Same-day storage segment: US stocks Micron +4.1%, Western Digital +3.1%, SanDisk +5.63%; A-shares Puya +11.38%, Biwin +8.36%, GigaDevice +7.77%. With the same price-increase logic, why is the A-share elasticity two to three times higher?
Same-day storage segment: US stocks Micron +4.1%, Western Digital +3.1%, SanDisk +5.63%; A-shares Puya +11.38%, Biwin +8.36%, GigaDevice +7.77%. With the same price-increase logic, why is the A-share elasticity two to three times higher?
同一天存储环节:美股美光+4.1%、西部数据+3.1%、闪迪+5.63%;A股普冉+11.38%、佰维+8.36%、兆易+7.77%。同样的涨价逻辑,A股弹性凭什么翻两三倍?
The desk answers已核实Walnut, from public filings
That day, the two-to-three-fold moves in A-share storage names were not the same price-increase logic being amplified — they were three different things colliding on the same day: the A-share line saw several company filings in a single week, while the three US storage names had not a single one these past few days. And looking purely at profit elasticity, Micron is fiercer than any A-share name — its quarterly net profit rose tenfold year-on-year, yet the stock rose only 4.1% that day.
Start with the US. Micron's most recent report was September 30: fourth-quarter revenue 542.29 yi dollars, versus 113.15 yi a year earlier; gross margin 86.8%; net profit 377.01 yi dollars, versus 32.01 yi a year earlier. Western Digital just reported its first quarter on October 22: revenue 43.81 yi dollars, up 55% year-on-year, gross margin 57.8%, next-quarter guidance of 51 yi dollars revenue and 60.5% gross margin — the price increase is still being passed through, but that story was told on the 22nd. SanDisk's most recent report was August 5, Fabrinet's was August 17; on October 28 neither had a new filing, and their 5.63% and 7.19% moves simply followed the sector.
What the US market was really pricing that day was Vertiv: quarterly revenue 43.13 yi dollars, up 61% year-on-year, organic orders up more than 100% year-on-year, full-year revenue guidance raised by 6.5 yi dollars, stock up 13.95%. US money was buying "data-center power and cooling," not "storage price increases." The 3% to 5% moves in the three storage names were sympathetic gains within the same AI-hardware move. Lam Research just reported its quarter on October 21: revenue 82.19 yi dollars, gross margin 52.3%, and rose 5.18% on the 28th — likewise an aftereffect of the sector.
On the A-share side, things were advancing within the week. Biwin issued a "recent production and operating conditions" announcement on October 24, in which the company itself wrote that "prices of major storage products such as DRAM and NAND Flash have risen significantly" — this was an announcement specifically responding to unusual stock-price movement, with no numbers and no guidance, but it turned the price increase from a research note into a company disclosure. Puya's merger and restructuring passed review on October 15 and its registration draft was posted on October 17: 2.47 yi to buy 49% of Noah Changtian, whose underlying asset is SkyHigh Memory, which makes 2D NAND. GigaDevice posted routine announcements such as buyback progress that day; its 7.77% was basically pure sector. Biwin's third-quarter report was only posted after the close on October 28; that day's 8.36% was not driven by the earnings report but by the announcement four days earlier confirming the price increase.
Below that is the base effect. Biwin's third-quarter single-quarter revenue was 61.42 yi, net profit 16.02 yi, up 525.34% year-on-year; a year earlier it was 26.63 yi revenue and 2.56 yi net profit (estimated), with net margin rising from 9.6% to 26.1% (estimated). Lifting net margin by one point on 61 yi of revenue is 0.61 yi of profit, which is 3.8% of 16 yi of profit (estimated); on Micron's side, revenue is 542.29 yi dollars and net margin about 69.5% (estimated), so one point is 5.42 yi dollars, only 1.4% of 377 yi dollars of profit (estimated). Same magnitude of price increase, different percentage impact on per-share profit. This is arithmetic, not skill.
What really determines elasticity is what it uses to meet the price increase. At the end of the third quarter, Biwin had inventory of 240.53 yi (78.68 yi at the start of the year), prepayments of 73.75 yi (1.55 yi at the start of the year), and operating cash flow of -100.41 yi for the first three quarters; first-three-quarter operating cost was 105.51 yi and revenue 217.17 yi, giving a gross margin of 51.4% (calculated). This table shows: before the price increase, it converted cash into wafers, then sold at the new price. Micron's 86.8% gross margin is the result of owning its own fabs and deciding its own capacity; Biwin's 51.4% is bought, earning the time gap between "bought before the price rose" and "sold after the price rose." Both things are called gross margin on the statements, but their quality is not the same.
The trading layer is the same. A 20% price limit, retail-driven pricing, a small free float — the same event plays out in one day in A-shares but takes a month in the US. This is not fundamentals, but it does determine how many points it moves that day.
The strongest opposing argument: of the high profits at these three A-share names, how much is inventory price differential rather than pricing power? Biwin has 314 yi (inventory plus prepayments) tied up in wafers and inventory, five times single-quarter revenue. Foundries can take back this gain at any time in the next round of price increases; module makers are just pass-through hands. Once prices flatten quarter-on-quarter, the same 314 yi turns from appreciation to impairment, and profit falls as fast as it rose — so-called elasticity is two sides of the same position.
What would count as me being wrong: if Biwin's annual report shows full-year gross margin above the first-three-quarter 51.4% (meaning the fourth quarter alone was still rising), while inventory stops increasing and operating cash flow turns from negative to positive, that would show its profit comes from sustained selling-price pass-through and turnover improvement, not the inventory time gap I described, and this explanation would be wrong. Conversely, if inventory is still rising while gross margin starts moving sideways, that is what the time gap looks like when it is almost used up.
Sources:
Vertiv 8-K, 2026-10-28, 0001628280-26-067707, EX-99.1 (q32026exhibit991vrt10282026.htm)
Micron 8-K, 2026-09-30, 0000723125-26-000018, EX-99.1 (a2026q4ex991-pressrelease.htm)
Western Digital 8-K, 2026-10-22, 0001628280-26-066882, EX-99.1 (a4ex991-pressreleaseq127.htm)
Lam Research 8-K, 2026-10-21, 0000707549-26-000052, EX-99.1 (lrcx_exhibitx991xq1x2027.htm)
SanDisk 8-K, 2026-08-05, 0001628280-26-053346 (latest earnings announcement before 10/28)
Fabrinet 8-K, 2026-08-17, 0001408710-26-000026 (latest earnings announcement before 10/28)
Biwin Storage "Announcement on the Company's Recent Production and Operation Status", 2026-10-24, Announcement No. 2026-100
Biwin Storage "2026 Third Quarter Report", disclosed 2026-10-29
Puya Semiconductor "Summary of the Report on the Acquisition of Assets by Issuing Shares, Convertible Corporate Bonds and Paying Cash and Raising Matching Funds (Registration Draft)", 2026-10-17; "Announcement on Passing the Review of the M&A and Restructuring Review Committee of the Shanghai Stock Exchange", 2026-10-15
那天存储环节A股的两三倍,不是同一个涨价逻辑被放大,是三件不同的事撞在同一天:A股这条线一周内连着出了几份公司文件,美股存储三家这几天一份都没有。而单看利润弹性,美光比A股任何一家都猛——它一个季度净利同比涨了十倍,那天股价只涨4.1%。
先看美股。美光最近一份财报是9月30日:第四财季收入542.29亿美元,去年同期113.15亿;毛利率86.8%;净利377.01亿美元,去年同期32.01亿。西部数据10月22日刚报完第一财季,收入43.81亿美元、同比增55%,毛利率57.8%,下季指引51亿美元收入、60.5%毛利率——涨价还在传,但这话22日说完了。闪迪最近一份财报是8月5日,Fabrinet是8月17日,10月28日这两家都没有新文件,5.63%和7.19%是跟着板块走的。
那天美股真正在定价的是维谛技术:单季收入43.13亿美元、同比增61%,有机订单同比增超100%,全年收入指引上调6.5亿美元,股价13.95%。美股的钱在买“数据中心用电、散热”,不是“存储涨价”。存储那三家的3%到5%,是这波AI硬件同向里的跟涨。泛林10月21日刚报完季,收入82.19亿美元、毛利率52.3%,28日涨5.18%,同样是板块余波。
A股这边,一周内有东西在推进。佰维10月24日发了一份“近期生产经营情况”公告,公司自己写“DRAM、NAND Flash等主要存储产品价格显著上涨”——这是专门回应股价异动的公告,没有数字也没有指引,但把涨价从研报变成了公司披露。普冉10月15日并购重组过会、10月17日挂出注册稿,2.47亿买诺亚长天49%股权,标的底下就是做2D NAND的SkyHigh Memory。兆易那天挂的是回购进展这类例行公告,7.77%基本是纯板块。佰维的三季报是10月28日收盘后才挂出来的,那天的8.36%不是财报推的,推它的是四天前那份确认涨价的公告。
再往下是基数。佰维三季度单季收入61.42亿、净利16.02亿,同比增525.34%;去年同期是26.63亿收入和2.56亿净利(估算),净利率从9.6%抬到26.1%(估算)。同样抬一个点的净利率,在61亿收入上是0.61亿利润,对16亿的利润是3.8%(估算);美光那边收入542.29亿美元、净利率约69.5%(估算),抬一个点是5.42亿美元,对377亿美元利润只有1.4%(估算)。涨价幅度一样,落到每股利润上的百分比不一样。这是算术,不是本事。
真正决定弹性的,是它拿什么去接涨价。佰维三季度末存货240.53亿(年初78.68亿)、预付款项73.75亿(年初1.55亿)、前三季经营现金流-100.41亿;前三季营业成本105.51亿、收入217.17亿,毛利率51.4%(算出来)。这张表说明:它在涨价之前把现金换成了晶圆,再按新价出手。美光86.8%的毛利率是自己有厂、自己决定产能的结果;佰维的51.4%是买来的,赚的是“买的时候还没涨”和“卖的时候已经涨了”之间的时间差。两件事在报表上都叫毛利率,含金量不一样。
交易层面也是。20%的涨跌幅、散户定价、流通盘小,同一件事A股一天走完,美股要走一个月。这不属于基本面,但确实决定了当天涨几个点。
反方最强的一条:A股这三家的高利润里,有多少是库存差价而不是定价权?佰维把314亿(存货加预付款)压在晶圆和库存上,是单季收入的五倍。晶圆厂随时能在下一轮涨价里收回这块收益,模组厂只是过手。价格一旦环比走平,同样的314亿就从增值变成减值,利润掉起来和涨起来一样快——所谓弹性,是同一个头寸的两面。
什么情况算我看错:如果佰维年报里全年毛利率高于前三季的51.4%(等于四季度单季还在往上走),同时存货不再增加、经营现金流由负转正,那说明它的利润来自持续的售价传导和周转改善,不是我说的囤货时间差,这条解释就错了。反过来,存货还在涨、毛利率却开始横着走,那就是时间差快用完的样子。
来源:
维谛技术8-K,2026-10-28,0001628280-26-067707,EX-99.1(q32026exhibit991vrt10282026.htm)
美光8-K,2026-09-30,0000723125-26-000018,EX-99.1(a2026q4ex991-pressrelease.htm)
西部数据8-K,2026-10-22,0001628280-26-066882,EX-99.1(a4ex991-pressreleaseq127.htm)
泛林8-K,2026-10-21,0000707549-26-000052,EX-99.1(lrcx_exhibitx991xq1x2027.htm)
闪迪8-K,2026-08-05,0001628280-26-053346(10/28前最新业绩公告)
Fabrinet 8-K,2026-08-17,0001408710-26-000026(10/28前最新业绩公告)
佰维存储《关于公司近期生产经营情况的公告》,2026-10-24,公告编号2026-100
佰维存储《2026年第三季度报告》,2026-10-29披露
普冉股份《发行股份、可转换公司债券及支付现金购买资产并募集配套资金报告书摘要(注册稿)》,2026-10-17;《获得上交所并购重组审核委员会审核通过的公告》,2026-10-15
## 钱停在哪一层
存储这条链是:晶圆和颗粒 → 模组 → 渠道 → 终端。美光、西数、闪迪卖颗粒;佰维、江波龙、德明利买颗粒做成内存条、固态盘、嵌入式存储;兆易、普冉、君正是设计公司,买代工产能做 NOR、利基 DRAM、2D NAND 这些小品类。
三者的毛利率差得很远:美光 86.8%,闪迪 84.6%,西数 57.8%;佰维单季 42.8%,德明利 32.9%;设计公司里兆易二季度 66.6%、君正三季度 65.9%。
越靠近晶圆越高。这不是巧合,是"谁手里有产能"的直接标价。
## 卡点在原厂,原厂还在锁产能
闪迪第四财季收入环比涨 51%,公司自己拆过:约三分之二来自价格,三分之一来自出货量。它在提价,不在放量。
同一个财季里它又签了 5 份新商业模式协议,累计 10 份;美光把"战略客户协议"写进了 CEO 的表态里。西数给下季的指引是收入 51 亿美元、毛利率 60.5%。
原厂三家给下季的毛利率指引都还站在高位:美光 85.95%,闪迪 83% 到 85%,西数 60.5%、比本季的 58.1%(非 GAAP)还高 2.4 个点。没有一家给出掉头向下的指引。
模组厂拿不到增量产能,只能在存量里抢料。所以同一个涨价,原厂是收钱的一方,模组是付钱的一方。
## 三家A股,压的不是同一样东西
佰维是模组加封测。三季度末存货 240.53 亿,是当季收入的 3.9 倍;预付款 73.75 亿,年初只有 1.55 亿,三季度单季就加了 52.8 亿。它赚的是"买的时候还没涨、卖的时候已经涨了"。
预付款这一项值得单独看:钱先付出去,料后到,价格在付款时就谈定了。所以三季度这 73.75 亿,买的是四季度的成本,价已经锁在合同里。
兆易是设计公司,二季度末存货 41.84 亿,只有当季收入的 0.57 倍。它赚的是售价涨得比代工成本快:二季度收入环比 +76.2%,营业成本只 +37.2%。
普冉也是设计公司(存货 0.85 个季度),但它在 10 月 28 日之前多了一样东西——10 月 15 日过会、17 日挂注册稿的 2.47 亿元收购,买回诺亚长天 49% 少数股权。标的是 2D NAND(SLC、eMMC、MCP),增值率 2.04%,没有业绩承诺,配套募资 0.77 亿。
这是事件落地,不是涨价弹性。三家的涨幅被同一个标签扫到一起,生意却不在同一条链的同一环上。
## 三季度的成本追上来了
| 公司(最新季) | 单季收入(亿) | 环比(%) | 单季毛利率(%) | 毛利率环比(pct) |
|---|---|---|---|---|
| 佰维(2026Q3) | 61.42 | (29.9) | 42.81 | (13.17) |
| 德明利(2026Q3) | 116.45 | 24.3 | 32.89 | (5.95) |
| 江波龙(2026Q2) | 141.80 | 43.1 | 61.26 | 5.73 |
| 兆易(2026Q2) | 73.78 | 76.2 | 66.57 | 9.49 |
| 普冉(2026Q2) | 25.13 | 73.7 | 54.87 | 10.19 |
| 君正(2026Q3) | 31.45 | 31.2 | 65.93 | 8.81 |
把六家放进同一张横截面,差异就出来了:两家已经出了三季报的模组厂,毛利率都环比掉了。佰维单季营业成本环比只降 8.9%,收入却降了 29.9%;德明利收入环比 +24.3%,成本 +36.4%。
佰维收入掉三成、成本只掉 8.9%;德明利成本涨 36.4%、收入只涨 24.3%——两边指向同一件事:新进的料已经是涨价后的价。二季度的 55.98%,来自那批还没涨价的库存。
这条传导链的时滞也在这里:晶圆涨价先反映到售价,再反映到营业成本,中间隔着一批库存。涨价时它送利润,涨价停下时它送亏损,同一批货,方向反过来而已。
```chart
type: line
title: 佰维存储单季毛利率:两个季度冲到56%后回落13个点
unit: %
labels: 2025Q4, 2026Q1, 2026Q2, 2026Q3
values: 31.91, 53.30, 55.98, 42.81
```
同一时间,原厂给的毛利率指引还都停在 60% 到 86% 的高位。涨价的钱停在上一环,没有流到模组。
设计公司这条支线暂时没被挤:君正三季度收入环比 +31.2%,营业成本只 +4.2%,毛利率还在升。
所以准确的排序不是"A股和美股",是原厂、设计、模组——模组在最下面,产能最松,涨价末期最先被挤压。
## 8.36%买的不是三季度
佰维的三季报是在 10 月 28 日收盘之后才披露的(公告日期 10 月 29 日)。白天那 8.36%,定价的是 10 月 24 日那份"生产经营情况"公告,和二季度 55.98% 的毛利率。
当晚的季报是另一组数:单季收入 61.42 亿、净利 16.02 亿,同比 +525.34%,环比 -62.5%。
同比还在翻五倍,环比已经腰斩。
这里还有一层时间差:美股三家的季报分别在 8 月 5 日、9 月 30 日、10 月 22 日就说完了,10 月 28 日没有新文件;A 股这边的三季报正挤在这几天落地,兆易和普冉的都还没出。同一个时点上,一边是已经兑现、已经被反复定价的利润,一边是还没被财报检验的预期。
再往下还有一层更机械的:佰维和普冉在科创板,一天能走 20%;兆易在主板,10% 封顶;盘子也比美股那三家小得多。同一件事,A 股常常一天走完,美股要分几天消化。这一层不改变生意,只改变当天的数字——但当天问的,就是当天的数字。
"弹性"的完整含义就在这里:它的分母是去年同期那个几乎不赚钱的基数——2025 年三季度净利只有 2.56 亿、净利率 9.6%;它的对照是上一个季度的 42.67 亿。同一天买进去的人,买的是前一个季度。
## 反方最强的一条
最硬的反驳是:42.81% 仍然是去年同期的两倍——2025 年三季度佰维单季毛利率只有 21.02%。拿 55.98% 当基准,本身就是在最高点上划线。
这条我接一半:我没说它会跌回去年,我说的是拿二季度的毛利率去推三、四季度的利润,方向是错的。
第二条:佰维前三季 AI 端侧存储收入 48.23 亿、同比 +389.64%,占收入 22.2%,客户里有 Meta、Google、阿里、小米。如果三季度毛利率下滑是产品结构变了,而不是料价追上,我的解释就只对一半。这条要等分产品数据才能分开。
第三条:三季度收入掉三成,也可能只是二季度客户提前拉货造成的交付节奏,不是需求掉了。
## 什么情况算我看错
- 佰维四季度单季毛利率回到 50% 以上、存货环比不再增加、经营现金流转正:三季度是节奏问题,不是成本问题。
- 兆易、普冉的三季报(本月底前)单季毛利率继续环比上升:那就不是"模组最先被挤",而是"只有模组被挤"。
- 原厂把下季毛利率指引明显调低(美光跌破 84%、闪迪跌破 82%、西数跌破 58%):涨价这个前提没了,上面全部作废。
这三条就是这条判断的证伪条件。任何一条成立,我都得改口,不能拿"周期还没走完"糊过去。
## 这笔账还没开始算
佰维三季度末 240.53 亿存货,前三季只计提了 0.968 亿存货跌价,占存货 0.40%。
意思是:到 9 月 30 日,绝大部分存货的成本仍低于可变现净值,这笔账只动了 0.4%,几乎还没开始赔。
这笔账不在毛利率里,在"资产减值损失"这个科目上。毛利率只记录已经卖出去的那部分,剩下的 240 亿要等价格走平才结账。
弹性从来不是"涨得多",是"押得多"。押的是货,赚的是时间差,赔的也是同一个时间差。
来源:
美光科技 Form 8-K,2026-09-30,0000723125-26-000018,EX-99.1(a2026q4ex991-pressrelease.htm)
西部数据 Form 8-K,2026-10-22,0001628280-26-066882,EX-99.1(a4ex991-pressreleaseq127.htm)
闪迪 Form 8-K,2026-08-05,0001628280-26-053346,EX-99.1(sndkq4-26ex991xpressrelease.htm)
佰维存储《2026年第三季度报告》,2026-10-29披露
佰维存储《关于2026年前三季度计提资产减值准备的公告》,2026-10-29,公告编号2026-101
佰维存储《关于公司近期生产经营情况的公告》,2026-10-24,公告编号2026-100
兆易创新《2026年半年度报告》,2026-08-19披露(截至10月28日三季度报告未披露)
普冉股份《发行股份、可转换公司债券及支付现金购买资产并募集配套资金报告书摘要(注册稿)》,2026-10-17;《获得上交所并购重组审核委员会审核通过的公告》,2026-10-15
德明利《2026年第三季度报告》,2026-10-26披露
北京君正《2026年第三季度报告》,2026-10-28披露
江波龙《2026年半年度报告》,2026-08-11披露
普冉股份《2026年半年度报告》,2026-08-19披露
佰维存储《2026年半年度报告》,2026-08-25披露
德明利《2026年半年度报告》,2026-08-28披露