A股做存储模组的德明利跌7.64%、成交53.8亿却没有公告，同一天美股西部数据涨6.85%、闪迪涨3.9%，存储涨价只…
- 前三季度收入 285.55 亿元、成本 167.56 亿元，毛利率 41.3%
- 毛利率的走向才是关键，得自己倒算：上半年收入 169.11 亿元、营业成本 89.42 亿元，毛利率 47.1%
- 下季度公司自己给的指引是收入 51 亿美元、毛利率 60% 到 61%
- 净利 27.29 亿元，同比涨 2902.75%
- 财务费用 5.56 亿元，同比涨 331.70%
A股做存储模组的德明利跌7.64%、成交53.8亿却没有公告，同一天美股西部数据涨6.85%、闪迪涨3.9%，存储涨价只传到美国去了？
On the A-share market, Demingli, which makes storage modules, fell 7.64% with a turnover of 5.38 billion without any announcement, while on the same day in the US market Western Digital rose 6.85% and SanDisk rose 3.9% — has the storage price increase only spread to the US?
The desk answers已核实Walnut, from public filings
结论：涨价没传丢，只是传到的位置不一样——西部数据卖的是产能，德明利买的是颗粒；而且德明利早就吃到了这波涨价，跌的这一天，是它利润率往下走的第一个季度被市场看见。
先看美国那头的一手文件。西部数据 10 月 22 日的季报（2027 财年第一季，截至 10 月 2 日）：收入 43.81 亿美元，同比涨 55%；毛利率 57.8%，去年同期 43.5%；下季度公司自己给的指引是收入 51 亿美元、毛利率 60% 到 61%。财报里 CFO 的原话是出货量和价格同时往上（robust exabyte growth and improved pricing）。硬盘这个行业全球就剩三家在供货，AI 数据中心要存下来的东西又必须落在盘上，所以它不是被动接受价格，它是在定价格。
德明利这边，三季报 10 月 26 日就发了（公告编号 2026-093）。当天没有新公告，但两天前的这份东西就是全部信息。第三季度单季收入 116.45 亿元，同比涨 356.64%；净利 27.29 亿元，同比涨 2902.75%。这个数字摆出来，"涨价没传到中国"是站不住的。
毛利率的走向才是关键，得自己倒算：上半年收入 169.11 亿元、营业成本 89.42 亿元，毛利率 47.1%；前三季度收入 285.55 亿元、成本 167.56 亿元，毛利率 41.3%；两个成本相减，第三季度单季成本约 78.15 亿元，对应单季毛利率约 32.9%（估算，公司没有单独披露单季成本）。净利率更直白：上半年 35.6%（估算），第三季度 23.4%。收入创新高，利润率从 35.6% 掉到 23.4%。
这就是模组这门生意的结构。它赚的不是加工费，是"买进来便宜、出手时贵"的这段时间差。上半年颗粒涨价，它手上是低价库存、卖的是高价，毛利率冲到 47.1%；涨到第三季度，补库存的价格追上来，售价涨幅跟不上，毛利率掉到 32.9%。三季报把这层写得很清楚：存货 289.86 亿元，年初 70.58 亿元；经营活动现金流 -136.91 亿元；短期借款 101.50 亿元（年初 34.72 亿元），长期借款 91.00 亿元（年初 9.52 亿元）；财务费用 5.56 亿元，同比涨 331.70%。翻译一下：它是借钱把颗粒囤起来，赌价格继续往上。价差扩，利润爆；价差收窄，利润掉；价差反过来，这 289.86 亿元存货就是最重的那块石头。市场卖的不是"没涨价"，是"这一层的钱已经赚在前面了"。
把这条链摊开，涨价是分三段落地的：颗粒原厂，涨价当季直接变成收入和毛利率；模组厂，先靠手上的低价库存吃一段价差，等补库存的价格追上来，涨价就翻成成本；下游整机，从头到尾只承担成本。德明利站在第二段，所以它的好日子比原厂晚一个季度到，也比原厂早一个季度走。同一件事在三段上方向相反，问"传到没传到"就问错了——该问的是传到了哪一段。
A 股那天跌的也不止存储。同一批票里：星宸科技 -14.56%、行云科技 -14.21%、联特科技 -13.71%、芯碁微装 -10.40%、协创数据 -10.21%、华正新材 -10.00%、本川智能 -9.30%，成交从 4.4 亿元到 40 亿元。这是整条 AI 硬件——光模块、覆铜板、设备、模组——在十月底出完三季报之后一起被卖，不是存储一个子行业的事。协创数据 10 月 27 日同一天发了两份东西：三季报，和一份计提减值公告（公告编号 2026-159），第三季度计提应收账款坏账 1.72 亿元、存货跌价 3501.28 万元，合计让利润总额少了 2.01 亿元。市场看到的不是"涨价"，是"囤货的公司开始计提"。
美股那边同样不是存储一个板块在涨。那天涨最猛的艾马克技术 +8.56%，是 10 月 26 日盘后的三季报：收入 19.91 亿美元，和去年同期 19.87 亿美元几乎一样，毛利率从 14.3% 升到 18.3%，计算类收入占比从 19% 升到 26%，通信类从 51% 降到 37%。它赚的是算力封装的订单结构，跟颗粒价格没有关系。Coherent 最近一份业绩公告是 8 月 12 日，Lumentum 是 8 月 11 日，Fabrinet 是 8 月 17 日，闪迪最近一份 8-K 是 9 月 16 日的高管变动——这几家那天是跟着板块走，不是自己出了数。
所以"同一天 A 股跌、美股涨"不是地理问题。涨的那一端是能定价的：硬盘原厂、封装、算力硬件，最新的财报刚证明它们能把成本推出去；跌的那一端是刚交完卷的：A 股 AI 硬件在十月底集中披露，利润率、存货、减值三样一起被翻出来看。位置不同，不是国界不同。
反方最强的一条：德明利并不是被价格压着走的那一方。公司自己在业绩预告里写"第三季度营业收入环比持续增长"，说明提价能转嫁出去；289.86 亿元存货是按更低的价钱买进来的，这部分价差要到四季度结算才真正体现，所以三季度毛利率回落可能只是季度节奏，我把一个季度当成了趋势。另外利润里压着一次性东西：前三季度股份支付 2.81 亿元，财务费用 5.56 亿元里有一块是汇兑损失，这两样不代表经营能力变差。如果四季度这两项不再扩大，毛利率自己会回来。
什么情况算我看错：看 2026 年年报的两个数。第一，四季度单季毛利率（用全年减前三季度倒算，估算）如果回到 40% 以上；第二，年报里存货跌价准备余额接近零、没有明显计提。这两条同时出现，说明三季度只是费用和节奏问题，我说的"价差收窄"就是错的。反过来，如果年报存货跌价准备明显计提，或者经营现金流继续大额为负而收入不再增长，这个判断才算被证实。
来源：
西部数据（WDC），a4ex991-pressreleaseq127.htm，EX-99.1 财报新闻稿《WD Reports Fiscal First Quarter 2027 Financial Results》，2026-10-22，accession 0001628280-26-066882。
艾马克技术（AMKR），amkr9302026erex-991.htm，EX-99.1 财报新闻稿《Amkor Technology Reports Financial Results for the Third Quarter 2026》，2026-10-26，accession 0001047127-26-000049。
德明利，《2026 年第三季度报告》，公告编号 2026-093，2026-10-26。
德明利，《2026 年半年度报告》，2026-08-28（上半年收入与营业成本出处）。
德明利，《2026 年前三季度业绩预告》，公告编号 2026-092，2026-10-17。
协创数据，《关于计提 2026 年第三季度信用减值损失及资产减值损失的公告》，公告编号 2026-159，2026-10-27。
SEC EDGAR 申报清单，CIK 820318（Coherent）、1633978（Lumentum）、1408710（Fabrinet）、2023554（Sandisk），核查截至 2026-10-28 各自最近一份定期报告与 8-K。
Conclusion: the price increase was not lost in transmission—only the point at which it lands differs: Western Digital is selling capacity, Demingli is buying chips; and Demingli had already captured this wave of price increases long ago. The day it fell was the first quarter in which its declining profit margin was seen by the market.
First, the primary documents from the US side. Western Digital's quarterly report on October 22 (fiscal 2027 first quarter, ended October 2): revenue USD 4.381 billion, up 55% year-on-year; gross margin 57.8%, versus 43.5% in the same period last year; the company's own guidance for next quarter is revenue USD 5.1 billion and gross margin 60% to 61%. In the financial report, the CFO's exact words were that both shipment volume and price moved up (robust exabyte growth and improved pricing). In the hard drive industry, only three companies worldwide remain supplying, and what AI data centers need to store must land on disks, so it is not passively accepting prices—it is setting prices.
On Demingli's side, the third-quarter report was released on October 26 (announcement no. 2026-093). There was no new announcement that day, but this document from two days earlier is all the information. Third-quarter single-quarter revenue was 11.645 billion yuan, up 356.64% year-on-year; net profit was 2.729 billion yuan, up 2902.75% year-on-year. With these numbers on the table, "the price increase was not transmitted to China" does not hold up.
The direction of gross margin is the key, and it has to be back-calculated: first-half revenue 16.911 billion yuan, operating cost 8.942 billion yuan, gross margin 47.1%; first three quarters revenue 28.555 billion yuan, cost 16.756 billion yuan, gross margin 41.3%; subtracting the two costs gives third-quarter single-quarter cost of about 7.815 billion yuan, corresponding to a single-quarter gross margin of about 32.9% (estimate; the company did not separately disclose single-quarter cost). Net margin is even more direct: first half 35.6% (estimate), third quarter 23.4%. Revenue hit a new high, and the profit margin fell from 35.6% to 23.4%.
This is the structure of the module business. What it earns is not a processing fee but the time spread of "buying cheap and selling dear." In the first half, chip prices rose; it held low-priced inventory and sold at high prices, and gross margin surged to 47.1%; by the third quarter, the cost of restocking caught up, the increase in selling prices could not keep pace, and gross margin fell to 32.9%. The third-quarter report states this layer very clearly: inventory 28.986 billion yuan, versus 7.058 billion yuan at the beginning of the year; operating cash flow -13.691 billion yuan; short-term borrowings 10.150 billion yuan (34.72 billion yuan at the beginning of the year—note: 3.472 billion yuan), long-term borrowings 9.100 billion yuan (952 million yuan at the beginning of the year); financial expenses 556 million yuan, up 331.70% year-on-year. In translation: it borrowed money to stockpile chips, betting that prices would continue to rise. If the spread widens, profit explodes; if the spread narrows, profit falls; if the spread reverses, this 28.986 billion yuan of inventory is the heaviest stone. What the market sold was not "no price increase" but "the money at this layer has already been earned earlier."
Laying out this chain, the price increase lands in three stages: at chip original manufacturers, the price increase directly becomes revenue and gross margin in the same quarter; at module makers, they first rely on low-priced inventory on hand to capture part of the spread, and when the cost of restocking catches up, the price increase turns into cost; at downstream finished devices, they bear cost from beginning to end. Demingli stands in the second stage, so its good days arrive one quarter later than the original manufacturers and leave one quarter earlier than the original manufacturers. The same event moves in opposite directions across the three stages; asking "was it transmitted or not" is the wrong question—what should be asked is which stage it was transmitted to.
On that day in the A-share market, more than storage fell. Among the same batch of stocks: SigmaStar -14.56%, Xingyun Technology -14.21%, Linktel Technologies -13.71%, Chipmore -10.40%, Xiechuang Data -10.21%, Huazheng New Material -10.00%, Benchuan Intelligent -9.30%, with turnover from 440 million yuan to 4 billion yuan. This was the entire AI hardware chain—optical modules, copper-clad laminates, equipment, modules—being sold together after third-quarter reports were out at the end of October, not a matter of one storage sub-industry. Xiechuang Data released two documents on the same day, October 27: its third-quarter report, and an impairment provision announcement (announcement no. 2026-159), with third-quarter provisions of 172 million yuan for accounts receivable bad debt and 35.0128 million yuan for inventory write-down, together reducing total profit by 201 million yuan. What the market saw was not "price increases" but "companies that stockpiled goods beginning to make provisions."
The US stock market is likewise not seeing only the memory sector rise. The biggest gainer that day was Amkor Technology +8.56%, from its Q3 report after the close on October 26: revenue 1.991 billion US dollars, almost the same as 1.987 billion US dollars in the same period last year, gross margin rising from 14.3% to 18.3%, computing revenue share rising from 19% to 26%, communications falling from 51% to 37%. What it earns is the order structure of compute packaging, which has nothing to do with chip price. Coherent's most recent earnings announcement was August 12, Lumentum's was August 11, Fabrinet's was August 17, and Sandisk's most recent 8-K was a September 16 executive change — these companies were following the sector that day, not reporting numbers of their own.
So "A-shares fall, US stocks rise on the same day" is not a geography problem. The rising end is the part that can price: hard drive manufacturers, packaging, compute hardware — their latest financial reports just proved they can pass costs through; the falling end is the part that just handed in its exam: A-share AI hardware disclosed in a concentrated wave at the end of October, with margin, inventory, and impairment all turned over and examined together. Different position, not different national border.
The strongest counterargument: Demingli is not the side being squeezed by price. The company itself wrote in its earnings preview that "third-quarter operating revenue continued to grow quarter-on-quarter," showing that price increases can be passed through; the 28.986 billion yuan of inventory was bought at lower prices, and this price difference will only truly show up when settled in Q4, so the Q3 gross margin pullback may just be quarterly rhythm — I mistook one quarter for a trend. In addition, there are one-off items weighing on profit: share-based payment of 281 million yuan in the first three quarters, and part of the 556 million yuan of financial expenses is exchange loss; these two do not mean operating capability has deteriorated. If these two items no longer expand in Q4, gross margin will come back on its own.
What would count as me being wrong: look at two numbers in the 2026 annual report. First, whether Q4 single-quarter gross margin (back-calculated as full year minus first three quarters, estimated) returns above 40%; second, whether the inventory write-down provision balance in the annual report is close to zero with no obvious provision. If both appear at the same time, it means Q3 was only a cost and rhythm problem, and my "price spread narrowing" is wrong. Conversely, if the annual report shows obvious inventory write-down provisions, or operating cash flow continues to be significantly negative while revenue no longer grows, then this judgment is confirmed.
Sources:
Western Digital (WDC), a4ex991-pressreleaseq127.htm, EX-99.1 earnings press release "WD Reports Fiscal First Quarter 2027 Financial Results," 2026-10-22, accession 0001628280-26-066882.
Amkor Technology (AMKR), amkr9302026erex-991.htm, EX-99.1 earnings press release "Amkor Technology Reports Financial Results for the Third Quarter 2026," 2026-10-26, accession 0001047127-26-000049.
Demingli, "2026 Third Quarter Report," announcement no. 2026-093, 2026-10-26.
Demingli, "2026 Semi-Annual Report," 2026-08-28 (source of first-half revenue and operating cost).
Demingli, "2026 First Three Quarters Earnings Preview," announcement no. 2026-092, 2026-10-17.
Xiechuang Data, "Announcement on Provision for Credit Impairment Losses and Asset Impairment Losses for the Third Quarter of 2026," announcement no. 2026-159, 2026-10-27.
SEC EDGAR filing list, CIK 820318 (Coherent), 1633978 (Lumentum), 1408710 (Fabrinet), 2023554 (Sandisk), checking each one's most recent periodic report and 8-K as of 2026-10-28.
上半年和第三季度直接比是 47.1% 到 32.9%，累计数会掩盖节奏。摊到单季是三个数：一季度 57.42%、二季度 38.84%、三季度 32.90%（一季度收入 75.38 亿、成本 32.10 亿；二季度 93.72 亿、57.32 亿；三季度 116.45 亿、78.15 亿，都由三季报和半年报的累计数相减算出）。三个季度掉了 24.5 个百分点。
但毛利率掉，不等于赚得少。三个单季的毛利额是 43.29 亿、36.40 亿、38.30 亿——三季度毛利额环比还多了 1.90 亿。真正吃掉利润的是这 1.90 亿的去向：单季财务费用从一季度 1.18 亿、二季度 1.46 亿，涨到三季度 2.92 亿（前三季 5.56 亿减上半年 2.64 亿倒算），一个季度多花 1.46 亿。三季度毛利额多出来的 1.90 亿，有 1.46 亿交了利息和汇兑。所以三季度净利 27.29 亿，只比二季度 26.71 亿多 0.58 亿。这家公司这个季度的故事，主要不是"卖得便宜了"，是"借钱变贵了"。
落到报表科目上得说清楚：成本压力全部走在"营业成本"里。单季营业成本环比 +36.4%（57.32 亿到 78.15 亿），收入环比只 +24.3%（93.72 亿到 116.45 亿），成本跑得比收入快 12 个点。不是"税金及附加"（前三季 4344 万），也不是"资产减值"（单季 0.20 亿、前三季 0.63 亿，相对 289.86 亿存货等于没提）。
同一天跌的四家，才是有可比性的同行。10 月 27 日：德明利 -7.64%、香农芯创 -4.15%、佰维存储 -3.65%、江波龙 -2.11%。而前一天 10 月 26 日，德明利反而是这组里跌得最少的（-3.38%，其余三家 -4.77% 到 -4.90%）。只盯一天会得出"德明利出事了"的错觉，两天累计才看得清：四家全跌，德明利跌得最多，和跌得最少的江波龙差 4 个点，不是 30 个点。
那为什么它跌得最多？答案不在利润表，在资金表。同一波涨价，三家存储流通公司的现金流分成了两边：江波龙 10 月 15 日的业绩预告里，一边说"DRAM 和 NAND Flash 价格涨幅较前期明显收窄，存储价格上涨对公司盈利的边际贡献显著趋缓"，一边写"实现第三季度经营性现金流量净额转正的重大跨越，并且实现期末资产负债率低于50%"——涨价放缓它认了，但它把现金流转正、把负债率压到 50% 以下。香农芯创前三季经营活动现金流净额 +54.38 亿。德明利前三季是 -136.91 亿，资产负债率 70.10%，有息负债 192.50 亿（短借 101.50 亿、长借 91.00 亿），货币资金只有 9.51 亿。同一件事，一家在收钱，一家在借钱。
把链条上各层的毛利率摆出来，方向比水平更有信息。闪迪（NAND 原厂）2026 财年第四季毛利率 84.6%，环比 +6.2 个点；西部数据（硬盘原厂）2027 财年第一季 57.8%，环比 +3.7 个点，指引下季 60% 到 61%；德明利（模组）三季度 32.90%，环比 -5.94 个点。原厂在升，模组在降，而且原厂的下季指引还在往上指。跨层比毛利率水平没意义，但方向可以比——同一段时间，价格上升的好处在上游被放大，在下游被成本化。所谓"涨价只传到美国"，实际是涨价在链条两端符号相反：对卖颗粒的人是收入，对买颗粒的人是成本。
时间差是这门生意的全部内容。原厂当季兑现，因为它的售价就是当期现货；模组滞后一到两个季度，因为三季报结转的营业成本对应的是二季度及更早的采购价，而售价是按三季度现货定的。这也解释了同一天一个价格信号为什么两边反向：西部数据 10 月 22 日的指引是下季收入同比 +66% 到 +72%，它交易的是未来两个季度的价格；A 股十月底交易的是已经封账的三季报。一个在定未来的价，一个在结算过去。
已经定进去多少了，时间线摆出来就清楚。10 月 17 日德明利发业绩预告，前三季净利 86.5 亿到 89.5 亿、三季度单季 26.33 亿到 29.33 亿——利润先给了。市场买了：10 月 16 日收盘 206.01 元，10 月 23 日 251.00 元，五个交易日 +21.84%。10 月 26 日三季报落地，第一次给出 289.86 亿存货、-136.91 亿经营现金流、192.50 亿有息负债——钱的来路给了。市场卖：251.00 元到 224.00 元，两天 -10.76%。先买利润，再卖钱的来路，这是同一条信息的两次定价，不是两件事。
反方最强的一条，我替它讲完整：存货按加权平均计价，三季度结转的成本是二季度的采购价，而西部数据指引下季价格还在改善，那么四季度"售价减成本"的差会重新张开，32.90% 就是这轮的低点。支持它的证据是下降的斜率在收敛——一季度到二季度掉 18.58 个点，二季度到三季度只掉 5.94 个点。公司自己在三季报里也把存货增加 310.66% 归因为"主要系公司为企业级存储业务备货所致"，前三季企业级存储收入 82.37 亿（业绩预告），不是纯粹赌价。
这条反方的软肋在合同负债。公司说在手订单快速增长、为企业级业务备货，但合同负债从年初 6.24 亿降到 4.05 亿（-35.06%），公司自己给的原因是"预收客户货款减少"。订单真在快速增长，客户预付款该往上走，不该往下走。同时应收账款 51.37 亿（+610.89%）、应付账款 59.25 亿（+710.56%）、预付款项 1.51 亿（-49.09%）——这三组数凑在一起说明这轮备货的钱从哪来：给客户账期、拿供应商账期、再向银行借。客户没有拿钱来锁货。
什么情况算我看错，三条，都能查。第一，2026 年年报出来，四季度单季毛利率（全年减前三季倒算）回到 40% 以上，且年报存货跌价准备余额接近零——那说明三季度只是费用和节奏，我说的"成本追上售价"是错的。第二，江波龙和佰维存储的三季报（截至 10 月 27 日都没披露，江波龙只有 10 月 15 日预告，佰维 10 月 24 日那份公告里没有一个数字）如果单季毛利率也在 33% 附近、经营现金流同样转负，那"资金表差异"这个解释就站不住，得改成"整个模组层被一起重新定价"——这两份报告十月底之前就会出来。第三，四季度财务费用单季再增 1 亿以上而收入不再环比增长，说明靠负债接涨价这条路已经走到头，这个判断也就被证实。
回到问题本身：涨价没有只传到美国。它 2026 年上半年就在中国的模组厂报表上炸过一次——德明利上半年净利 60.17 亿，江波龙上半年销售净利率 44.49%。第三季度炸的是同一个故事的另一半：为了接住涨价，德明利把存货从 70.58 亿加到 289.86 亿，把有息负债从 44.24 亿加到 192.50 亿，然后在一个季度里多付 1.46 亿利息。市场卖的不是"没涨价"，是"涨价的账单开始由自己付"。同一天美股那两家在涨，因为账单不在它们手上。
来源：
西部数据（WDC），EX-99.1《WD Reports Fiscal First Quarter 2027 Financial Results》，2026-10-22，accession 0001628280-26-066882。
闪迪（Sandisk），EX-99.1《Sandisk Reports Fiscal Fourth Quarter 2026 Financial Results》，2026-08-05，accession 0001628280-26-053346。
德明利，《2026 年第三季度报告》，公告编号 2026-093，2026-10-26。
德明利，《2026 年半年度报告》，2026-08-28。
德明利，《2026 年前三季度业绩预告》，公告编号 2026-092，2026-10-17。
江波龙，《2026 年前三季度业绩预告》，公告编号 2026-117，2026-10-15。
香农芯创，《2026 年第三季度报告》，公告编号 2026-098，2026-10-27。
佰维存储，《关于公司近期生产经营情况的公告》，公告编号 2026-100，2026-10-24。
单季倒算：tushare 利润表（001309.SZ，2026Q1-Q3）；行情：2026-10-27 收盘。