A-Share Market Recap | 2026-08-05 Wednesday

Data-basis disclosure first (please read this section first)

1. Quote basis: All single-stock closing prices, percentage changes, volume ratios, turnover rates, turnover value, main-force net inflow, market cap and trailing P/E in this report are as of the 2026-08-05 close, taken from the Tencent Finance single-stock snapshot (stock_zh_a_spot_tx, 5,537 records). The limit-up pool / failed-limit pool / limit-down pool / performance of yesterday's limit-ups are taken from East Money (stock_zt_pool_em, etc.). Industry percentage changes and industry net inflows are taken from the Tonghuashun industry basis (90 industries) — industry net inflow and single-stock main-force net inflow belong to two different bases and cannot be added together. Index closes are taken from the Sina index snapshot.
2. Price-limit bands differ by board, and every stock in this report is tagged with its board: Shanghai/Shenzhen main board ±10%, ChiNext (30/301) ±20%, STAR Market (688) ±20%, Beijing Stock Exchange (starting with 8/9) ±30%. "Limit-up" in this report has been counted separately against each board's own limit, not uniformly at 10%.
3. Items not obtained / left blank in this report:
   • Northbound single-day net inflow: cannot be provided. Since August 2024 the exchanges have discontinued real-time/post-close disclosure of northbound single-day net buying; the "net buy turnover" figures returned today by the interface for both Shanghai Connect and Shenzhen Connect are 0.00, which is a missing disclosure basis, not a zero net figure for the day. Section 5 of this report substitutes industry fund flows and single-stock main-force flows, and separately gives the advance/decline counts for Stock Connect names.
   • 2026-08-05 Dragon-Tiger List: not yet published by the exchanges at press time; this report does not list today's Dragon-Tiger stocks or seats. What Section 5 gives is the obtainable substitute basis (main-force net inflow, industry net inflow), and no inference whatsoever is drawn about "active-seat preferences".
4. This report corrects one factual error from yesterday's pre-market brief, see Section 2.4: the pre-market brief said "ChangXin Memory is not a listed company"; in fact its listed entity 长鑫科技 (CXMT, 688825, STAR Market) was already listed on the STAR Market on 2026-07-27.
5. A major exogenous piece of news broke during today's session (the US FCC intends to ban imports of new-model Chinese optical transceiver modules). That news was released at 08-04 21:36 (Beijing time) by domestic media citing a Reuters report, falling inside the news time window that this morning's pre-market list declared for itself (08-04 15:00 → 08-05 07:00), yet the pre-market list did not include it. This is the single most important item in this report's reconciliation; see Section 2.3.

⚠️ Data-pull and implementation details for this run (internal only):

• The East Money full-market quote interface stock_zh_a_spot_em was unavailable for the fifth consecutive trading day (RemoteDisconnected); we continued with Tencent's stock_zh_a_spot_tx (5,537 records). This recommendation has now been raised five days running: please just make the Tencent source the default primary and demote East Money to backup — trial-and-erroring it here every day is pure waste.
• The East Money index interface stock_zh_index_daily_em also returned RemoteDisconnected today; Sina's stock_zh_index_daily (daily bars) is only updated through 08-04, so using it would write yesterday's close into the report as today's. In the end we used the Sina real-time snapshot stock_zh_index_spot_sina (562 records) to get the 08-05 close, and cross-checked it via "previous close 3822.2846 = the 08-04 close" to confirm it is same-day data. This is today's most dangerous trap: the daily-bar interface throws no error, it is simply silently one day stale. Recommend adding an assertion in the generate pipeline — the index snapshot's "previous close" must equal the prior trading day's close, otherwise abort.
• The East Money concept-board interface stock_board_concept_name_em returned HTTP 502; Tonghuashun's concept interface stock_board_concept_name_ths returns only names + codes with no quotes. Branch verification in Section 3 of this report was therefore done entirely on the Tonghuashun 90-industry basis, with no concept-board basis used — this does not affect the conclusions, but the basis source should be known.
• The Dragon-Tiger List genuinely could not be obtained today; this is not a fault on our side: stock_lhb_detail_em(20260805) returned None (TypeError), stock_lhb_hyyyb_em likewise None, and Sina's stock_lhb_detail_daily_sina(20260805) raised KeyError: 股票代码. The same set of interfaces returned 91 rows normally for 20260804, showing the exchanges simply have not published for the day rather than the interfaces being broken. Per the iron rules we wrote "cannot yet be confirmed" in the body, and did not pass off the trailing-one-month statistics list (stock_lhb_stock_statistic_em, 842 rows, latest listing date only through 08-04) as today's Dragon-Tiger List.
• The search engine gave a misleading summary on "Zhongji Innolight added to a US Department of Defense list": it merged the 1260H military-linked company list of 2026-06-10 (an old event, to which the company had already responded at the time that it had "no substantive impact on operations") with today's FCC optical-module import ban (a new event) into a single item. We verified the release dates of the two separately before writing. This is another instance of "an aggregated summary stitching an old event into a new one", of the same family as the news-source code error hit in yesterday's pre-market brief — whenever a summary contains "the company responded", you must go back to the original report and verify the date of that response.
• Foxconn Industrial Internet (601138) hit limit-up today, and AI-generated copy attributed it to "earnings + buyback + dividend". Checking each release date item by item: the earnings pre-announcement was 7/9, the dividend implementation 7/23, the buyback plan 7/27 — all old news, not one of them today or last night. Section 3.4 of this report therefore characterises it as "capital seeking a route" rather than "news-driven", and does not accept that after-the-fact attribution.
• The pre-market error "ChangXin Memory is not a listed company" was caught in this report only when running the main-force net-outflow ranking and running into 688825 长鑫科技 20-day +527%. Recommend adding a step to the pre-market flow: whenever writing "XYZ is unlisted / not a listed company", first reverse-check the short name once against the quote snapshot — the cost is trivial, and this time it directly caused the pre-market brief to miss the most direct corresponding name for the day's news.

---

0. One-Line Recap

Today's sentiment temperature: a sharp turn from freezing point back to warming; the whole session was a textbook deep V. All four major indices gapped down together at the open (SSE Composite −0.19%, SZSE Component −1.73%, ChiNext Index −3.35%), the direct cause being the Reuters disclosure at 08-04 21:36 that the US FCC is drafting a ban to prohibit imports of new-model Chinese optical transceiver modules; at the close the SSE Composite was +1.47%, SZSE Component +1.86%, ChiNext Index +1.32%, and the STAR 50 led with +4.78%, with combined turnover of RMB 2.68 trillion (up about 21% from yesterday's RMB 2.21 trillion) and 3,725 advancers / 1,621 decliners market-wide.

The strongest main line is not yesterday's. What genuinely led today was semiconductor equipment/materials (electronic chemicals +7.53%, semiconductors +5.63%) and non-ferrous metals (precious metals +7.72%, minor metals +6.12%) — and this is precisely the branch that this morning's pre-market list judged to be "the absent third leg" and on which it put 中微公司 (AMEC, 688012, STAR Market) on Pass. Meanwhile 中际旭创 (Zhongji Innolight, 300308, ChiNext) and 新易盛 (Eoptolink, 300502, ChiNext), ranked 1st and 4th pre-market, closed −7.27%, −5.29% respectively.

Pre-market list hit-rate overview: of the 18 long names, 6 clearly delivered (33.3%), 7 partially delivered, and 5 were clearly wrong; of the Top 5 recommendations, 2 clearly delivered (40%). And among the 6 names put on Pass / watch-only, 中微公司 +12.67% and 有研新材 (GRINM Advanced Materials, 600206, main board) limit-up were among the strongest of the entire session. The self-criticism in this report is not "we got the direction wrong" but that the pre-market brief spent enormous space debunking old news yet missed the one new item that fell inside its own time window and decided the whole session that very day (see 2.3).

Tone for tomorrow: the main line is not dead, but "which leg" has been forcibly switched. Capital's choice today was extremely clear — out of optical modules with US export exposure, into domestic equipment/materials, PCB, optical fibre/cable and whole-machine contract manufacturing that have no such exposure. Tomorrow, watch whether this "ban-avoidance" logic continues, rather than whether optical modules can rebound.

---

1. Market Overview

1.1 Indices

• Combined turnover of the two markets was RMB 2,679.9 bn (Shanghai RMB 1,208.6 bn, Shenzhen RMB 1,450.8 bn, Beijing RMB 20.5 bn), up 21.1% from RMB 2,213.6 bn on 08-04.
• All four indices opened at their session low and closed near their session high, with the ChiNext Index swinging 6.08% across the day (from −3.35% to +1.32%). This is a classic re-allocation of capital after a one-off clearing of bad news, not an ordinary gap-up-fade or gap-down-fade.
• Today happened to trigger both of the opposing thresholds set in Section 8.2 of the pre-market brief at the same time: turnover RMB 2.68 trillion ≥ RMB 2.4 trillion (defined pre-market as "incremental confirmation, bullish") was triggered; the failed-limit rate 28.8% ≥ 25% (defined pre-market as "showing that yesterday's 9.8% was an outlier") was likewise triggered. Both signals holding at once means today was "incremental money present, but money rotating hands and rotating direction", not a simple expansion or contraction of sentiment.

1.2 Sentiment Indicators

Sentiment reading: warming, but the structure is pickier than yesterday. Three facts hold simultaneously:

1. Limit-up count fell from 138 to 104 and the failed-limit rate spiked from 9.8% to 28.8% — the difficulty of sealing a limit-up rose markedly; yesterday's "best failed-limit rate in three days" was indeed an outlier, and the pre-market brief's scepticism on this point was right.
2. But yesterday's 138 limit-up stocks still averaged +5.46% today (median +5.61%), 121 up / 17 down, with a 23.2% promotion rate — follow-through quality was very good. Yesterday's limit-ups were not a one-day affair.
3. The share of first-time limit-ups fell from 86% to 69%, and second consecutive limit-ups rose from 0 names to 27 — a ladder has been built where there was none, which is exactly the signal required by pre-market Section 8.2's "if more than 2 second-consecutive limit-ups appear, a relay ladder is beginning to form"; in fact 27 appeared.

In sum: today was a combination of "sentiment warming + a higher bar for sealing limit-ups + a ladder beginning to take shape" — a healthier structure than yesterday, at the cost of a narrowing broad-based money-making effect.

---

2. Pre-Market List Reconciliation

2.1 Item-by-item reconciliation of the 24-name master list

Delivery basis (declared up front to avoid hindsight bias): benchmarked against the SSE Composite's +1.47%. ✅ delivered = gain ≥ +4.47% (beating the benchmark by more than 3 percentage points); ⚠️ partially delivered = +1.47% to +4.47%; ❌ not delivered = < +1.47%. For "watch only / Pass" names, what is judged is whether the avoid recommendation was correct.

2.2 Pass-list check (19 items)

Pass-list performance: 13 of 19 correct (68.4%), 4 clearly wrong, 2 too conservative.

Worth a separate note is 顺钠股份 (Shunna Electric, 000533): passed pre-market on the grounds that "a turnover rate as high as 20% is classic high-level game-playing"; today its turnover rate rose further to 30.01% while the share price gained only 0.79% — a textbook sample of the day's "record volume, no gain". This class of Pass, based on turnover rate rather than on whether the news is true, was correct across the board today.

2.3 Hit rate and self-criticism

Hit-rate summary:

One-line self-criticism: the pre-market brief spent 95% of its compute on "distinguishing old news" and "dissecting financial quality", and did that well; but it missed the one new item that fell inside its own declared time window and decided the whole session from the opening bell.

Specifically, in three layers:

① The most serious item: missing the core bad news inside the window. Section 1.1 of the pre-market list explicitly declared the news time window as 08-04 15:00 → 08-05 07:00, and used an entire section (1.5, debunking old news) to verify the true release dates of 9 items and strike them all from today's catalysts — that work was itself of high quality. But within the very same window, Reuters reported on 08-04 that the US FCC is drafting a ban to prohibit imports of new-model Chinese optical transceiver modules, relayed by domestic media at 08-04 21:36 (Beijing time), more than 9 hours before the window closed, yet the list did not include it at all.

The consequence was direct: pre-market ranked 中际旭创 (300308) 1st (total score 87) and 新易盛 (300502) 4th (total score 83), and this news hit precisely those two — on the Counterpoint basis, 中际旭创 holds roughly 27% of the global data-centre optical transceiver market, the largest exposure of any company. Today the two fell −7.27% and −5.29% respectively, with 中际旭创 opening −13.90%.

How to fix the pre-market process next time (actionable): the current process checks "old news inside aggregated copy" very carefully, but news retrieval leans toward domestic financial aggregators in its coverage. A fixed new action must be added: for the top name of every main line, separately search overnight foreign-media coverage of "that name / that category + export controls / ban / tariffs / entity list". Today's item was broken first by Reuters, with domestic aggregated copy relaying it only in the evening; using domestic aggregators as the single entry point guarantees a miss.

② The second item: reading "a branch missing a leg" as bearish, when it was in fact tomorrow's main line. Item 3 of pre-market Section 2.5.2 listed "the clear absence of semiconductor equipment/materials" (中微公司 +2.55%, 沪硅产业 +3.03%, 汇顶科技 +0.63%) as one of the hardest pieces of bearish evidence, and on that basis put 中微公司 on Pass and judged 有研新材 "watch only". Today that "absent leg" became the strongest thing on the board: electronic chemicals +7.53%, semiconductors +5.63%, with 中微公司 +12.67%, 有研新材 limit-up, 沪硅产业 (National Silicon Industry Group) +7.05%, 南大光电 (Nata Opto-electronic) +7.48%, 北方华创 (Naura Technology) +6.87%.

This is not "bad luck" but a reasoning chain pointed the wrong way: when one link in a branch lags, it can be read as "demand is not real" (the pre-market reading) or as "there is still catch-up room" (today's reading). The pre-market brief chose between the two readings but never wrote "what if it is the latter" into a triggerable observation point. The only trigger condition it did write was Section 8.2's "中微公司 opening auction gapping up >5% = the third leg catching up", and today 中微公司 gapped up only +1.57% in the auction (below the threshold) yet closed +12.67% — the threshold was hung on the auction, the moment most easily contaminated by overnight sentiment, and on a gap-down day it was bound to fail.

How to fix the pre-market process next time: catch-up observation points should not be hung on the opening auction; hang them instead on relative strength at 10:30 a.m. (stock gain − industry gain), a metric unaffected by the opening gap.

③ The third item: the watershed set pre-market pointed the right way; only the trigger mechanism was unforeseen. Pre-market Section 8.3 set 胜宏科技 (300476) as the "unconstrained readout" for PCB, on the criterion that "gapping up >8% = bullish; flat or gapping down = confirmation of a regime illusion, bearish". Today it gapped down −5.30% (which by the pre-market criterion should read as bearish), yet closed +17.12% with amplitude 25.26%. The insight that "a ±20% name will give a more truthful readout than a main-board limit-up" was entirely correct — it was the single biggest winner on the whole list; what was wrong was taking the readout in the first minute of trading. The same error shares a root with item ②.

④ The parts the pre-market brief got right, which should be recorded fairly:

• The Pass on 蜂助手 (301382) was correct: today's turnover rate 20.07%, volume ratio 2.89, amplitude 15.15%, closing +2.15% with main-force net outflow of RMB 200 mn — huge volume without price gain, exactly matching the twelve-point breakdown in pre-market Section 5.12.
• The "watch only" on 中孚实业 (600595) was correct: −0.89% against the trend on a day with 3,725 advancers.
• The direction on 天孚通信 (300394), "if the sector splits it falls the most", was right: it was the only one of the three optical-module majors to close green (+2.29%), because the pre-market brief correctly identified it as the one that had "risen the most and was the most expensive".
• The correction of 6 wrong stock codes remains valuable work — checked today against the correct codes, both 有研新材 (600206) and 江南新材 (603124) were indeed strong names, and trading on the aggregated copy's wrong codes would have bought the wrong tickets.
• Every Pass made on the grounds of "excessive turnover rate = high-level game-playing" was correct today without exception (顺钠股份, 蜂助手, and 红板科技 (Hongban Technology) with a 28.07% turnover rate and only a 4.46% gain). This is the most stable and effective criterion in this process; recommend raising its weight.

2.4 One pre-market factual error that must be corrected: ChangXin Memory is already listed

Section 5.9 of the pre-market list (兆易创新) read verbatim:

"ChangXin Memory is not a listed company; 兆易创新 and ChangXin are on the same side of the domestic memory ecosystem's supply chain, and this is not '兆易创新 winning an HP order'."

The grading logic in the second half is right and still holds today; but the first half is a factual error. ChangXin Memory's listed entity 长鑫科技 (CXMT, 688825, STAR Market ±20%) was already listed on the STAR Market on 2026-07-27, at an issue price of RMB 8.66, opening up 471.59%, the largest IPO in STAR Market history (raising more than RMB 57.9 bn, surpassing SMIC's RMB 53.2 bn in 2020).

The real cost of this error: last night's "ChangXin entering the HP/ASUS/Acer PC supply chain" was judged pre-market to be genuine new news within the time window, yet the most directly corresponding name for that news (长鑫科技 itself) never entered the list; the list contained only the indirectly mapped 兆易创新.

The two moved in exactly opposite directions today, and this must be recorded faithfully:

That is: the directly benefiting entity fell and suffered the largest main-force net outflow market-wide, while the indirectly mapped name surged. This report draws no conclusion on the cause (长鑫科技 has been listed for only 7 trading days, is up +527% over 20 days, and has a total market cap of RMB 3.63 trillion; the chip structure of a newly listed stock and news-driven moves are hard to separate, so this cannot yet be confirmed), but the result itself is the best footnote to the omission of "missing the direct name" — adding it would not necessarily have meant recommending it, but at least its enormous net outflow would not have been treated as non-existent.

---

3. Verification of Today's Main Lines

3.1 Branch verification master table

3.2 Today's most important finding: the market split "optical communications" precisely in two

This is the most informative price action of the day, and a point the pre-market brief's insufficient classification granularity failed to anticipate.

The FCC draft targets optical transceivers, i.e. the devices doing high-speed optical-electrical conversion inside data centres; it does not touch optical fibre, cable or submarine cable. Today the market priced this boundary down to the individual stock:

All belong to the same "communications equipment" industry (+3.06%, net inflow of only RMB 643 mn, down 96% from RMB 16.114 bn yesterday), yet inside it net outflows and limit-ups coexisted. Pre-market these two categories were merged under a single "AI computing hardware · optical modules/CPO" branch, with module makers listed among the representative stocks — on a day with no policy shock such a merger is fine, but the moment an external policy cuts along category lines, a merged classification makes the conclusions for the entire branch fail together.

The direct implication for tomorrow: do not judge optical communications as a single whole for continuation or ebbing; it must be regrouped by "does it have US export exposure or not".

3.3 The unexpected main line the pre-market brief missed entirely: non-ferrous metals

Precious metals +7.72% (14 up, 0 down, net inflow RMB 6.513 bn, 2nd of the session), minor metals +6.12% (29 up, 0 down), industrial metals +3.11% (net inflow RMB 6.059 bn, 3rd of the session), energy metals +3.76% (12 up, 0 down) — all four non-ferrous-related industries had zero decliners, and 紫金矿业 (Zijin Mining, 601899, main board) took RMB 2.847 bn of main-force net inflow, 2nd market-wide, behind only 工业富联.

The pre-market list did not mention non-ferrous metals from start to finish. Public reporting suggests the driver came mainly from the macro side: signals of easing US-Iran tensions (overnight Brent crude fell more than 6%) depressed inflation and rate-hike expectations, with the Fed's September rate-hike probability falling from 68% to 59%, and falling real rates being bullish for precious metals; at the same time, spillover from loosened positioning in tech also flowed into non-ferrous.

It must be fairly noted that Section 1.1 of the pre-market list did in fact include the source of this news — item 18, "US after-hours: oil prices tumbled, Trump said the US and Iran will hold talks and that he had cancelled the strike plan on Iran" — but its impact level was only graded B, the branches listed were "oil (bearish) / aviation and chemicals (bullish)", and it was never mapped onto the "rates → precious metals" chain.

This is a process improvement that can be hard-coded: for geopolitical-easing / oil-slump news, in addition to mapping to oil and downstream chemicals, one must simultaneously map the macro chain "inflation expectations → rate-hike probability → real rates → precious metals/non-ferrous". The strength of that chain today (RMB 6.513 bn net inflow, 14 up 0 down) far exceeded the B grade given pre-market.

34 工业富联601138today's largest single fund flow, but not news-driven · Foxconn Industrial Internet

工业富联 (601138, main board ±10%) hit limit-up +9.99% and closed at RMB 65.92, with turnover of RMB 18.55 bn, main-force net inflow of RMB 4.266 bn, 1st market-wide, a volume ratio of 2.12 and a total market cap of RMB 1.31 trillion.

What must be made clear: it had no new news today. Checking item by item the actual release dates of the three pieces of good news cited in the market's attribution:

The cited "good news"	Actual release date	Days ago
H1 earnings pre-announcement: net profit RMB 23.4–24.4 bn (+93% to +101%), AI server revenue +230%	2026-07-09	19th trading day
2025 dividend implementation (RMB 6.5 per 10 shares, full-year dividend RMB 19.45 bn)	2026-07-23	9th trading day
RMB 1–2 bn buyback plan	2026-07-27	7th trading day

Table scrolls horizontally

All three are old news; not one of them fell today or last night. This report therefore does not accept the after-the-fact attribution of "earnings + buyback + dividend driving the limit-up", and instead characterises it as: after optical modules were sold off over export exposure, AI computing-power bulls needed a substitute slot with no US export-control exposure and a market cap large enough to absorb big money — 工业富联 (predominantly domestic manufacturing and assembly, with a P/E of 32.2, on the low side among AI hardware leaders) became that slot.

This judgment has a falsifiable point, written here for checking tomorrow: if the characterisation holds, 工业富联's strength should persist in step with optical modules' weakness; once FCC ban risk is lifted and optical modules rebound, its excess return should converge.

---

4. Limit-Up Ladder and High Boards

4.1 High boards (≥3 consecutive limit-ups)

• The highest board, 传智教育, advanced to 8 boards (7 yesterday), but with 4 failed limits, sealing only at 14:21, and an 18.79% turnover rate — the worst sealing quality of any high board today, with a high probability of divergence tomorrow.
• 欣天科技 (300615) is the only tech name among the high boards, and it is a 3rd consecutive limit-up on a ±20% board (+19.97%), opening at the limit and sealed at 09:25, zero failed limits, turnover rate only 2.67% — the best sealing quality of any consecutive-limit stock today. But it is a small-cap elasticity name within communications equipment (turnover of only RMB 61 mn), highly identifiable but with low carrying capacity.
• 神雾节能 (Shenwu Energy Saving, 000820), on 4 boards yesterday, broke its streak today at −3.29%, with a 31.61% turnover rate — the risk signal in pre-market Section 8.4, "an ebb in yesterday's strong names will suppress market-wide sentiment", was triggered, yet the index still printed a deep V today, showing that leadership of sentiment has passed from hot-money themes to institutional main lines.

4.2 The second-consecutive-limit-up ladder (27 names, versus 0 yesterday)

This is the biggest structural change today. Yesterday there were zero consecutive limit-ups within the main line; today 27 second-consecutive limit-ups appeared, concentrated in the main-line branches:

Pre-market Section 8.2 set the condition "if components/communications equipment produce more than 2 second-consecutive limit-ups, a relay ladder is beginning to form" — in fact components produced 4 and communications equipment 1 (plus 欣天科技 on 3 boards), so the signal holds. This does not contradict "the limit-up count falling from 138 to 104": the market moved from yesterday's broad-based first-board diffusion into today's differentiated ladder relay, which is a more mature structure.

4.3 Board and industry distribution of the limit-up pool

Non-main-board limit-up stocks (counted against each board's own limit, 10 in total): 欣天科技 (300615) +19.97%, 万集科技 (Wanji Technology, 300552) +20.02%, 智动力 (Zhidongli, 300686) +20.00%, 新开普 (Newcapec, 300248) +20.02%, 天禄科技 (Tianlu Technology, 301045) +20.00%, 浩通科技 (Haotong Technology, 301026) +19.99%, 正帆科技 (688596) +20.00%, 中巨芯 (688549) +20.01%, 皖仪科技 (Wanyi Technology, 688600) +19.98%, 龙鑫智能 (Longxin Intelligent, 920117, Beijing Stock Exchange) +29.97%.

Basis reminder: if limit-ups were judged uniformly at ±10%, all 10 of the above would be missed, while a large number of ChiNext/STAR names up +10% to +20% would be wrongly recorded as limit-ups. This report's limit-up count of 104 and failed-limit rate of 28.8% have both been computed on the banded basis.

Top industries by limit-up count: auto parts 7, consumer electronics 7, minor metals 6, optics and optoelectronics 6, communications equipment 5, components 5, grid equipment 4, semiconductors 4, chemical products 4, general equipment 4.

Limit-up stocks with the largest turnover: 工业富联 (601138) RMB 18.55 bn, 长飞光纤 (601869) RMB 4.61 bn, 中钨高新 (000657) RMB 4.59 bn, 有研新材 (600206) RMB 3.44 bn, 利通电子 (603629) RMB 3.05 bn, 金安国纪 (002636) RMB 2.85 bn, 宏和科技 (Honghe Technology, 603256) RMB 2.85 bn, 星网锐捷 (Star-net Ruijie, 002396) RMB 2.72 bn.

4.4 Broken boards and failed limits

42 failed limits, a 28.8% failed-limit rate, distributed by industry as: components 6, the most, semiconductors 5, IT services 4, communications equipment 2, computer equipment 2, defence electronics 2, auto parts 2.

Failed-limit stocks with the largest turnover: 亨通光电 (600487) +8.24% (2 failed limits), 紫光股份 (Unisplendour, 000938) +7.63% (19 failed limits), 风华高科 (Fenghua Advanced Technology, 000636) +6.39%, 中国巨石 (China Jushi, 600176) +8.54%, 生益科技 (600183) +9.41% (10 failed limits), 太极实业 (Taiji Industry, 600667) +7.58% (turnover rate 22.47%).

Note: 生益科技, one of the pre-market Top 5, closed at +9.41% only after 10 failed limits. Although it is recorded as "✅ delivered" on this report's basis, failing to hold the limit 10 times shows that overhead selling pressure is real, and it cannot be read as "limit-up-grade strength".

Only 1 limit-down: 联创光电 (Lianchuang Optoelectronic, 600363, main board, consumer electronics), −10.01%, 1 consecutive limit-down day.

---

5. Fund Flows

5.1 Northbound funds: single-day net figure cannot be provided (missing disclosure basis)

Since August 2024 the exchanges have discontinued disclosure of northbound single-day net buying; the "net buy turnover" fields returned today by the interface for Shanghai Connect and Shenzhen Connect are both 0.00, and this is a missing basis, not a zero net figure for the day. This report draws no inference whatsoever about the northbound direction.

Obtainable substitute information (only advance/decline counts for Stock Connect names, not fund flows):

On the southbound side: Stock Connect (Shanghai) net outflow of RMB 1.806 bn, Stock Connect (Shenzhen) net inflow of RMB 1.200 bn, with a daily quota of RMB 42.0 bn each; the Hang Seng Index was +0.24%.

5.2 Dragon-Tiger List: 2026-08-05 not yet published

At press time the exchanges had not yet published the 08-05 Dragon-Tiger List data (the same set of interfaces returns 08-04 normally). This report lists no stocks on today's list and infers nothing about active-seat preferences. If that data is needed for tomorrow's pre-market brief, it should be pulled again before the next day's open.

5.3 Industry fund flows (Tonghuashun industry basis, unit: RMB bn)

Top net inflows:

Top net outflows: banks −4.159 (+4 up, 38 down, −1.15%, sold for a second straight day), grid equipment −2.545, power −2.404, baijiu −1.871, construction and decoration −1.631, traditional Chinese medicine −1.296, minor metals −0.773 (up 6.12% yet net outflow — retail-driven), chemical pharma −0.671.

Three key readings:

1. Communications equipment net inflow collapsed from RMB 16.114 bn to RMB 643 mn (−96%), the largest decline of any industry — pre-market Section 8.2 asked "can semiconductors sustain net inflows"; the answer is that semiconductors did sustain them (RMB 9.011 bn, still 1st, though down 61%), while communications equipment went nearly to zero. This maps precisely onto the exposure distribution of the FCC ban.
2. Minor metals rose 6.12% yet saw a net outflow of RMB 773 mn, in contrast with precious metals (+7.72%, net inflow RMB 6.513 bn) — within non-ferrous, precious metals are being bought by institutional money, while minor metals look more like a bandwagon.
3. Banks saw large net outflows for a second consecutive trading day (yesterday −RMB 8.045 bn, today −RMB 4.159 bn), so the "high-to-low rotation funding source" logic judged pre-market continues to hold today; but today the SSE 50's +1.58% actually beat the SSE Composite, showing that the heavyweight end has moved from one-way bleeding to differentiation.

5.4 Single-stock main-force flows

Top 10 net inflows (RMB bn): 工业富联 (601138) +4.266, 紫金矿业 (601899) +2.847, 亨通光电 (600487) +1.838, 兆易创新 (603986) +1.774, 胜宏科技 (300476) +1.523, 紫光股份 (000938) +1.363, 生益科技 (600183) +1.293, 洛阳钼业 (603993) +1.234, 北方华创 (002371) +0.980, 芯原股份 (VeriSilicon, 688521) +0.964.

Top 10 net outflows (RMB bn): 长鑫科技 (688825) −2.542, 新易盛 (300502) −1.991, 中际旭创 (300308) −1.627, 立讯精密 (Luxshare Precision, 002475) −1.367, 药明康德 (603259) −0.886, 贵州茅台 (Kweichow Moutai, 600519) −0.831, 行云科技 (Xingyun Technology, 300209) −0.749, 东山精密 (Dongshan Precision, 002384) −0.744, 光迅科技 (002281) −0.724, 永鼎股份 (Yongding, 600105) −0.590.

The structure is extremely clear: the net-inflow list is monopolised by "the links of AI hardware with no US export exposure + non-ferrous"; the net-outflow list is monopolised by "optical modules + consumer-electronics contract manufacturing + CXO leaders". This is neither a broad rally nor a broad decline; it is a reallocation of capital along the line of "export-control exposure".

Also to be recorded: 药明康德 (603259) was +4.35% today yet had a main-force net outflow of RMB 886 mn on a volume ratio of 2.20 — rising on expanding volume while the main force flows out is a classic profit-taking signature, consistent with the pre-market judgment that "the catalyst has already been delivered; what is being bought today is the expectation of a second board".

---

6. Outlook for Tomorrow (2026-08-06 Thursday)

6.1 Judgment on main-line continuation

Core judgment: the AI computing-power main line is not over, but its "tradable portion" has been forcibly switched once by policy. Tomorrow's key is not guessing an optical-module rebound, but confirming whether the "ban-avoidance" fund flow can carry into a second day.

Evidence supporting continuation:

• Combined turnover of RMB 2.68 trillion, up 21% from yesterday, genuinely incremental money rather than a stock-of-funds game (this stands in clear contrast with yesterday's stock-of-funds feature of "volume ratio ≈ 1.0": today 中际旭创's volume ratio was 1.41, 胜宏科技's 1.84, 工业富联's 2.12).
• Yesterday's 138 limit-up stocks still averaged +5.46% today, 121 up 17 down, solid follow-through.
• Second consecutive limit-ups rose from 0 to 27; the ladder has formed.
• The semiconductor industry ranked 1st market-wide in net inflow for a second consecutive day (RMB 23.368 bn → RMB 9.011 bn).

Evidence calling for caution:

• A failed-limit rate of 28.8% (9.8% yesterday); sealing a limit-up has become markedly harder, with the components industry topping the list at 6 failed limits.
• The limit-up count fell from 138 to 104; the money-making effect is narrowing.
• The FCC ban remains an unresolved event — it is currently only at the draft stage, and the Reuters original explicitly states that "the FCC may still modify or shelve the restrictions in question", with US officials "hoping to publish and implement within 2026". This means uncertainty will persist rather than having been cleared out in one go today.

6.2 Key focus tomorrow

6.3 Avoid tomorrow

6.4 Input prompts for tomorrow morning's pre-market list

1. Top priority: verify the latest developments on the FCC ban. This is the single biggest variable for tomorrow's session. Searches needed: whether the FCC has issued a formal proposal (NPRM), whether there is a hearing/comment timetable, whether any company announcement discloses substantive impact, and whether 中际旭创 (300308) and peers have issued clarifications or risk warnings. Until this is clear, any rebound in optical modules can only be characterised as an oversold repair, and cannot be upgraded to a main-line return.

2. Add one fixed search action (today's lesson): for the top name of every main line, separately search overnight foreign-media coverage of "the name/category + export controls / ban / tariffs / entity list". Today's bad news was broken first by Reuters, with domestic aggregated copy relaying it only in the evening, so using domestic financial aggregators as the single entry point guarantees a miss.

3. Fix the timestamp used for catch-up observation points (today's lesson): do not hang the "is it catching up" threshold on the opening call auction (today 中微公司's auction was only +1.57%, below the 5% threshold set pre-market, yet it closed +12.67%; 胜宏科技 gapped down −5.30%, which by the pre-market criterion should read as bearish, yet closed +17.12%). Switch to relative strength as "stock gain − industry gain at 10:30 a.m.", a metric uncontaminated by overnight gaps.

4. Branch classification needs one more cut along "policy exposure": today proved that "optical modules" and "optical fibre/cable" must be separated. Tomorrow morning, AI computing hardware should be listed in three tiers — ① with US export exposure (optical modules) ② domestic substitution with no exposure (semiconductor equipment/materials) ③ domestic supporting supply with no exposure (PCB, whole-machine contract manufacturing, optical fibre/cable) — rather than merged into one branch with a single strength rating.

5. Fill in the macro mapping chain: for geopolitical-easing / oil-slump news, besides mapping to oil and downstream chemicals, one must simultaneously map "inflation expectations → rate-hike probability → real rates → precious metals/non-ferrous". Today's pre-market brief included the US-Iran news but graded it only B and mapped it only to oil and aviation/chemicals, missing precious metals with RMB 6.513 bn of net inflow and 14 up, 0 down.

6. Add 长鑫科技 (CXMT, 688825, STAR Market ±20%) to the tracking pool of the domestic memory branch (listed on 2026-07-27, total market cap RMB 3.63 trillion). Today it was −1.27% with a main-force net outflow of RMB 2.542 bn, the largest market-wide, while the indirectly mapped 兆易创新 was +8.26% — this divergence itself needs an explanation tomorrow morning and can no longer be ignored.

7. Raise the weight of the "high turnover, no price gain" criterion. Every Pass made on that criterion today (顺钠股份 with a 30.01% turnover rate for a 0.79% gain, 蜂助手 with a 20.07% turnover rate and net outflow, 红板科技 with a 28.07% turnover rate for a 4.46% gain) was correct without exception, making it the most stable rule in this process at present.

8. Three arbitration points in the first 30 minutes of tomorrow's session: ① whether 胜宏科技 (300476) can close green on expanding volume (whether the PCB main line holds); ② whether 中微公司 (688012) can hold RMB 358 (whether the new main line continues); ③ the follow-through volume in 中际旭创 (300308) after its RMB 67.5 bn record — if turnover stays above RMB 40 bn tomorrow while the price does not rise, the position-cutting is not over.

---

⚠️ Risk warning: this recap is a post-close compilation of information and observations only and does not constitute investment advice. Data may differ in timeliness or basis; please refer to exchange/company disclosures as authoritative, and do not use this directly as a basis for trading.