating from 128 layers towards 300-plus, and per-wafer consumption scales close to linearly with layer count; AI chips use roughly 3x as much as ordinary logic chips. The transmission chain has only one link and is very direct.

CSSC Specialty Gases' earnings delivery is also real (back-solving Q2 from the Q1 report):

Q2 net profit attributable to parent of RMB 247 million is close to 72% of full-year 2025 (RMB 345.5 million); whereas the 2023–2025 three-year CAGR of net profit attributable to parent was only 1.4%. The inflection is real, and it happened in April–June.

But the following counter-evidence means it cannot rank first (the second item has been downgraded to "to be verified" on risk-control advice):

1. The supply-side narrative is denied by the parties themselves (see 1.2). The first branch cannot rest on a falsified pivot.
2. The price trend is "to be verified", not "already stalled". The latest verifiable data point is 2026-06-11 (5N at RMB 1,670–1,810/kg; 6N at RMB 2.2–3.0 million/tonne in June, only +2.1% month on month on the Longzhong convention), with no traceable quote for nearly two months since. ⚠️ Two months of missing data can only yield "unknown", not "stalled" — this brief withdraws the first draft's phrase "price momentum has stalled", replacing it with: neither assume it keeps rising, nor assert that it has rolled over. The related verifiable fact is that upstream wolframite concentrate halved from RMB 1.05 million/tonne on 3/27 to RMB 400,000 in May and returned to RMB 527,500 on 6/10, i.e. tightness at the raw-material end has indeed eased.
3. Gross margin barely expanded: 30.35% → 31.79%, only +1.44pp; operating costs +79.34% versus revenue +83.13%. The overwhelming majority of this growth is volume-and-price amplification from "raw material prices rise → pass through to selling prices" — pass-through, not windfall. The real profit elasticity comes from operating leverage (expense ratio 14.60% → 11.10%).
4. Yesterday's limit-up structure is itself a topping signal: CSSC Specialty Gases, with the most solid fundamentals, rose only +7.57%, while Zhongjuxin-U (+19.99%), Do-Fluoride (+10.00%) and Heyuan Gas (+10.00%), whose WF6 revenue is zero or near zero, all sealed limit-up. This is the classic terminal shape of money spreading from earnings names to low-priced concepts.

There is also something the company itself said that must be quoted (verbatim from the 2026-07-28 abnormal-movement announcement):

"The company has never publicly disclosed any information about tungsten hexafluoride product prices." "If the price of upstream tungsten materials for tungsten hexafluoride rises, it may increase the company's raw material procurement costs and compress that product's profit margin." "Do not over-amplify the impact of a single product on the company's results; make rational decisions and be mindful of investment risk."

And one medium-term threat, which the company wrote into its interim report risk factors (p.46): "Advanced-process storage chips are advancing molybdenum-for-tungsten metal deposition schemes, and new precursors are entering customer qualification… which may affect demand in the company's existing cleaning-gas market." — This is the flip side of this morning's "Samsung introduces a molybdenum process in 9th-gen 3D NAND": bullish for molybdenum, a medium-term substitution threat to WF6.

---

3. Overall Single-Stock Catalyst Strength Ranking (descending by stock)

Board tags: Main = SH/SZ main board ±10%; ChiNext = ChiNext ±20%; STAR = STAR Market ±20%. Prices and percentage moves are all on the 2026-08-06 close convention.

---

4. Single-Stock Scoring Model (out of 100)

4.1 Top 6 score breakdown

4.2 A valuation-convention footnote that runs through the whole brief (read before reading any P/E)

Every P/E (TTM) in this brief is on the 2026Q1 reporting-period convention, with a denominator that excludes H1 pre-announced earnings not yet booked. During the interim-report window this systematically overstates P/E. For the names most affected, the "annualised including the H1 pre-announcement" comparison is given alongside:

This changes none of the conclusions (Yuanjie is still a priority deep-dive, Yunnan Germanium is still watch only, Beijing Haohua is still at the 97th percentile), but the first draft did have a "newest numerator paired with the stalest denominator" problem: it used the H1 pre-announcement to give earnings elasticity full marks while deducting risk points using a P/E that excludes that pre-announcement. Disclosed here together to avoid misleading readers. ⚠️ Annualising is linear extrapolation, not forecasting — whether H2 can sustain H1's slope is an unverified assumption for every name above.

4.3 Two rows that need separate explanation

Two rows in this table deserve separate comment today. CSSC Specialty Gases: its "directness of benefit" and "earnings elasticity" are both close to full marks (Q2 net profit attributable to parent +170.5% is one of the hardest single-quarter numbers in the market), but source authority falls from 15 to 6 — because once the supply-side narrative propping up its valuation was denied by the companies involved, all that remains to cite is the company's own announcements; expectation gap 2 pts — the interim report was disclosed on 7/18 and the market already priced it with a limit-down; risk deduction −14 — +666% year to date, P/E 316×, operating cash flow −RMB 241 million. A company's earnings can be real while its stock is un-buyable, and the reason is not in the earnings. Yunnan Germanium: source and directness are close to full marks (a real announcement, a real contract), but fundamentals 2 pts and expectation gap 0 pts — a 2025 ex-non-recurring loss of RMB 6.807 million, three straight years of negative operating cash flow, P/E (TTM) 3119×, and P/B 40.4× at the 99.3rd historical percentile. "News strength" and "holdability" must be scored separately, or the ranking gets hijacked by a single dimension.

---

5. Detailed Analysis of the Top 10 Stocks

51 源杰科技 Yuanjie Semiconductor688498the cleanest earnings delivery, and the smallest gain yesterday · STAR Market ±20%

Related news: 2026-07-21 earnings pre-announcement (voluntary disclosure) — H1 revenue of RMB 900–950 million, +339.13% to +363.53%; net profit attributable to parent of RMB 600–650 million, +1,196.91% to +1,304.98%. Reasons for the growth: higher data-centre revenue, improved gross margin, lower period expense ratio, plus non-recurring gains from fair-value changes on private fund investments.

Bullish logic: direct. AI compute → 800G/1.6T optical modules → upstream optical chips is the tightest link in the whole chain. The company is the only A-share IDM whose main business is exclusively high-speed optical chips, supplying 100G EML in volume into 800G optical modules, with customers including Innolight and Eoptolink. Indium phosphide is the core substrate for its EML chips — it is downstream of Yunnan Germanium, and it is the link on this chain that actually converts the cycle into profit.

Stage of the branch: mid-fermentation. The indium phosphide / optical chip sector's sentiment ignited from 8/4 (that day's Securities Market Weekly piece "Optical chip stocks stir again, indium phosphide materials chased by capital"), with Yunnan Germanium serving as sentiment leader on 3 consecutive limit-ups, but the earnings leader has not yet moved.

Fundamental verification: net profit growth of +1,197% to +1,305% is the highest in the chain. But one thing must be tagged: the pre-announcement explicitly mentions that profit includes non-recurring gains from "fair-value changes on private fund investments", and the exact share coming from operating profit awaits confirmation in the interim report body; this brief tags it as to be verified. This is the largest unverified item in this recommendation.

Industry position: niche leader (high-speed optical chip IDM). Over 90% of global indium phosphide substrate capacity is concentrated in three overseas companies, and domestic self-sufficiency is below 5% — the import-substitution runway is real.

Technicals and sentiment: closed at RMB 1333.00 (only +2.70% on 8/6), 5-day +31.1%, 20-day −23.6%, +201.1% year to date, P/E 466×, turnover only 4.1%. When the indium phosphide branch surged on 8/6 it rose only 2.70%, clearly lagging Yunnan Germanium's limit-up.

Final verdict: priority deep-dive. In a window driven by earnings delivery, it is the combination within its branch with the most real earnings, the smallest same-day gain, and a position in the middle of a 20-day pullback. The risks are a high absolute share price (a four-digit stock, amplifying volatility), a P/E of 466× that is still not cheap, and profit that includes non-recurring items. On this chain, its evidence quality is higher than Yunnan Germanium's, even if the story is less sexy.

---

52 云南锗业 Yunnan Germanium002428the news is real, the stock is dangerous · main board ±10%

This is the name that has to be written up most heavily today, because the gap between "looks strongest" and "is actually riskiest" is largest here.

Related news (corrected): in a 2026-07-23 announcement, controlled subsidiary Yunnan Xinyao Semiconductor signed an indium phosphide wafer (substrate) supply agreement with a customer, with a VAT-inclusive contract value of RMB 570 million – 855 million (53.48%–80.23% of 2025 audited revenue of RMB 1.066 billion), with a performance period of 2026-08-01 to 2027-12-31.

⚠️ Important correction: the widely circulated "RMB 5.7 billion – 8.55 billion" is a billion/亿 mistranslation from an English summary; the actual figure is RMB 570 million – 855 million, a 10x difference. Check: 10.66 × 53.48% = 5.70.

Nature of the contract: volume locked, price not locked — quantities are fixed, with settlement at market price ±20%. The customer name is undisclosed. The announcement's risk warning: "affected by performance progress, cost fluctuations and other factors, the revenue and profit actually contributed by the contract are uncertain".

Fundamental verification (the entire basis for the conclusion):

Metric	2025 annual report	2026Q1
Total operating revenue	RMB 1.0659 billion (+38.89%)	RMB 288.9 million (+20.31%)
Net profit attributable to parent	RMB 20.146 million (−62.06%)	RMB 9.107 million (−10.71%)
Ex-non-recurring net profit	−RMB 6.807 million (a loss)	RMB 4.847 million
Gross margin	21.12% (−7.4pct versus the prior year)	20.29%
Net operating cash flow	−RMB 187.3 million (negative for three straight years, 2023–2025)	+RMB 22.76 million
Gearing ratio	55.31%	55.85%

Table scrolls horizontally

• Finance costs of RMB 59.28 million are 3x that year's net profit attributable to parent (RMB 20.15 million).
• A counter-intuitive but decisive fact: rising germanium prices are a cost to the company, not revenue. Germanium ingot is up about 75% year to date, but the earnings pre-announcement explicitly says "the increase in unit cost exceeded the increase in selling price" — the gross margin on material-grade germanium collapsed from 29.00% in 2024 to 4.32%. Treating "surging germanium prices" as bullish for Yunnan Germanium is a directional misreading.
• Revenue mix: the four germanium product lines total RMB 881.8 million, or 82.7%; compound semiconductor materials (including indium phosphide, gallium arsenide and others, with InP not broken out separately) are RMB 147.3 million, or 13.82%.
• Off-balance-sheet pressure: controlled subsidiary Xinyao has a (contingent) repurchase obligation to minority shareholders of about RMB 504 million as of end-2026Q1, for which RMB 25.64 million of unrecognised financing costs were amortised in 2025.
• Goodwill is zero (clean). Accounts receivable went RMB 241.1 million in 2024 → RMB 352.7 million in 2025 → RMB 407.7 million in 2026Q1, growing faster than revenue.

The only hard bull evidence: the 2026-07-14 pre-announcement — H1 net profit attributable to parent of RMB 55–80 million, +148.31% to +261.18%, ex-non-recurring of RMB 52–67 million, +576.26% to +771.34%, implying Q2 net profit attributable to parent of RMB 45.89–70.89 million, the highest single quarter in its history. The stated driver is "higher demand for high-speed optical modules → both volume and average price of compound semiconductor materials rose". This is genuinely the first substantive profit inflection in a decade, and the driver is indium phosphide rather than germanium prices.

Timing mismatch between capacity and the contract (a hard flaw): existing capacity is 150,000 wafers/year (2–4 inch), of which 6-inch is only 6,000 wafers/year; an expansion started in April 2026 (total investment RMB 189 million, an 18-month build) takes capacity to 450,000 wafers/year once ramped, with full ramp only in H2 2027. But the contract's performance period starts in 2026-08 — the first 12 months can only be delivered from the existing 150,000-wafer capacity.

"The only 6-inch volume producer in the A-share market" needs to be toned down: the company explicitly replied on 2025-08-25 that "neither the company nor Xinyao currently has a specific plan for volume production of 6-inch indium phosphide wafers"; of the 300,000-wafer expansion, only 6,000 wafers are 6-inch (2%). In its public disclosures the company never calls itself a "leader" or "the only one".

Valuation (own calculation, 2026-08-06 close of RMB 90.98): total share capital of 653.12 million shares → market cap of about RMB 59.42 billion; P/E (TTM) about 3,119×; P/B about 40.4×; P/S about 55.7×. At the 99.26th historical percentile. The all-time closing high was RMB 130.10 (2026-06-25), so it is currently 30% below the peak; +186.6% year to date (Tencent snapshot zdf_y convention).

Technicals and sentiment: 3 consecutive limit-ups (8/4–8/6), with sealing capital of RMB 585 million, the largest in the market, and 8/6 Dragon-Tiger List net buying of RMB 1.154 billion, the largest in the market; but on the same day the East Money main-force convention shows net inflow of only +RMB 286 million (about 4% of its RMB 7.063 billion of turnover), not even top ten in the market. The two conventions differ by 4x, and the former only counts the top five seats — citing RMB 1.154 billion alone would overstate how much money endorses this stock. The shareholder count went from 141,200 in April to 289,200 in July (doubling in three months), and with the whole float unlocked there is no lockup cushion.

Attribution note: attributing the 3 consecutive limit-ups to the 7/23 contract does not work chronologically (the stock rose only +1.84% on the announcement day and fell 19% over the following week). The real ignition point was sentiment in the indium phosphide / optical chip sector on 8/4.

Final verdict: watch only. The news is real, the Q2 inflection is real, the money is real; but a RMB 59.4 billion market cap sits against a 2025 ex-non-recurring loss, three straight years of negative operating cash flow, a 40x P/B and the 99.3rd historical percentile. Even if the contract is delivered in full at the most optimistic high margin, the annualised contribution to net profit attributable to parent is only on the order of RMB 70–130 million. Current pricing reflects the long-dated narrative of "indium phosphide growing from 13.8% of revenue into the main business", not any realised earnings base.

---

53 锡业股份 Yunnan Tin000960the most comfortable position in the market · main board ±10%

Related news: ① the "compute metals" trade heating up (8/7 01:03, Securities Daily): tin +40% and indium +60% in H1; the company has the world's largest primary indium reserves and is China's only fully integrated tin value chain platform, and is a core raw-material supplier to domestic indium phosphide substrate makers. ② The 2026-07-15 earnings pre-announcement: H1 net profit attributable to parent of RMB 1.47–1.57 billion, +38.43% to +47.85%; ex-non-recurring of RMB 1.88–1.98 billion, +44.23% to +51.91%.

This pre-announcement has one easily overlooked highlight: ex-non-recurring net profit is higher than headline, and ex-non-recurring is growing faster than headline — which says operating profit quality is better than it appears, the opposite shape from pre-announcements propped up by investment income or asset disposals.

Bullish logic: indium is the upstream raw material for indium phosphide substrates, and tin is the core metal in AI-server solder. The company sits on both the "compute metals" and the "indium phosphide upstream" main lines at once, and on the resource end rather than the concept end.

Technicals and sentiment: closed at RMB 39.25, only +1.68% on 8/6, 5-day +24.8%, 20-day +5.3% (a rare positive on this list), 60-day −2.8%, +42.6% year to date, P/E (TTM) only 27.7×, P/B 3.18×, turnover 5.6%.

Final verdict: priority deep-dive. In a list full of "P/E in the hundreds, up six-fold year to date, down thirty percent over 20 days", it is the only name that simultaneously has a real earnings pre-announcement, good ex-non-recurring quality, a valuation under 30x, a 20-day gain that is positive but not extreme, and a position at the intersection of two main lines. The drawbacks are less elasticity than pure theme stocks and ever-present commodity cycle risk.

---

54 沪电股份 WUS Printed Circuit002463生益电子 Shengyi Electronics (688183, STAR Market ±20%) / 深南电路 Shennan Circuits (002916, main board ±10%) — the genuine beneficiaries of the only in-window S-grade bullish catalyst · main board ±10%

Related news: Goldman Sachs sharply raised its AI-server PCB/CCL forecasts — the 2027 global AI-server PCB market to USD 37.5 billion (raised 38% from the prior forecast) and 2028 to USD 84 billion; CCL 2027 to USD 22.1 billion (raised 18%) and 2028 to USD 48 billion; 2026–2028 CAGRs of 148% and 161% respectively; and the 2026 global AI PCB market raised 35% to USD 13.56 billion. Growth comes not only from server shipments but also from higher layer counts, rising HDI penetration, CCL material upgrades and greater PCB content per rack; demand from cloud vendors' in-house ASIC servers is a more important source of increment than NVIDIA GPUs.

Why these three rather than the batch that hit limit-up yesterday: in 2026Q1 — WUS Printed Circuit revenue +53.91%, net profit +62.90%; Shennan Circuits revenue +37.90%, net profit +73.01%; Shengyi Electronics revenue +52.62%, net profit +122.16%. These three are the companies that have actually put the AI PCB cycle onto their income statements. The batch that hit or broke limit-up in the PCB/components sector yesterday (see Section 6) is either issuing clarifications or contradicted by its own financials.

Position and flows (both conventions must be shown side by side):

Stock	8/6 change	Net main-force inflow (East Money convention)	Last 20 days	P/E (TTM)	Market cap
沪电股份 WUS Printed Circuit (002463)	+1.38%	+RMB 160 million	−14.6%	52.4×	RMB 225.5 billion
生益电子 Shengyi Electronics (688183)	+0.62%	+RMB 65 million	−14.3%	52.8×	RMB 90.6 billion
深南电路 Shennan Circuits (002916)	+1.38%	+RMB 255 million	−20.0%	66.3×	RMB 241.1 billion

Table scrolls horizontally

All three are the combination of "a small gain on the day + net main-force inflow + a 14%–20% 20-day pullback + a P/E of 50–66x" — neither position nor valuation is expensive, and money is already flowing in.

An important correction on "chip structure": on 8/6 the components industry was +2.97% while the THS industry convention showed net outflow of RMB 3.085 billion (the largest in the market), and Jiangnan New Material, Litong Electronic and Huazheng New Material all broke their limit-ups — from which the first draft concluded "the worst chip structure in the market". But the East Money single-stock convention says exactly the opposite: JCET Group +RMB 2.297 billion (largest in the market), T&S Communications +RMB 1.197 billion, Kinwong Electronic +RMB 1.060 billion, Founder Technology +RMB 975 million, Victory Giant Technology +RMB 890 million, and even the broken-limit-up names Jiangnan New Material (+RMB 190 million), Huazheng New Material (+RMB 245 million) and Litong Electronic (+RMB 74 million) were all net inflows.

The correct description is "severe divergence within the industry, money concentrating into the leaders", not "bad chip structure". The flip side of the industry net outflow is that a large number of small and mid caps are being sold to buy leaders. For the Goldman catalyst that is a tailwind, not a headwind.

Final verdict: watch closely. The catalyst is today's only in-window S-grade bullish item, single-stock money is already flowing in, and the position is mid-way through a 20-day pullback. How to watch: see who leads. If real-earnings names like WUS / Shengyi / Shennan lead and yesterday's clarification names gap down, money is continuing to concentrate into leaders, which is healthy; if instead it is yesterday's clarification names and broken-limit-up high-priced stocks that surge again, that is the terminal stage of the spread.

---

55 中船特气 CSSC Specialty Gases688146the hardest earnings, and the last thing you should chase today · STAR Market ±20%

Related news: the 2026 interim report (disclosure date 2026-07-18, not 8/6; the company made zero announcements on 8/6): revenue of RMB 1.9043 billion, +83.13%, net profit attributable to parent of RMB 347.9 million, +95.63%, ex-non-recurring of RMB 326.4 million, +117.08%; tungsten hexafluoride revenue in the reporting period grew nearly 3x year on year (the company's own words, with no absolute amount disclosed — the interim report explicitly ticks "breakdown of operating revenue: not applicable").

The earnings are real, and better than they look on the surface: Q2 revenue of RMB 1.203 billion (+129.5%), net profit attributable to parent of RMB 247 million (+170.5%), net margin 20.50%; ex-non-recurring / headline = 94% (only 69% for full-year 2025); a gearing ratio of just 21.42%, no short-term borrowings, and net cash of about RMB 1.983 billion. Q2 net profit attributable to parent alone is close to 72% of full-year 2025, while the 2023–2025 three-year CAGR was only 1.4%.

But five counter-facts dictate "watch only":

1. The market has already priced it, and it did so with a limit-down. The interim report was disclosed on 7/18 (Saturday), and on the next trading day 7/20 it closed at RMB 200.80, exactly −20.00% from RMB 251.00 on 7/17 (a 20cm limit-down), with a further intraday probe to 161.37 on 7/21.
2. The reason for that limit-down still applies today: operating cash flow of −RMB 240.7 million (versus +RMB 333.5 million in the year-earlier period, −172.19% year on year), with net-profit-to-cash conversion of −0.69; accounts receivable of RMB 950 million (+89.29%, 49.9% of half-year revenue), inventory of RMB 657 million (+95.00%), and prepayments +346.53%. The company explains this as "stockpiling plus rising raw material prices led to large cash payments for procurement, and receivables within credit terms increased". This is the classic "stockpile expensive raw material + extend credit terms to grab share" pattern of a price-increase cycle; if WF6 prices peak and roll over, that expensive inventory becomes impairment risk.
3. The supply narrative underpinning the valuation is denied by the parties involved (see 1.2, 2.3).
4. Extreme valuation + a tiny float: market cap of RMB 163.06 billion, P/E (TTM) 316.3×, P/B 27.1×, P/S (2025A) 72.2×; +840.3% over 12 months and +666.1% year to date; the free float is only 27.4%. Even annualising the most optimistic Q2 run rate (RMB 986 million), the P/E is still 165×.
5. The company itself keeps cooling things down: it published 6 abnormal-movement / risk-warning announcements within 2 months, and halted for verification on 6/23–6/25. Verbatim from the 7/28 announcement: "do not over-amplify the impact of a single product on the company's results".

Industry position is beyond dispute: on the Linx Consulting convention, its 2025 integrated-circuit electronic specialty gas sales revenue ranked ninth globally and first in China; WF6 capacity is 2,000 tonnes/year and entirely 6N grade, with customers including TSMC, Micron, SK Hynix, Infineon, Kioxia, GlobalFoundries, SMIC, YMTC, ChangXin Memory, Hua Hong and CR Micro. Note: Samsung does not appear on the customer list.

Medium-term threat (the company's own words, interim report p.46): "advanced-process storage chips are advancing molybdenum-for-tungsten metal deposition schemes" — which corroborates this morning's "Samsung introduces a molybdenum process in 9th-gen 3D NAND".

Final verdict: watch only. Its fundamentals are among the hardest in the market, but today it has no new catalyst, its position is on a +666% year-to-date plateau, its cash flow is negative, and the supply narrative that carried the imagination has been falsified. If it gaps up with the sector today, the risk-reward is not symmetric — no new catalyst and an extremely high position.

---

56 洛阳钼业 CMOC Group603993how it really benefits from the ban is the opposite of the intuition · main board ±10%

Related news: an administrative order from the DR Congo (signed 6/29) surfaced on 8/6 banning copper and cobalt concentrate exports outright. The company's response: its local products are cathode copper and cobalt hydroxide, which are not on the banned list.

The logic must be graded (the crux of this entry): the ban is not a company-event benefit — the company itself says it is unaffected. The way it benefits is indirect but sturdier — the ban hits the model of "ship concentrate out and smelt it elsewhere", whereas CMOC is integrated across mining and smelting and turns concentrate into cathode copper / cobalt hydroxide locally. Competitors are constrained, it is not, and at the same time product prices rise on the supply contraction — three effects stacking.

Magnitude: the shock to cobalt is far larger than to copper — DR Congo copper concentrate exports are about 370,000 tonnes of copper equivalent, roughly 1.3% of global supply, while cobalt export quotas have already been compressed to about 52% of 2024 output.

Technicals and sentiment: closed at RMB 20.37, −0.49% on 8/6 (it did not rise), 20-day +18.5%, +3.3% year to date, P/E (TTM) 18.0×.

Final verdict: watch closely. A textbook case of "grading the logic" today: it looks bearish (an export ban) but is actually bullish (integration + prices), and it did not rise on 8/6 and is not expensively valued. The risk is that the enforcement intensity and duration of the ban are unknown.

---

57 华友钴业 Huayou Cobalt603799the most elastic and lowest-positioned name under the same ban · main board ±10%

Related news: the same DR Congo cobalt concentrate ban; the company has a deep footprint in the DR Congo, and with new Indonesian projects coming onstream in 2026 its own annual cobalt capacity may exceed 30,000 tonnes.

Technicals and sentiment: closed at RMB 41.40, −1.48% on 8/6, 20-day −1.9%, 60-day −33.5%, −38.9% year to date, P/E (TTM) 10.7×, P/B 1.58×.

Final verdict: watch closely. The lowest-positioned main-line name on the whole list — down nearly forty percent year to date, on a P/E below 11. If cobalt prices genuinely rise on the ban, the profit elasticity is significant. The risks are that cobalt prices have been extremely volatile historically and that the company was already badly hurt by the earlier cobalt price slump — a "high odds, win rate to be proven" type.

---

58 昊华能源 Beijing Haohua Energy601101the only coal name that stands up · main board ±10%

Related news: the 2026-07-14 earnings pre-announcement, with H1 net profit attributable to parent of RMB 742.55 million, +65.68%, and ex-non-recurring of RMB 745.27 million, +65.77% (ex-non-recurring slightly above headline, with no reliance on one-off gains). Reasons: higher year-on-year selling prices for coal and methanol; commercial coal sales volume up 390,000 tonnes year on year.

A genuine expectation gap: the company's 2025 annual report disclosed a 2026 operating target of RMB 583 million of net profit attributable to parent. H1 alone, at RMB 742.55 million, is 127% of the full-year target.

Fundamental verification: 2026Q1 revenue of RMB 2.605 billion, net profit attributable to parent of RMB 319.8 million (+15.6%), gross margin 32.56%, gearing 47.26%, and operating cash flow of RMB 892 million / net profit attributable to parent = 2.79× (flawless conversion of profit into cash).

Three counter-facts: ① partly priced already — from the pre-announcement close of RMB 10.75 on 7/13 to the RMB 12.72 close on 8/6, +18.3%. ② A very high position — the 12-month price percentile is 97%, with P/E (TTM) of 32.96× (Q1 convention; about 12–14× once the H1 pre-announcement is annualised in, see 4.2). ③ The "high-dividend coal" label no longer holds — the dividend itself was halved: RMB 2.50 per 10 shares for FY2025, down 45% from RMB 4.58 per 10 shares for FY2024.

⚠️ The dividend-yield convention must be spelled out: computed at the same current price of RMB 12.72, the FY2024 dividend corresponds to 3.29% and FY2025 to 1.97% (against a 10-year government bond at 1.71%, a premium of just 26bp). The first draft's "down from 6.02% to 1.97%" used two different share prices — the 6.02% used the then-low price, so that decline mixes in a +67% share-price gain and exaggerates the "collapse in dividend yield". The truth is: the dividend was halved by 45%, and together with the share-price rise that pushed the yield down to 1.97%.

China Shenhua at 4.57% and Shaanxi Coal at 3.81% still have a cushion, and these two rose the least on 8/6 (+2.26% / +4.14%).

One more piece of background that should be stated: a P/E (TTM) of 32.96× implies full-year 2025 net profit attributable to parent of only about RMB 261 million, while 2025H1 alone was RMB 448 million — i.e. 2025H2 was roughly a loss. That is exactly why "the 2026 operating target was set at only RMB 583 million", and it is the source of the denominator against which H1 delivered 127%. The low base is a fact, and the real growth is also a fact; the two have to be read together.

Technicals and sentiment: limit-up, first sealed at 09:44, the earliest in the sector, but with sealing capital of only RMB 4 million (extremely weak).

Final verdict: watch closely. It is the only coal name whose rise is justified, but the sector as a whole is sentiment beta; and it already sealed limit-up yesterday, with the current price at the 97th percentile of its 12-month range and sealing capital of just RMB 4 million.

---

59 金钼股份 Jinduicheng Molybdenum601958today's newest and least-priced narrative · main board ±10%

Related news: Cailianpress 8/7 06:57, "Research Picks · research-note data": Samsung Electronics has introduced a molybdenum metallization process in its ninth-generation 3D NAND, and the molybdenum supply-demand gap is expected to keep widening.

Logic: it shares a root with WF6 — both are material increments arising from rising 3D NAND stack counts. Molybdenum replacing tungsten in word-line metallization can lower resistance and support higher layer counts. If the roadmap is established, this is the second "storage capex → demand for a specific metal" transmission chain after WF6; and for WF6 it is a substitution threat (CSSC Specialty Gases has already written this into its interim report risk factors).

The strength of the evidence must be labelled honestly, and here there are two gaps, not one:

1. First layer: this is research-note sourcing, not a Samsung announcement or industry data; this brief obtained no first-hand confirmation from Samsung, and tags it as to be verified.
2. Second layer (more important, and missed in the first draft): even if the molybdenum roadmap holds, what Samsung is introducing is semiconductor-grade molybdenum precursors/targets (high-purity compounds of the MoO₂Cl₂ / MoCl₅ type), whereas Jinduicheng Molybdenum's main business is molybdenum concentrate, ferromolybdenum and molybdenum powder, aimed mainly at steel and metallurgy. Whether the company has semiconductor-grade molybdenum products, and whether it is in the relevant supply chain, is something this brief obtained no first-hand confirmation of.

Under this brief's rules it therefore counts as a second-order mapping (the industry trend may be bullish → but whether this company can capture it is unverified), not a "direct beneficiary". Accordingly the catalyst grade has been cut from B+ to B and the total score from 60 to 54.

Final verdict: watch closely. The narrative is the newest (published this morning) and has not yet fermented (no unusual move on 8/6), but its evidence tier is the lowest on the whole list — it lacks both first-hand factual confirmation and confirmation of the company's qualifications. Suitable only as a direction to watch, not to be treated as established logic.

---

510 光库科技 Advanced Fiber Resources300620the second tier of the indium phosphide branch · ChiNext ±20%

Related news: the indium phosphide / optical device cycle; H1 net profit of RMB 140–150 million. Its main business is fibre-optic devices and lithium niobate modulators, benefiting from the ramp of 800G/1.6T optical modules and CPO.

Technicals and sentiment: closed at RMB 275.67, +5.63% on 8/6, 5-day +44.2%, 20-day −12.6%, +87.7% year to date, P/E 326×.

Final verdict: watch closely. Within the indium phosphide / optical chip branch its earnings certainty is below Yuanjie Semiconductor's and its position is slightly higher than Yunnan Tin's, making it the second tier within the branch rather than the first choice. Being +44.2% over 5 sessions says money has already come in, so the risk-reward of chasing it is mediocre.

---

6. Pass List

6.1 Appendix: why China Tungsten & Hightech is a Pass (supporting evidence)

2026Q1 revenue of RMB 7.007 billion (+106.47%) and net profit attributable to parent of RMB 921 million (+264.44%), with an H1 pre-announcement of RMB 1.97–2.17 billion (+261% to +298%) — the financials are extremely pretty. But:

• Operating cash flow of −RMB 2.136 billion (against net profit attributable to parent of +RMB 921 million in the same period), i.e. profit-to-cash conversion of −232%
• Inventory from RMB 4.891 billion → RMB 8.326 billion (+70%), equal to 29.2% of total assets and 78% of net assets attributable to parent
• Prepayments from RMB 358 million → RMB 2.022 billion (5.6x); short-term borrowings from RMB 1.995 billion → RMB 3.889 billion
• The gearing ratio rose from 50.78% to 59.38% in a single quarter
• P/B 13.33×, P/E (TTM) 73.1×; −22.3% over the last month, from a peak of 108.10 (6/22) → an 8/3 low of 47.99 = −55.6%

This is the balance sheet of "using leverage to stockpile raw material and earn the spread": superb in an up-cycle, excruciating in a down-cycle. And August long-term contract prices have already fallen another 8% month on month — the profit is being delivered on the income statement at exactly the moment prices turn.

6.2 Appendix: why Ellington Electronics is a Pass (the most valuable falsification in this brief)

2026Q1 was the fiercest quarter of the AI compute PCB cycle. If Ellington had any substantive AI-server exposure, its Q1 results could not possibly look like this.

Four supporting points:

1. H1 earnings pre-announcement (7/15): net profit attributable to parent of RMB 92.5 million – 118.5 million, −54.54% to −64.51% year on year.
2. The fundraising project's IRR is only 13.01% with a payback period of 8.51 years — a genuine AI high-speed board project does not produce that return (compare WUS / Shengyi Electronics at ROE 28–30%). The project is expected to break ground in 2026-12 with a 21-month build, contributing ≈ 0 to 2026/2027 profit.
3. The customers are all automotive: the announcement names Valeo, BYD, Leapmotor and Yanfeng, with not a single server / cloud vendor / ODM. The full set of announcements never mentions any M6–M9 material grade or layer-count metric.
4. The company's own classification says the most: the 20 benchmark companies chosen in its equity incentive draft are uniformly low- to mid-end / automotive board makers, and not one of WUS, Shennan or Shengyi Electronics is on the list; moreover the vesting conditions set the 2027 ex-non-recurring target at about RMB 337 million — 24.5% below the RMB 447 million actually earned in 2025.

The controlling shareholder's subscription is more likely about consolidating control than confidence in earnings: Jiuzhou Group holds 30.00% and Ellington Investment 28.98%, a gap of only 1.02pct; subscribing RMB 1 billion would widen it to 10.8pct, while subscribing only RMB 500 million would instead dilute it to 28.1%. And Jiuzhou will not participate in the bidding, accepts market pricing, and will not subscribe at all if no one else bids — providing no price guarantee whatsoever for the placement's success.

6.3 Appendix: grading "genuine beneficiaries vs pure concepts" in the WF6 branch (the table that most needs to be clear today)

The most jarring thing about this table: the size of the gain is perfectly inversely correlated with relevance — CSSC Specialty Gases, the most relevant, rose the least, while the three with zero relevance all hit limit-up. That is the direct tape-based evidence for this brief cutting the branch from S to A and from No. 1 to No. 3.

---

7. Intra-Branch Rankings

A note covering all three branch tables: the rankings below are by industrial relevance / hardness of evidence, and are not the same as attention priority. For attention priority see the overall ranking in Section 3 and Section 9. It is possible to rank first within a branch while the conclusion is "watch only" (e.g. CSSC Specialty Gases in 7.3) — please do not read "first in the branch" as "today's top pick".

7.1 Indium phosphide / optical chips (first branch)

Hardest single stock: Yuanjie Semiconductor. Highest trading recognisability: Yunnan Germanium (but the worst fundamentals — precisely the core mismatch in this branch). Removed: Grinm Advanced Materials.

7.2 AI-server PCB / CCL (second branch)

Hardest single stock: Shengyi Electronics. Who is following: Jiangnan New Material, Huazheng New Material, Litong Electronic. Who is a Pass: Kexiang, Baoding Technology, Ellington Electronics.

7.3 Electronic specialty gases / WF6 (third branch)

Hardest single stock: CSSC Specialty Gases (the only one). For the other five, on today's tape the size of the gain is perfectly inversely correlated with fundamental relevance — the classic signature of a branch topping out.

---

8. Verification Signals at Today's Open

8.1 Auction signals (09:15–09:25)

• First checkpoint: the auction in the six names from yesterday's "wave of clarifications". Tongyu Communication (002792), Olai New Material (688530), Bojay Electronics (002975), Baoding Technology (002552), Kexiang (300903), Jiangxi Tungsten Equipment (600397) — if they gap down more than 3% as a group with shrinking auction volume, money has accepted it and risk at highs is being released in an orderly way, which is good for the whole market; if instead they gap up or even lock limit-up on the open, hot money is still fighting the fundamentals, the game of pass-the-parcel goes one more day, and you should watch for a collective afternoon plunge.
• Second checkpoint: the "quality correction" in the WF6 branch. Yesterday the three zero-revenue concept names Heyuan Gas, Do-Fluoride and Zhongjuxin hit limit-up while CSSC Specialty Gases rose only 7.57%. If those three gap down today while CSSC Specialty Gases holds up, the market is starting to re-rank on fundamentals; if the three seal limit-up again, the branch is still in a purely sentiment-driven phase and even a gap-up in CSSC Specialty Gases cannot be trusted.
• 爱丽家居 Aile Home (603221) issued a risk warning after 10 consecutive limit-ups; whether it locks limit-up on the open or gaps up and breaks is the thermometer for the meme-stock line.
• 云南锗业 Yunnan Germanium (002428): after 3 consecutive limit-ups, whether sealing orders can stay above RMB 300 million. On 8/6 sealing orders were RMB 585 million with Dragon-Tiger List net buying of RMB 1.154 billion; if it gaps up in the auction but sealing orders collapse below RMB 100 million, that is the classic distribution shape.
• 聆达股份 Lingda Group (300125) on its first day back after having its warning removed is purely event-driven and unrelated to the main lines; do not use it to read sentiment.

8.2 Sector signals

• Healthy shape: the PCB/CCL sector gaps up and is then led by real-earnings names such as WUS Printed Circuit, Shengyi Electronics and Shennan Circuits, while yesterday's broken-limit-up names Jiangnan New Material, Litong Electronic and Huazheng New Material lag — that says the Goldman note is being used for a quality rotation, confirming on the tape what the China Securities Journal called "a new earnings-driven cycle".
• Dangerous shape: PCB gaps up but is still led by yesterday's broken-limit-up high-priced names while WUS, Shennan and Shengyi lag. That means the new research note is being used to carry high-level names, the classic signal of a branch topping out.
• Flow-driven candidates (a line the first draft of this brief missed yesterday and that must be watched separately today): among the top net main-force inflows on East Money on 8/6 there is a batch of names with "huge net inflow + still deeply down over 20 days", the classic shape of money building positions against the trend — JCET Group (600584, main board, limit-up, net inflow RMB 2.297 billion, largest in the market, 20-day −26.7%, P/E 82×), T&S Communications (300394, ChiNext, +3.89%, +RMB 1.197 billion, 20-day −17.0%), Founder Technology (600601, main board, limit-up, +RMB 975 million, 20-day −15.7%), Victory Giant Technology (300476, ChiNext, +5.84%, +RMB 890 million, 20-day −15.7%). Of these, JCET Group is in semiconductor packaging and testing, a direction this brief's branch table does not cover. If this batch keeps rising on volume today, the main line is shifting from "news-driven" to "flow-driven", and packaging / advanced packaging must be added to the watch pool intraday.
• Coal: absorption in the first half hour is the key. Yesterday 6 names sealed late in the session with sealing orders of RMB 1–4 million; if they gap down today and cannot re-seal quickly, yesterday's late ramp is confirmed as a one-day wonder. Watch Liaoning Energy (600758) and Dayou Energy (600403) in particular — if they plunge first, that is the leading indicator of the sector rolling over.
• Storage chain: overnight Western Digital was −13% and SK Hynix −5%, but A-share storage is already down 40%–55% over 20 days (GigaDevice −42.0%, Longsys −40.9%, Biwin Storage −44.4%, Shenzhen Deren −54.9%). Watch whether it is "gap down then rally" or "gap down then keep falling" — the former says the bad news is exhausted. Musk's "memory is the biggest bottleneck" and NVIDIA cutting Rubin Ultra's memory, both from this morning, are the bulls' handholds.
• Communications equipment (most limit-ups yesterday, 8): this brief does not list it as a bullish branch (reasons in 2.0), but it was yesterday's sentiment battleground. If Xintian Technology (300615) and Tongyu Communication (002792) gap down and weaken together today, the market-wide limit-up count and money-making effect will be visibly dragged down — treat it as a sentiment indicator, not as an opportunity.
• Humanoid robots (this afternoon's sentiment variable): Unitree Robotics (688836) holds its online roadshow 14:00–17:00, with an issue P/E of 219.23× versus an industry average of 38.56× and DeepSeek and Tencent in the strategic placement; layered on top of the overnight Tesla/SpaceX TeraFab news that "about 25% of the compute is for Optimus". Two same-direction items land on the same day, but the net effect may go either way: the roadshow may lift robotics sentiment, or it may weigh on liquidity because of "a high issue P/E + the 8/10 subscription draining funds". This brief's branch table does not cover robotics, so this afternoon needs live observation with no preset direction.

8.3 Single-stock signals

• Absorption in the first 5 minutes: watch the opening five-minute volume in the two top picks, Yuanjie Semiconductor and Yunnan Tin. Both rose only marginally yesterday (+2.70%, +1.68%), so if they start today on visibly heavier volume, that is the most direct evidence of "money rotating from sentiment names into earnings names".
• Sealing fast without breaking: of yesterday's 79 limit-ups, 41 sealed between 09:25 and 10:00; if that ratio falls today, the collective push is weakening.
• Being blocked by a stronger name: watch the money see-saw between Yunnan Germanium and Yuanjie Semiconductor — if Yunnan Germanium breaks its limit-up while Yuanjie Semiconductor rises on volume, that is the healthiest gear change this branch could make.
• Large-order net inflow: focus on whether the electronic chemicals industry can post a second consecutive day of net inflow (yesterday's +RMB 1.811 billion was the largest in the market). Two consecutive days of net inflow is the minimum bar for a genuine main line; if it turns to net outflow today, then yesterday's inflow was absorbed by concept stocks sealing limit-up, and the branch should be downgraded.

8.4 Risk signals (if these appear, lower your expectations for today's opportunities)

1. Gap up then plunge: the indices gap up and turn negative within half an hour, with fewer than 20 limit-ups before 10:30.
2. A lone limit-up: only the leader in a branch seals, while all the followers turn negative.
3. The core names do not follow: if WUS Printed Circuit / Shennan Circuits (PCB) and CSSC Specialty Gases (specialty gases) stay persistently weaker than the small caps, this is pure theme speculation rather than industry allocation.
4. Yesterday's strong themes rolling over: coal and tungsten opening green and selling off on volume. Tungsten is the direction most likely to break down first, but the basis is the commodity side (August long-term contract prices down another 8%, July APT −15.28%, quotas +3.45%), not the flow side — the two flow conventions point opposite ways (Dragon-Tiger List net selling of RMB 1.9 billion versus East Money main-force net inflow of +RMB 385 million), which is not enough on its own to support a "distribution" conclusion.
5. External: US July non-farm payrolls tonight at 20:30. Against a 55% probability of a Fed hike in September and Bank of America expecting three consecutive hikes, strong data would reinforce hike expectations and could trigger risk-avoidance selling into the close. This is today's only risk that "goes off in the afternoon".
6. Typhoon "White Dolphin": affects East China 8/8–8/11 (Shanghai gusts of force 8–9, cumulative rainfall possibly above 100 mm). No direct impact today, but East China shipping, airport and outdoor construction names may react in advance.

---

9. Final Recommendations

① The 5 stocks most worth watching today

Convention note: the "8/6 net main-force inflow" column below is on the East Money single-stock convention; P/E (TTM) is on the 2026Q1 reporting-period convention, excluding H1 pre-announcements not yet booked (see the footnote in 4.2). The three names with net main-force outflow on the day are flagged honestly and not hidden.

⚠️ A self-contradiction that must be spelled out: this brief uses fund flows to reject tungsten and some concept stocks on the one hand, while on the other recommending three names that saw net main-force outflow yesterday (Yunnan Tin, CMOC Group, Huayou Cobalt). This is not a slip, it is a trade-off — these three are recommended on earnings pre-announcements and valuation position, not on flow momentum; and "money flowed out yesterday" is the flip side of the fact that they have not yet been chased. But readers have a right to know about this contradiction, so it is set out honestly. If they continue to see heavy outflows today, the basis for these recommendations is weakened.

This list deliberately excludes yesterday's two most glamorous names: Yunnan Germanium (largest sealing orders, largest Dragon-Tiger List net buying in the market) and CSSC Specialty Gases (the hardest earnings). The reasons for Yunnan Germanium are in 5.2 — a RMB 59.4 billion market cap against an ex-non-recurring loss, three straight years of negative operating cash flow, a 40x P/B and the 99.3rd historical percentile. The reasons for CSSC Specialty Gases are in 5.5 — its earnings are real, but the supply narrative underpinning its valuation is denied by the companies involved, the interim report was already priced with a limit-down back on 7/18, operating cash flow is −RMB 241 million, and it is up 666% year to date. This is the single judgement this brief most wants readers to take away: today, "the one money most endorses", "the one with the hardest earnings" and "the one most worth holding" are three different stocks.

② Today's 3 strongest branches

Two sentences that must accompany the third branch (details in 2.3 and 6.3): it falls from an initial No. 1 / S-grade to No. 3 / A-grade — the demand logic is still hard and CSSC Specialty Gases' Q2 earnings are real, but the supply-side narrative is unverified, WF6 prices have been consolidating for two months, and yesterday's tape showed the topping signature of "the less relevant to WF6, the better it rose". It is in the top three because of demand logic and fund flows, not because it can be bought today — 4 of the 6 names in the branch are on this brief's Pass list.

Right behind them, the 4th and 5th branches are compute metals (tantalum/indium/molybdenum) and copper / cobalt (the DR Congo ban); the latter ranks only 5th, yet its representative names CMOC Group (603993) and Huayou Cobalt (603799) made this brief's Top 5 stock list on "low position, low valuation, no rise on 8/6" — branch strength and single-stock risk-reward are decoupled today, so please do not conflate them.

③ Directions to treat with great caution today

1. Every name in the WF6 branch except CSSC Specialty Gases — Heyuan Gas (no qualification obtained, no orders signed, the company itself says no material impact for 2–3 years, has clarified for the 5th time), Do-Fluoride (main business is lithium hexafluorophosphate, explicitly "no volume production or sales of WF6"), Haohua Technology (WF6 is 0.13% of revenue), Zhongjuxin-U (no substantive orders and loss-making itself). Three of these four hit limit-up yesterday and one was +9.5%, completely decoupled from their WF6 fundamentals.
2. Tungsten (China Tungsten & Hightech, Jiangxi Tungsten Equipment, Xiamen Tungsten) — August long-term contract guidance down another 8% month on month, July APT −15.28%, quotas +3.45% year on year (loosening), Dragon-Tiger List net selling of nearly RMB 1.9 billion. The commodity and the share price are running in opposite directions.
3. The earnings-divergent coal names (Liaoning Energy, Zhengzhou Coal Industry & Electric Power, Dayou Energy) — H1 respectively a first-ever loss of −3282%, a loss widening 52.27%, and a loss of RMB 635 million. Liaoning Energy's pre-announcement says in black and white that "coal output and selling prices fell year on year", yet it hit limit-up in a "coal prices are rising" rally.
4. The six high-flying stocks that issued clarifications yesterday — Tongyu Communication, Olai New Material, Bojay Electronics, Baoding Technology, Kexiang, Jiangxi Tungsten Equipment. A clarification announcement + a high position + large sealing orders is the classic combination for gapping up and fading the next day.
5. Ellington Electronics — 2026Q1 net profit −67.60% against peers at +63%~+122%; a fundraising IRR of only 13.01%; customers all automotive. This is a routine capacity-replacement project wrapped in a compute narrative.
6. Pure sentiment high-flyers such as Aile Home and Xintian Technology — 10 / 4 consecutive limit-ups, negative P/E, and risk warnings issued by the companies themselves.
7. Chasing the storage chain — overnight Western Digital −13% and SanDisk −6%, with guidance rather than results doing the killing. The overnight direction is clearly under pressure; but the A-share chain is already down 40%–55% over 20 days, so certainty is low in both directions.

④ The final one-sentence judgement

Today is the first trading day on which "earnings delivery" replaces "concepts" as the driver, and the market will complete the switch by way of two falsifications — the hottest branch (WF6) has had its supply-side story denied by the Japanese companies themselves, and the most explosive stock (Yunnan Germanium) has an ex-non-recurring loss and three straight years of negative operating cash flow for fundamentals; meanwhile the only fresh catalyst (Goldman raising PCB/CCL) has landed on a sector where "the industry reading looks ugly while the leading single stocks are being bought heavily", and JCET Group's RMB 2.297 billion, the largest net inflow in the market, happened right there. So today's main line is not "can good news rescue bad chips", but "will the concentration of money into leaders accelerate"; put your attention on names like Yuanjie Semiconductor and WUS Printed Circuit that "rose little yesterday, have earnings on the ground and are already seeing inflows", rather than on the batch that "had the largest sealing orders yesterday but is issuing clarifications this morning, or whose gains are inversely correlated with fundamentals".

---

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