SanDisk fell 13.5% in one day, and weakening NAND prices have been put on the table. If prices do weaken, in Micron Technology's (MU) mid-December quarterly report, will gross margin drop first, or will inventory write-down provisions jump first?
- The next day the stock fell −13.47%, and Micron followed down 5.3%.
- Cost is only 13.2% of selling price
- For estimated selling price to fall to cost, it would have to fall 86%.
SanDisk fell 13.5% in one day, and weakening NAND prices have been put on the table. If prices do weaken, in Micron Technology's (MU) mid-December quarterly report, will gross margin drop first, or will inventory write-down provisions jump first?
闪迪一天跌 13.5%，NAND 价格转弱被摆上台面。如果价格真转弱，美光科技（MU）12 月中旬那份季报里，是毛利率先掉，还是存货跌价准备先跳升？
The desk answersverifiedWalnut, from public filings
Conclusion: the first thing to fall is gross margin, and it does not equal inventory write-downs — in that mid-December quarterly report, gross margin will move from 86.8% toward 85.95%, less than one point, and there will not be a single line of inventory write-down. With Micron's current cost structure, for write-downs to appear, gross margin would first have to fall from 86% all the way to near zero; that is not something "prices turning weak" can do — that is a "price collapse."
Micron's quarter ended September 3 (the 8-K filed September 30) had revenue of 542.29 yi dollars, GAAP gross margin of 86.8%, and cost of revenue of 71.82 yi. Out of every 100 of revenue, 13.2 is cost. The mid-December report covers late August to late November; the guidance the company gave on September 30 was revenue of 615 yi and GAAP gross margin of 85.95%, implying cost of 86.4 yi.
Gross margin equals 1 minus cost divided by revenue. When cost does not move, revenue falling x% makes gross margin fall by (1 minus gross margin) times x%. Right now 1 minus 86% leaves only 0.14, so revenue falling 10% makes gross margin fall 1.4 points. Write-downs run on another track: the estimated selling price of inventory below cost. Cost is only 13.2% of selling price; for estimated selling price to fall to cost, it would have to fall 86%. These two things are an order of magnitude apart, not one quarter apart.
How much NAND accounts for determines the lethality of "turning weak." In the 10-K, FY26 DRAM revenue was 1006.8 yi, NAND 317.9 yi, total 1331.88 yi, with NAND at 23.9%. On that share, a 10% decline across all NAND categories would cut revenue 2.4%, cut gross margin 0.34 points, and reduce gross profit by 14.7 yi dollars. It lands on gross profit, not on the inventory line item.
The inventory structure is even more direct. Note 6: inventory on September 3 was 103.72 yi, finished goods 9.58 yi, work in process 82.52 yi, raw materials 11.62 yi. Finished goods are only 9%, equivalent to 12 days of cost of revenue. A year earlier in the same period, finished goods were 10.94 yi — revenue was already nearly fifteen times the prior crash level (FY23 second quarter 36.93 yi), yet finished goods were lower. The object of impairment is finished goods and work in process, and this company ships what it makes; there is no pile in the warehouse waiting to be revalued at market price.
There is also a lock at the accounting-rule level. The accounting policy in the same 10-K says DRAM, NAND and other memory inventory are combined into one group, and that group is compared against cost and net realizable value. NAND turning weak on its own cannot, under the rules, trigger impairment — as long as DRAM is still making money (Q4 gross margins in core data center and mobile client were both 90%), the combined group's selling price is far above cost, and impairment cannot be calculated. To know when impairment comes, what to watch is not NAND, it is DRAM.
The prior cycle's sequence is fully written in the filings. FY23 second quarter (ended March 2, 2023) gross margin was −32.7%, and that quarter booked 14.30 yi; the third quarter booked another 4.01 yi, 18.31 yi for the full year, with full-year gross margin of −9%. But one quarter before that, gross margin had already fallen from 46.7% to 21.9%, and revenue had fallen from 86.42 yi to 40.85 yi, straight in half, while write-downs were zero. Gross margin fell for a full quarter, nearly 25 points, before the inventory line item moved.
This account has a two-year tail. FY23 booked 18.31 yi; in the same fiscal year, because what was sold was already written-down inventory at lower cost, 8.44 yi was reversed; FY24 reversed another 9.87 yi, exactly offsetting. First a one-time hit, then a slow reversal — a lagged account, not a synchronized one.
How real is "prices turning weak" itself. SanDisk reported after the close on October 29: revenue 110.31 yi, quarter-on-quarter +23%, gross margin 85.0%, next-quarter guidance 84.4% to 85.4%, essentially flat, and on the call the CFO said prices would continue to rise modestly through the year. The next day the stock fell −13.47%, and Micron followed down 5.3%. The only real soft spot in the filing: SanDisk's consumer business was 3.97 yi, quarter-on-quarter −29%, year-on-year −39%, which the company said was the market adapting to higher prices. That is high prices killing low-end demand, still a layer away from contract prices falling. 50% of FY27 bits are already under long-term contracts, and FY28 is two-thirds.
Micron's protection is written in the same 10-K: strategic customer agreements are take-or-pay, multi-year, fixed-price or periodically renegotiated, most with floors and ceilings; the company says that even at the low end of the range, gross margin is clearly above the single-quarter peak of any historical cycle. The historical single-quarter peak was 60.6% in the third quarter of fiscal 2018 (I calculated it from the XBRL quarterly series). In plain terms: the gross margin floor on the contracted volume is above 60%.
In the December report, only two things will move: gross margin shifting from 86.8% to 85.95%, that 0.85 points — this is the guidance given on September 30, not new news; and the guidance in the February quarterly report. If prices really loosen, the order is: first revise guidance, then revise next quarter's revenue, and only last does it reach the inventory line item.
The strongest opposing argument: the December report's gross margin only recognizes goods already shipped from end-August to end-November, while the write-down provision is recalculated using the inventory as of end-November and expectations of future selling prices — one looks backward, one looks forward. If NAND spot prices fall sharply in late November, and at year-end testing the post-February selling prices are revised down, it is entirely possible to book a provision before revenue has fallen much; the 10-K itself writes "in advance of when inventories are sold." And that lock is two-way: on October 30, Western Digital −7.4%, Seagate −8.0%, Micron −5.3%, all four storage names moved together; once it is not just NAND falling, both sides of the combined group are pressed at the same time, and the lock is gone. The absolute inventory amount also rose from 83.55 yi to 103.72 yi — the base is a notch larger, so if a provision really is taken, it will not be a rounding amount.
What counts as me being wrong: if in the December 8-K's earnings release or cost notes there appears a provision explicitly stated as written down to net realizable value, while the same report's gross margin is still above 84% — the order is reversed, I am wrong. Or one step earlier: Micron cuts gross margin guidance from 85.95% by more than 2 points, while inventory continues to increase quarter-on-quarter and the finished-goods share rises from 9% — the transmission is faster than I calculated, the warehouse is starting to pile up, and February will have to change its tune. There is also an institutional one: if Micron switches to testing DRAM and NAND for impairment separately, my core argument of "combined-group protection" directly fails. The test of the premise is equally hard: if Sandisk's actual gross margin next quarter is below 84.4% (the lower bound of its own guidance), then the statement that prices are rising moderately is wrong, and the price weakening is real, not sentiment.
Sources:
Micron Technology, Inc. Form 10-K (fiscal year ended 2026-09-03), filed 2026-10-09, accession 0000723125-26-000023, main file mu-20260903.htm — Note 6 Inventory, Inventory and Impairment in Accounting Policies, Revenue and Gross Margin in MD&A, Strategic Customer Agreement.
Micron Form 8-K, 2026-09-30, accession 0000723125-26-000018, EX-99.1 a2026q4ex991-pressrelease.htm — FY26 fourth quarter results and FY2027 first quarter guidance.
Micron Form 8-K, 2026-06-24, accession 0000723125-26-000013, EX-99.1 a2026q3ex991-pressrelease.htm — FY2026 third quarter data.
Micron Form 10-K (fiscal year ended 2023-08-31), filed 2023-10-06, accession 0000723125-23-000054, main file mu-20230831.htm — inventory net realizable value provision table, 5% selling price sensitivity.
Micron XBRL companyfacts, CIK 0000723125, InventoryWriteDown and quarterly revenue/gross margin/cost series, retrieved 2026-11-02.
Sandisk Corporation Form 8-K, 2026-10-29, accession 0001628280-26-068243, EX-99.1 sndkq1-27ex991xpressrelease.htm — FY2027 first quarter results and second quarter guidance.
Sandisk Corporation Form 10-Q, 2026-10-30, accession 0001628280-26-068601, main file sndk-20261002.htm — inventory details and inventory expense.
Sandisk FY2027 first quarter earnings call transcript, 2026-10-29 (Motley Fool transcription).
结论：先掉的是毛利率，而且它不等跌价准备 —— 12 月中旬那份季报里，毛利率会从 86.8% 往 85.95% 挪不到一个点，存货跌价准备连一行都不会有。以美光现在的成本结构，跌价准备要出现，得先让毛利率从 86% 一路跌到接近零，这不是「价格转弱」能干的事，是「价格崩塌」。
美光 9 月 3 日结束的那季（9 月 30 日发的 8-K）收入 542.29 亿美元，GAAP 毛利率 86.8%，营业成本 71.82 亿。每 100 块收入里，13 块 2 是成本。12 月中旬那份报表覆盖 8 月底到 11 月底，公司 9 月 30 日给的指引是收入 615 亿、GAAP 毛利率 85.95%，反推成本 86.4 亿。
毛利率等于 1 减成本除以收入。成本不动的时候，收入跌 x%，毛利率掉（1 减毛利率）乘 x%。现在 1 减 86% 只剩 0.14，所以收入跌 10%，毛利率掉 1.4 个点。跌价准备走的是另一条线：存货的预计售价低于成本。成本只占售价 13.2%，要让预计售价掉到成本，得跌 86%。这两件事隔着一个量级，不是一个季度。
NAND 占多少，决定「转弱」的杀伤力。10-K 里 FY26 DRAM 收入 1006.8 亿、NAND 317.9 亿、合计 1331.88 亿，NAND 占 23.9%。按这个占比估，NAND 全品类降 10%，收入掉 2.4%，毛利率掉 0.34 个点，毛利少 14.7 亿美元。它落在毛利上，不落在存货科目上。
存货结构更直接。附注 6：9 月 3 日存货 103.72 亿，成品 9.58 亿、在产品 82.52 亿、原材料 11.62 亿。成品只占 9%，相当于 12 天的营业成本。一年前同期成品是 10.94 亿 —— 收入已经是上一轮崩盘时（FY23 第二季 36.93 亿）的近十五倍，成品反而更少。减值的对象是成品和在产品，而这家公司做出来就拉走，仓库里没有一堆等着按市价重估的东西。
制度上还有一道锁。同一份 10-K 的会计政策写着，DRAM、NAND 和其他存储的存货合并成一个组，拿这个组去比成本和可变现净值。NAND 单独转弱，在规则上就触发不了减值 —— 只要 DRAM 那边还在赚钱（Q4 核心数据中心和移动客户端毛利率都是 90%），合并组的售价远高于成本，减值算不出来。想知道减值什么时候来，盯的不是 NAND，是 DRAM。
上一轮的顺序文件里写全了。FY23 第二季（截至 2023 年 3 月 2 日）毛利率 −32.7%，那一季计提 14.30 亿；第三季再提 4.01 亿，全年 18.31 亿，全年毛利率 −9%。但在这之前一季，毛利率已经从 46.7% 掉到 21.9%，收入从 86.42 亿掉到 40.85 亿、直接腰斩，跌价准备是零。毛利率先掉了一整季、掉了近 25 个点，存货科目才动。
这账有个两年的尾巴。FY23 提了 18.31 亿，同一财年因为卖的是已减记的存货、成本变低，返还 8.44 亿；FY24 再返还 9.87 亿，正好抵平。先一次性砸，再慢慢返 — 滞后的账，不是同步的账。
「价格转弱」这件事本身有多真。10 月 29 日盘后闪迪出季报：收入 110.31 亿、环比 +23%，毛利率 85.0%，下一季指引 84.4% 到 85.4% 基本走平，电话会上 CFO 说的是年内价格继续温和上行。第二天股价 −13.47%，美光跟跌 5.3%。文件里真正的软点只有一个：闪迪消费者业务 3.97 亿，环比 −29%、同比 −39%，公司说是市场在适应更高的价格。这是高价把低端需求打掉了，离合约价下跌还隔着一层。FY27 位元 50% 已签长约，FY28 是三分之二。
美光这边的保护写在同一份 10-K：战略客户协议是 take-or-pay、多年期、固定价或定期议价，多数带上下限；公司说即使按区间下限的价格，毛利率也明显高于历史上任何周期的单季峰值。历史单季峰值是 2018 财年三季度的 60.6%（我从 XBRL 的季度序列里算的）。换成大白话：合约那部分量的毛利率地板在 60% 以上。
12 月那份报表里会动的只有两样：毛利率从 86.8% 挪到 85.95% 那 0.85 个点 —— 这是 9 月 30 日给的指引，不是新消息；以及 2 月那份季报的指引。价格如果真松，顺序是先改指引，再改下一季收入，最后才轮到存货科目。
反方最强的一条：12 月报表的毛利率只认 8 月底到 11 月底已经发出去的货，跌价准备却是拿 11 月底那天的存货、按对未来售价的预期重算一遍 —— 一个往后看，一个往前看。如果 NAND 现货在 11 月下旬急跌，年末测试时把 2 月以后的售价往下调，完全可能在收入还没怎么跌时先提一笔，10-K 自己写的就是「in advance of when inventories are sold」。而且那道锁是双向的：10 月 30 日西数 −7.4%、希捷 −8.0%、美光 −5.3%，存储四家一起动，一旦跌的不只是 NAND，合并组两边同时被压，锁就没了。存货绝对额也从 83.55 亿涨到 103.72 亿，基数大了一档，真要提，不会是个零头。
什么情况算我看错：12 月那份 8-K 的业绩稿或成本附注里出现一笔明确写着按可变现净值减记的计提，而同一份报表毛利率还在 84% 以上 —— 顺序反了，我错。或者更早一步：美光把毛利率指引从 85.95% 下调 2 个点以上，同时存货环比继续增加、成品占比从 9% 往上走 —— 传导比我算的快，仓库开始积货，2 月就得改口。还有一条制度性的：如果美光改成 DRAM 和 NAND 分开做减值测试，我这条「合并组保护」的核心论据直接失效。对前提的检验同样硬：闪迪下一季实际毛利率若低于 84.4%（它自己指引的下限），那句价格温和上行就是错的，价格转弱是真的，不是情绪。
- 01Micron Technology, Inc. Form 10-K（财政年度截至 2026-09-03），2026-10-09 提交，accession 0000723125-26-000023，主文件 mu-20260903.htm —— 附注 6 存货、会计政策中的存货与减值、MD&A 的收入与毛利率、战略客户协议
- 02Micron Form 8-K，2026-09-30，accession 0000723125-26-000018，EX-99.1 a2026q4ex991-pressrelease.htm —— FY26 第四季业绩与 2027 财年第一季指引
- 03Micron Form 8-K，2026-06-24，accession 0000723125-26-000013，EX-99.1 a2026q3ex991-pressrelease.htm —— 2026 财年第三季数据
- 04Micron Form 10-K（财政年度截至 2023-08-31），2023-10-06 提交，accession 0000723125-23-000054，主文件 mu-20230831.htm —— 存货可变现净值计提表、5% 售价敏感性
- 05Micron XBRL companyfacts，CIK 0000723125，InventoryWriteDown 与季度收入/毛利率/成本序列，2026-11-02 取数
- 06Sandisk Corporation Form 8-K，2026-10-29，accession 0001628280-26-068243，EX-99.1 sndkq1-27ex991xpressrelease.htm —— 2027 财年第一季业绩与第二季指引
- 07Sandisk Corporation Form 10-Q，2026-10-30，accession 0001628280-26-068601，主文件 sndk-20261002.htm —— 存货明细与存货费用
- 08Sandisk 2027 财年第一季业绩电话会记录，2026-10-29（Motley Fool 转写）
收入和成本走的是两个时钟。售价按周重定价，今天谈的价今天就进收入；成本锁在已经做出来的货里，9 月 3 日那 103.72 亿存货按当季 71.82 亿营业成本算是 131 天的量。一季只有 91 天，价格要落进成本，得先穿过四个多月的库存。跌价准备是最慢的时钟，季末才测一次，测的是 11 月底那批货对往后售价的预期。最短的是收入，所以顺序没有悬念。
## 指引里没给价格下跌留位置
9 月 30 日那份指引，收入 615 亿、毛利率 85.95%，反推成本 86.4 亿，比这一季多两成。电话会上 CFO 那句是钥匙：第一季指引假设位元出货环比只增个位数。收入要增 13.4%，量只增个位数，多出来的全是价——指引里价格是往上走的，一分钱下跌都没算。
所以 12 月那份报表出来，毛利率下滑是算术结果，不是新闻。价格跌 20%，毛利率掉到 83.4%；跌 30%，81.1%；跌 50%，73.5%。成本率只有 13.2%，分母一缩，毛利率同比例缩，中间没有缓冲垫。而跌价准备的门槛在另一边——售价要跌穿成本才算，那是 87% 的跌幅，不是一两个季度的事。
## 同行四家，没有一家在转弱
十月底三家交了季报，季末和美光只差一个月，横着看。
| 公司 | 单季收入(亿美元) | GAAP毛利率 | 环比 | 存货/单季成本(天) |
|---|---|---|---|---|
| 美光 FQ4-26 | 542.29 | 86.8 | +2.2 | 131 |
| 闪迪 Q1-27 | 110.31 | 85.0 | +0.4 | 166 |
| 西数 Q1-27 | 43.81 | 57.8 | +3.7 | 88 |
| 希捷 Q1-27 | 42.33 | 57.3 | +5.0 | 86 |
机械硬盘那两家五十七个点上下，还在环比往上走；NAND 纯玩家闪迪八十五个点，环比走平。所谓"存储价格转弱"，在这四份已披露的报表里一行都没有。
四家的成本率更说明问题：NAND 两家是 13.2% 和 15.0%，硬盘两家是 42.2% 和 42.7%。同样是存储，NAND 的售价是成本七八倍，硬盘只有两倍多——不是 NAND 更赚钱，是它的价格离成本最远：掉起来最自由，要掉到成本线也最远。
真正转弱的是别的东西：闪迪消费者分部收入 3.97 亿，环比 −29%，公司说这是"市场在适应更高的价格"——量被打掉，不是价往下走。同一场电话会，CFO 给下一季毛利率指引 84.5% 到 85.5%，明确"包含温和的同比提价"。价格转弱的证据目前只在链条最末端：零售和渠道。
```chart
type: line
title: 美光单季 GAAP 毛利率爬到 86.8% 后开始走平
unit: %
labels: 2026Q1, 2026Q2, 2026Q3, 2026Q4, 2027Q1指引
values: 56.0, 74.4, 84.6, 86.8, 85.95
```
## 价格先到的是最脆的那一端
传导链是五步：现货与渠道价 → 模组和零售 → 长约与战略客户协议 → 美光的季度实现均价 → 收入与毛利率。存货减值不在这条链上，它是另一条线：季末的存货，乘上对明年售价的预期，再和成本比大小。
第三步是这条路最堵的地方。10 月 1 日电话会上说得很清楚：2027 财年超过 75% 的出货已在客户分配或合同之下，26 份战略客户协议覆盖到 2030 年约 35% 的位元产量。闪迪更直接，2027 财年 50%、2028 财年三分之二的位元已签长约。说白了，明年的价格已经卖掉了。NAND 现货今天转弱，落不进 12 月那份报表的收入，它落的是 2028 财年重签长约的起点。
这也是 12 月中旬没什么看头的原因：真正会被改写的两个数——2027 财年下半年实现均价、2028 财年合约价——都还在后面，这份报表只能验证指引准不准。
## 两个信号不是一个量级
先看跌价准备的基数。10-K 附注 6 里，成品 9.58 亿、在产品 82.52 亿、原材料 11.62 亿，在产占七成九。NAND 的存货成本没单独披露，按收入占比 23.9% 推是 24.8 亿；NAND 毛利率低于 DRAM，按成本占比三成半推更合理，约 36 亿。
就算 NAND 全归零、公司改成分产品测试，减值上限也就三十几亿美元，相当于单季 615 亿收入的 4% 到 6%，毛利率从 86.8% 掉到 81%。
> U.S. GAAP provides for products to be grouped into categories in order to compare costs to net realizable values … we categorize all inventories (including DRAM, NAND, and other memory) as a single group.
更硬的一道锁在这段政策里。公司把三类存储合成一个组，拿整组成本比整组可变现净值。按 10-K 比例粗算：DRAM 占存货成本约七成、72.6 亿，按合并成本率算售价是成本的 7.5 倍，这块富余 475.6 亿；NAND 按三成成本率算，富余 72.6 亿。
NAND 售价跌 85%，那块从富余 72.6 亿变成倒挂 15.5 亿，整组仍是 563.8 亿对 103.7 亿成本，差五倍多。要让这个组合破，必须 DRAM 跟着跌，且跌到和 NAND 一个量级。
量级摆在一起：毛利率这边，价格跌两成就是 3.4 个点；跌价准备那边，最先能动的是 NAND 那块，四到六个点。笔头大的那个是小概率，笔头小的那个是大概率。
## 反方最强的一条：我用的是合并的尺子
这套算法有个洞：13.2% 的合并成本率是 DRAM 带出来的，DRAM 占收入 75.6%、数据中心毛利率 90%，NAND 自己的成本率可能高得多。敏感度：NAND 成本率 30%，需要的跌幅是 70%；50%，跌一半就够；65%，跌 35% 就穿。这是我这条判断唯一能被真正推翻的地方——不是"会不会先掉"，而是"跌多少才轮到"。NAND 单季毛利率哪天单独披露，这把尺子立刻能校准。
时点上的那条反驳只能挪动一个季度：跌价准备测的是 11 月底的存货和往后 12 个月的售价，10-K 原话是提前认、不等卖出去。11 月下旬现货若急跌、公司把明年售价曲线往下拉，可能在收入还没掉时先提一笔。但金额受上面基数约束，量级变不了。
## 12 月那份报表，先看三行
- 存货：103.72 亿再涨 15% 以上，成品占比从 9.2% 升过 15%——仓库开始积压，2 月那份就要提。
- 毛利率：掉 1.5 个点以内算兑现指引；掉 3 个点以上，说明价格转弱真进了报表，但减值的钟还没响。
- 数据中心分部毛利率：这一季 90%，跌破 85% 才是 DRAM 那头松动的先行信号，也是组合锁失效的前兆。
硬证伪两条：12 月的现金流量表或存货附注里出现 10 亿美元以上、明确按可变现净值减记的计提，而毛利率还高于 84%——顺序反了；或者公司改成 DRAM 和 NAND 分开做减值测试——上面那套组合算术作废。
反向也一样硬：闪迪二季实际毛利率若跌破自己给的非 GAAP 下限 84.5%，"温和提价"那句话就是错的，价格转弱是真的。可即便如此，它离美光的存货科目还隔着一整条 131 天的成本通道和一年多已经锁掉的长约。
所以 12 月中旬那份季报，第一行要看的不是毛利率，是存货。价格转弱先体现在仓库变重，再体现在毛利率变薄，最后才轮到那笔一次性减记。这个顺序只有在一种情况下会翻：DRAM 先倒，而且是带着 NAND 一起倒。
来源：
Micron Form 10-K（财年截至 2026-09-03），2026-10-09，0000723125-26-000023，mu-20260903.htm——附注 6 存货、存货会计政策与分组、MD&A 毛利率、风险因素中的五年 ASP 区间、战略客户协议。
Micron Form 8-K，2026-09-30，0000723125-26-000018，EX-99.1 a2026q4ex991-pressrelease.htm——FY26 第四季业绩与 2027 财年第一季指引。
Micron 2026 财年第四季电话会记录，2026-10-01（Motley Fool 转写）——位元增速、75% 出货锁定、26 份协议。
Micron XBRL companyfacts，CIK 0000723125，单季序列，2026-11-02 取数。
Sandisk Form 8-K，2026-10-29，0001628280-26-068243，EX-99.1 sndkq1-27ex991xpressrelease.htm——Q1-27 业绩、分终端收入、Q2-27 指引。
Sandisk 2027 财年第一季电话会记录，2026-10-30（Motley Fool 转写）——温和提价、消费者分部、位元长约比例。
Western Digital Form 8-K，2026-10-22，0001628280-26-066882，EX-99.1 a4ex991-pressreleaseq127.htm——Q1-27 业绩与毛利率。
Seagate Form 8-K，2026-10-27，0001137789-26-000300，EX-99.1 stxq12027pressreleasefinan.htm——FQ1-2027 业绩与毛利率。