寒武纪跌5.68%、成交77.1亿，是当天全场最大的一笔钱，可它一条公告都没有；同一天美股AMD涨5.2%、台积电涨3.…
- 三季度毛利率58.69%，连续三个季度往上走（54.33%、56.10%、58.69%），比去年同期54.24%高4.45个点
- 2026年股权激励的收入目标是135亿，前三季度做了95.26亿，四季度要做到39.74亿，比三季度的35.31亿再增12.6%
- 存货 44.97亿 → 82.48亿 → 116.65亿
- 一季度 营收28.85亿／毛利率54.33%／归母10.13亿
- 合同负债 3.96亿 → 1.88亿 → 1.41亿
寒武纪跌5.68%、成交77.1亿，是当天全场最大的一笔钱，可它一条公告都没有；同一天美股AMD涨5.2%、台积电涨3.0%，这77亿到底在跟谁对着干？
Cambricon fell 5.68% with turnover of 7.71 billion yuan, the largest single sum in the whole market that day, yet it did not issue a single announcement; on the same day, US stocks AMD rose 5.2% and TSMC rose 3.0% — who exactly is this 7.71 billion yuan fighting against?
The desk answers已核实Walnut, from public filings
这77亿没跟AMD对着干——时间上根本对不上；它跟的是周一那134亿买盘，和自家仓库里116.65亿存货。
先把问题里三个前提改掉。
第一，"一条公告都没有"不成立。10月24日（周六）寒武纪一口气发了8条公告，其中三条是关键：三季报、计提资产减值准备、关于公司近期生产经营情况的公告。周一10月26日是三季报后第一个交易日，它涨了3.84%，成交134.24亿，盘中冲到985元。所以不是"没消息"，是消息在周一被当成好消息买过一遍，周二被重新读了一遍。
第二，"当天全场最大的一笔钱"不成立。77.11亿在10月27日全市场排第六。前面是长鑫科技151.09亿、中际旭创140.11亿、新易盛114.33亿、宁德时代88.14亿、药明康德85.77亿。在半导体里它也只排第二，第一是长鑫科技。而且跟它自己比，周一是134.24亿，周二是77.11亿，只有前一天的57%。
第三，"跟AMD对着干"在时间上不可能。A股10月27日15:00收盘，美股当天21:30才开盘，中间差六个半小时。A股那天早上能看到的隔夜信息是10月26日的美股：AMD收531.28美元涨2.80%，台积电收424.37涨1.40%，英伟达收216.00涨2.39%。隔夜是暖的。你看到的AMD涨5.26%（559.20美元）是10月27日美东盘中15:31的价，那个涨幅出现在寒武纪跌完六个半小时之后，它不可能是原因。
那周二到底发生了什么，按小时排一遍就清楚了。
10月23日周五，董事会通过三季报和减值议案。10月24日周六，8条公告挂网。10月26日周一，股价+3.84%。10月26日15:00到17:07，盘后业绩说明会，45条回复。10月27日周二，-5.68%，77.11亿，缩量。
说明会开在收盘后，所以周一那根阳线是说明会之前定的价，周二的跌是说明会之后定的价。这条时间线比任何"资金博弈"的说法都硬。
再看周二是不是寒武纪一个人在跌。不是。上证-1.33%，沪深300-1.32%，科创50-4.81%，全市场606只涨、4922只跌。半导体200只平均跌4.56%。当天红的是银行，+1.34%。寒武纪跌5.68%，只比科创50多跌0.87个点。
更要紧的是同行业里有人涨。长鑫科技，7月27日上市，10月27日成交151.09亿，换手7.65%，收涨0.80%。板块里的钱没有离场，它换了个地方待着——从讲了两年的AI芯片龙头，换到上市刚三个月的新股。这77亿不是被谁打出去的，是没人接。
现在说报表。把三期单季拆出来（累计相减，一季报直接用累计值），损益表里找不出一处是坏的：
一季度 营收28.85亿／毛利率54.33%／归母10.13亿
二季度 营收31.11亿／毛利率56.10%／归母12.98亿
三季度 营收35.31亿／毛利率58.69%／归母14.04亿
三季度毛利率58.69%，连续三个季度往上走（54.33%、56.10%、58.69%），比去年同期54.24%高4.45个点；单季收入35.31亿、归母14.04亿，都是公司上市以来最高的一个季度。收入同比+104.5%，环比+13.50%（这个环比数是公司在说明会上自己给的，二季度是+7.84%）。单季净利率从41.7%掉到39.8%，掉的原因也不是减值——三季度单季资产减值1.05亿，比二季度的1.51亿还少——是研发费用一个季度从3.78亿加到5.23亿，多了1.44亿，增幅38.1%。
所以"财报变差了"这句话是错的。变差的不是赚钱能力，是资产周转。
一季度到三季度，资产负债表上这几行是这样走的：
存货 44.97亿 → 82.48亿 → 116.65亿
预付款项 18.97亿 → 29.14亿 → 30.44亿
应付票据 7.50亿 → 30.85亿 → 49.66亿
合同负债 3.96亿 → 1.88亿 → 1.41亿
单季经营现金流 +8.34亿 → -5.23亿 → -0.84亿
三句话能说清这里面在干什么。
一是116.65亿存货，按三季度单季营业成本14.58亿算，是八个季度的成本压在仓库里；而客户先付的钱（合同负债）只有1.41亿，不到四天的收入。存货比年初的49.44亿多了136%。赚钱的公司压货不奇怪，但压到八个季度的成本、客户先付的钱不到四天收入，这个组合只有在产能最紧的时候才敢做。
二是这堆货不是用现金堆的，是用票据堆的。三季度应付票据加应付账款合计67.93亿，比二季度末的41.61亿多了26.32亿；货币资金只剩5.55亿，理财38.11亿。公司没借银行一分钱短期借款，它是拿票据（银行承兑汇票，期限一般不超过六个月）去换一批要八个季度才周转完的存货。
三是减值已经开始了。前三季度计提存货跌价5.02亿，是去年同期0.87亿的5.77倍。公司自己的原话是"主要系公司前期对原材料进行了战略备货，随着原材料库龄增长，计提的存货跌价准备相应增加"。这句话是整份文件里最有信息量的一句：战略备货是有保质期的，不是免费的。
那说明会为什么没救回来？因为两小时、45条问答，没有新增一个能用的数字。问存货，答的是公告原文照抄；问采购计划，答"公司根据客户订单需求以及对市场未来需求的预测情况制定采购计划"，一个在手订单数都没有；被问到市场上流传的"机构交流纪要"，公司答"公司从未进行过相关沟通"，并称那属于"来源不明或未经核实的信息"。整场唯一的新增信息是两个：三季度环比+13.50%，以及前三季度收入构成里不包含智能计算集群系统业务。
第二条其实是被低估的好消息。集群业务是给地方政企建智算中心，账期长、毛利低、看财政。95.26亿收入里一块都没有，全是云端产品线，公司说这条线主要面向各类型商业客户——这意味着增长的质量比市场以为的干净。
所以反方最强的论点是什么？就是上面这两条：毛利率连续三个季度往上走、收入环比还在加速、经营现金流从二季度的-5.23亿收窄到三季度的-0.84亿、收入里没有智能计算集群系统这一块。四条都在往好的方向走。一个聪明人会说：你拿八个季度的存货说事，那叫产能锁定，Fabless公司抢晶圆和封测产能本来就要提前打款压货，用票据而不是现金去压，说明它对上游还有账期；等这批货按58%的毛利卖出去，116亿就是弹药，不是包袱。
这个反驳成立，而且现在没有文件能证伪它。三季报只有存货总额，分原材料、在产品、库存商品的明细和库龄要等年报，所以现在只能停在"两种解释都说得通"。
所以真正还没被定价的，不是"跌了多少"，是"这批货什么时候变成收入"。
2026年股权激励的收入目标是135亿，前三季度做了95.26亿，四季度要做到39.74亿，比三季度的35.31亿再增12.6%。做到了，116亿存货就是弹药，二季度那笔-5.23亿的现金流就是扩张成本；做不到，同一批原材料在2027年上半年会再计提一次跌价。这跟股价无关，跟两个数有关：年报里存货跌价准备的余额，和应付票据的到期结构。
证伪条件写三条，都很硬。一，四季度单季收入做到39.74亿以上，备货论成立，我这套"周转开始紧"的判断就错了。二，年报存货跌价准备余额显著高于5.02亿，积压论成立。三，出现单季经营现金流回正且存货环比下降，说明票据换存货的阶段结束，判断要改。
最后补一句当天的市场环境，避免把这77亿想得太有意图。10月27日是缩量普跌：4900多只下跌，科创50跌4.81%，钱往银行那边去，没有任何一条针对算力板块的政策或事件。我查了一手公告和当日全市场成交，找不到那天有哪条消息专门打在寒武纪身上。所以77亿更接近一次没人愿意出价的下跌，而不是一次有人蓄意的对决。真正的对决在前一天下午那两小时里——公司拿着95.26亿收入和37.15亿利润，没能回答"116亿存货什么时候变成收入"。市场第二天用缩量跌了5.68%投了这一票。
来源：中科寒武纪科技股份有限公司《2026年第三季度报告》，2026年10月24日（定期报告，无公告编号）；《关于2026年前三季度计提资产减值准备的公告》，2026年10月24日，公告编号2026-038；《关于公司近期生产经营情况的公告》，2026年10月24日，公告编号2026-042；《关于召开2026年第三季度业绩说明会的公告》，2026年10月17日，公告编号2026-037；上海证券交易所上证路演中心《寒武纪2026年第三季度业绩说明会》互动问答记录（45条回复），2026年10月26日15:00-17:07。单季数据由上述三季报累计数相减得出：单季营业成本=前三季营业成本41.41亿-半年报营业成本26.83亿=14.58亿，单季毛利率=(35.31-14.58)/35.31=58.69%；单季经营现金流=-0.84亿为三季报利润表直接列示的"本报告期"数。成交额、涨跌幅、换手率与全市场排名取自2026年10月27日收盘全市场快照（5570只）；AMD、TSMC、NVDA为2026年10月27日美东盘中及10月26日收盘价，10月26日收盘价=10月27日现价减当日涨跌额。
These 7.7 billion did not go head-to-head with AMD—the timing simply does not line up at all; it followed Monday's 13.4 billion in buying, and the 11.665 billion of inventory in its own warehouse.
First, let's correct the three premises in the question.
First, "not a single announcement" does not hold. On October 24 (Saturday), Cambricon released 8 announcements in one go, three of which were key: the third-quarter report, the provision for asset impairment, and the announcement on the company's recent production and operating conditions. Monday, October 26, was the first trading day after the third-quarter report; it rose 3.84%, with turnover of 13.424 billion, and intraday hit 985 yuan. So it was not "no news"—the news was bought as good news once on Monday, and re-read once on Tuesday.
Second, "the single largest amount of money in the whole market that day" does not hold. The 7.711 billion ranked sixth across the entire market on October 27. Ahead of it were ChangXin Technology 15.109 billion, Zhongji Innolight 14.011 billion, Eoptolink 11.433 billion, CATL 8.814 billion, and WuXi AppTec 8.577 billion. Within semiconductors it also ranked only second, first being ChangXin Technology. And compared with itself, Monday was 13.424 billion, Tuesday was 7.711 billion, only 57% of the previous day.
Third, "going head-to-head with AMD" is impossible in time. The A-share market closed at 15:00 on October 27, while the U.S. market only opened at 21:30 that day, a gap of six and a half hours. The overnight information visible in the A-share market that morning was the U.S. market of October 26: AMD closed at 531.28 dollars, up 2.80%; TSMC closed at 424.37, up 1.40%; Nvidia closed at 216.00, up 2.39%. The overnight was warm. The AMD up 5.26% (559.20 dollars) you saw was the price at 15:31 Eastern time during the October 27 U.S. session; that gain appeared six and a half hours after Cambricon had already finished falling, so it cannot possibly be the cause.
So what exactly happened on Tuesday—laying it out hour by hour makes it clear.
Friday, October 23: the board approved the third-quarter report and the impairment proposal. Saturday, October 24: 8 announcements posted online. Monday, October 26: share price +3.84%. From 15:00 to 17:07 on October 26: post-close earnings briefing, 45 responses. Tuesday, October 27: -5.68%, 7.711 billion, shrinking volume.
The briefing was held after the close, so Monday's bullish candle was the price set before the briefing, and Tuesday's drop was the price set after the briefing. This timeline is harder than any "capital game" narrative.
Next, check whether Cambricon was falling alone on Tuesday. It was not. Shanghai Composite -1.33%, CSI 300 -1.32%, STAR 50 -4.81%, across the whole market 606 rose and 4,922 fell. The 200 semiconductor stocks fell 4.56% on average. What was red that day were banks, +1.34%. Cambricon fell 5.68%, only 0.87 points more than the STAR 50.
More important is that someone in the same industry rose. ChangXin Technology, listed on July 27, had turnover of 15.109 billion on October 27, turnover rate 7.65%, and closed up 0.80%. The money in the sector did not leave; it stayed somewhere else—moving from the AI chip leader that had been talked about for two years to a new stock listed just three months ago. These 7.7 billion were not knocked out by anyone; no one was there to catch them.
Now the statements. Breaking out the three single quarters (subtracting cumulative figures; for the first-quarter report use the cumulative value directly), not a single line in the income statement is bad:
First quarter: revenue 2.885 billion / gross margin 54.33% / net profit attributable to parent 1.013 billion
Second quarter: revenue 3.111 billion / gross margin 56.10% / net profit attributable to parent 1.298 billion
Third quarter: revenue 3.531 billion / gross margin 58.69% / net profit attributable to parent 1.404 billion
Third-quarter gross margin 58.69%, rising for three consecutive quarters (54.33%, 56.10%, 58.69%), 4.45 points higher than 54.24% in the same period last year; single-quarter revenue 3.531 billion and net profit attributable to parent 1.404 billion are both the highest quarter since the company's listing. Revenue +104.5% year-on-year, +13.50% quarter-on-quarter (this quarter-on-quarter figure was given by the company itself at the briefing; the second quarter was +7.84%). Single-quarter net margin fell from 41.7% to 39.8%, and the reason for the fall is not impairment either—third-quarter single-quarter asset impairment was 105 million, less than the second quarter's 151 million—it is R&D expenses rising from 378 million in one quarter to 523 million, an increase of 144 million, or 38.1%.
So the statement "the financial report got worse" is wrong. What got worse is not the ability to make money, but asset turnover.
From the first quarter to the third quarter, these lines on the balance sheet moved as follows:
Inventory 4.497 billion → 8.248 billion → 11.665 billion
Prepayments 1.897 billion → 2.914 billion → 3.044 billion
Notes payable 750 million → 3.085 billion → 4.966 billion
Contract liabilities 396 million → 188 million → 141 million
Single-quarter operating cash flow +834 million → -523 million → -84 million
Three sentences can explain what is going on here.
First, 11.665 billion yuan of inventory, calculated at the third quarter's single-quarter operating cost of 1.458 billion yuan, means eight quarters of cost are sitting in the warehouse; while the money customers paid upfront (contract liabilities) is only 141 million yuan, less than four days of revenue. Inventory is up 136% from 4.944 billion yuan at the start of the year. It's not strange for a profitable company to build up stock, but building up to eight quarters of cost while customer prepayments are less than four days of revenue—that combination is only dared when capacity is at its tightest.
Second, this pile of goods wasn't built with cash; it was built with notes. At the third quarter, notes payable plus accounts payable totaled 6.793 billion yuan, up 2.632 billion yuan from 4.161 billion yuan at the end of the second quarter; monetary funds were down to only 555 million yuan, with 3.811 billion yuan in wealth management products. The company borrowed not a single cent of short-term bank loans; it used notes (bank acceptance bills, typically with maturities of no more than six months) to swap for a batch of inventory that takes eight quarters to turn over.
Third, impairment has already begun. In the first three quarters, 502 million yuan of inventory write-downs were provisioned, 5.77 times last year's same-period 87 million yuan. The company's own words were "mainly because the company carried out strategic stockpiling of raw materials in the earlier period, and as the inventory age of raw materials grew, the corresponding inventory write-down provisions increased." This sentence is the most informative one in the entire document: strategic stockpiling has a shelf life; it is not free.
So why didn't the briefing save it? Because two hours and 45 Q&As did not add a single usable number. Asked about inventory, the answer was a verbatim copy of the announcement; asked about procurement plans, the answer was "the company formulates procurement plans based on customer order demand and forecasts of future market demand," without a single figure on orders in hand; asked about the "institutional exchange minutes" circulating in the market, the company answered "the company has never conducted such communication" and said it belonged to "information of unknown origin or unverified information." The only new information in the entire session was two items: third-quarter quarter-on-quarter +13.50%, and the first three quarters' revenue composition does not include the intelligent computing cluster system business.
The second point is actually an underappreciated piece of good news. The cluster business is about building intelligent computing centers for local governments and enterprises, with long payment cycles, low gross margins, and dependence on fiscal spending. Not a single piece of the 9.526 billion yuan in revenue comes from it; it is all the cloud product line, which the company says mainly serves various types of commercial customers—meaning the quality of growth is cleaner than the market thinks.
So what is the bull case's strongest argument? Exactly the two points above: gross margin rising for three consecutive quarters, revenue quarter-on-quarter still accelerating, operating cash flow narrowing from -523 million yuan in the second quarter to -84 million yuan in the third quarter, and no intelligent computing cluster system in revenue. All four are moving in the right direction. A smart person would say: you cite eight quarters of inventory, but that's capacity lock-in; a Fabless company grabbing wafer and packaging/testing capacity has to pay upfront and stock up in advance anyway, and using notes rather than cash to stock up shows it still has payment terms with upstream; when this batch of goods is sold at a 58% gross margin, 11.6 billion is ammunition, not a burden.
This rebuttal holds, and right now no document can disprove it. The third-quarter report only has total inventory; the breakdown into raw materials, work in progress, and finished goods, and the inventory age, must wait for the annual report, so for now we can only stop at "both explanations are plausible."
So what is truly not yet priced in is not "how much it fell," but "when this batch of goods turns into revenue."
The 2026 equity incentive revenue target is 13.5 billion yuan; the first three quarters did 9.526 billion yuan, so the fourth quarter must do 3.974 billion yuan, a further 12.6% increase over the third quarter's 3.531 billion yuan. If achieved, the 11.6 billion yuan of inventory is ammunition, and the second quarter's -523 million yuan cash flow was the cost of expansion; if not achieved, the same batch of raw materials will trigger another write-down in the first half of 2027. This has nothing to do with the share price and everything to do with two numbers: the balance of inventory write-down provisions in the annual report, and the maturity structure of notes payable.
Write three falsification conditions, all hard. One, fourth-quarter single-quarter revenue reaches 3.974 billion yuan or more—the stockpiling thesis holds, and my judgment that "turnover is starting to tighten" is wrong. Two, the year-end inventory write-down provision balance is significantly higher than 502 million yuan—the backlog thesis holds. Three, single-quarter operating cash flow turns positive and inventory declines quarter-on-quarter—indicating the phase of swapping notes for inventory is over, and the judgment must change.
Finally, a note on that day's market environment, to avoid reading too much intent into the 7.7 billion yuan. October 27 was a broad decline on shrinking volume: more than 4,900 stocks fell, the STAR 50 fell 4.81%, money flowed toward banks, and there was no policy or event targeting the computing power sector. I checked the primary announcements and that day's whole-market turnover and could not find any news item that hit Cambricon specifically that day. So the 7.7 billion yuan is closer to a decline where nobody was willing to bid than a deliberate showdown. The real showdown was in those two hours the previous afternoon—the company, holding 9.526 billion yuan in revenue and 3.715 billion yuan in profit, failed to answer "when will the 11.6 billion yuan of inventory turn into revenue." The market cast that vote the next day with a 5.68% decline on shrinking volume.
Source: Cambricon Technologies Corporation Limited, 2026 Third Quarter Report, October 24, 2026 (periodic report, no announcement number); Announcement on the Provision for Asset Impairment for the First Three Quarters of 2026, October 24, 2026, announcement number 2026-038; Announcement on the Company's Recent Production and Operating Conditions, October 24, 2026, announcement number 2026-042; Announcement on Convening the 2026 Third Quarter Results Briefing, October 17, 2026, announcement number 2026-037; Shanghai Stock Exchange SSE Roadshow Center "Cambricon 2026 Third Quarter Results Briefing" interactive Q&A record (45 replies), October 26, 2026, 15:00-17:07. Single-quarter data are derived by subtracting the cumulative figures in the above third quarter report: single-quarter operating cost = operating cost for the first three quarters 4.141 billion yuan - operating cost in the semi-annual report 2.683 billion yuan = 1.458 billion yuan; single-quarter gross margin = (3.531 - 1.458)/3.531 = 58.69%; single-quarter operating cash flow = -0.084 billion yuan is the "current reporting period" figure directly presented in the income statement of the third quarter report. Turnover, price change, turnover rate and whole-market ranking are taken from the whole-market snapshot at the close on October 27, 2026 (5570 stocks); AMD, TSMC and NVDA are intraday U.S. Eastern time prices on October 27, 2026 and closing prices on October 26; the closing price on October 26 = the current price on October 27 minus that day's price change.