AI & Infrastructure CapexAI 与基础设施资本开支
Micron Technology, Inc.
MU 4Price and target zones价格与目标区间
Above the sell price高于卖出价
What the company does公司做什么
Micron Technology, Inc. designs, develops, manufactures, and sells memory and storage products in the United States, Taiwan, Japan, Mainland China, Hong Kong, Europe, and internationally. It operates through the Cloud Memory Business Unit; Core Data Center Business Unit; Mobile and Client Business Unit; and Automotive and Embedded Business Unit segments.
Read the full description查看完整介绍
Micron Technology, Inc. designs, develops, manufactures, and sells memory and storage products in the United States, Taiwan, Japan, Mainland China, Hong Kong, Europe, and internationally. It operates through the Cloud Memory Business Unit; Core Data Center Business Unit; Mobile and Client Business Unit; and Automotive and Embedded Business Unit segments. The company provides memory products, including dynamic random access memory components and modules, CXL-based memory, LPDDR components and modules, graphics memory, high-bandwidth memory, and data center memory products; multichip packages (MCP) comprising embedded multimedia card-based, universal flash storage-based, and NAND-based MCPs; and technology leadership products that include 1y DRAM and G9 NAND technologies. It also offers storage products, such as data center solid-state drives (SSD), client SSD storage, and auto and industrial SSD storage; managed NAND; NAND flash; NOR flash; and memory cards. In addition, the company provides design tools, including FBGA and part decoders; DRAM power calculators; NAND power calculators; simulation models; chipset compatibility guides; serial presence-detection tools; cross-reference tools; UFSparm; SSD firmware; software and drivers; storage executive software; and obsolete part catalogs. It markets its semiconductor memory and storage products under the Micron and Crucial brands. The company serves the data center, PC, graphics, networking, automotive, industrial, and consumer embedded markets, as well as the smartphone and other mobile-device markets. It sells its products through its direct sales force, independent sales representatives, distributors, and retailers; web-based customer direct sales channel; and channel and distribution partners. Micron Technology, Inc. was founded in 1978 and is headquartered in Boise, Idaho.
Products and services产品与服务 from the vendor summary — not yet written up来自供应商简介——尚未整理成文
- The company provides memory products, including dynamic random access memory components and modules, CXL-based memory, LPDDR components and modules, graphics memory, high-bandwidth memory, and data center memory products
- Multichip packages (MCP) comprising embedded multimedia card-based, universal flash storage-based, and NAND-based MCPs
- Technology leadership products that include 1y DRAM and G9 NAND technologies It also offers storage products, such as data center solid-state drives (SSD), client SSD storage, and auto and industrial SSD storage
- Managed NAND
- NAND flash
- NOR flash
- Memory cards In addition, the company provides design tools, including FBGA and part decoders
- DRAM power calculators
- NAND power calculators
Company facts公司资料
- Founded成立时间
- 1978
- Headquarters总部
- Boise, Idaho
- Employees员工人数
- 53,000
- Industry行业
- Semiconductors
- Country国家
- United States
Investment thesis投资逻辑 from the watchlist来自关注列表
What's working
- Real, and credited at 5.1x the pre-HBM decade: 16 take-or-pay SCAs to 2030, ~$100B RPO, $22B collateral posted — and the price cap excludes HBM/DDR6/LPDDR6, so the AI line is not capped
- Debt $16.1B→$6.4B
- 25% DRAM share, 1pt behind SK Hynix
Red flags
- The killer number: cumulative FY21-25 FCF was $1.22B on $136.5B of revenue — a 0.90% five-year FCF margin. FY2025 was a record profit year and threw off $1.67B of cash; capex ate 98% of five years of operating cash flow. Median ROIC ~4%; FY2023 gross profit was −$1.42B (sold below cash cost)
- 5.74x fwd is not value — it is the market pricing a collapse, and not enough of one: reverse-engineered it discounts a 62% peak-to-normalized decay, when Micron's actual historical decay is 90-100%. It prices no CXMT and no capex reckoning
- The bull case fails its own best test — grant 3 more years at the run-rate and 100% cash return from 12/09, you collect 22-31% of the market cap and still hold the residual at 21-34x through-cycle earnings.
- CXMT ~350K wafer-starts/mo by end-26 (vs MU's ~375K) → 17% share by 2028, aimed squarely at the 60-80% of volume with no contract floor, while industry bit supply grows only 6-8%/yr
- Capex $27B FY26 → $40B+ FY27 = 40-50% of normalized revenue, landing 2028-30 — exactly when the SCAs expire
- Insiders sold ~$233M with zero open-market buying; CEO sold $46.3M at $1,128-1,192 two days after the record print; officer Arnzen sold 5x her prior clip size near the high
- The $18.6B "net cash" is largely customer money — non-debt liabilities +$15.5B vs cash +$14.8B; the ~$18B of deposits is refundable, back-end weighted to 2029-30
Watch
- ✂️ TRIM NOW / EXIT — analyzed 8/04, first file ever. The prior "top sell candidate" call is confirmed after crediting the bull case at full strength, not by ignoring it. 52wk $106.75 → $1,255 → $892: −28.9% off the high but +736% off the low — the exact shape of
pattern-ai-levered-fields-trade-above-own-band - But MU holds only single-digit-to-low-teens of HBM4/Vera Rubin vs SK Hynix's 60-70%, and SK Hynix removed its price cap entirely in July — MU bought the floor, SK Hynix kept the upside. ⚖️ Take-or-pay is enforceable (Hemlock v. SolarWorld, 6th Cir. 2017, $800M affirmed) — but Hemlock won and still collected almost nothing when the counterparty failed. The contracts compress the trough; they do not abolish it. Normalized EPS ~$20; FV $280-560 (mid ~$400) — spot is ~2.2x that.
- No trim multiple is set here on purpose: at a 5.74x forward multiple the convention would render a sell level of $1,244 — above the all-time high — and instruct holding through the top. Cyclicals at peak earnings need an instruction, not a multiple →
pitfall-multiple-trim-inverts-on-peak-cycle-cyclicals. Exit, or trim at minimum — still a loss, so tax-cheap, but the harvest value has already shrunk $577 as the price recovered. - Timing risk is real and acknowledged: 46 ratings, 0 sells, mean tgt $1,508, and an FQ4 guide of $50B/86%/$31.00 that will likely be met or beaten in September
🔴 [4.0] 8/04 分析 — 确认减持/清仓，而且是在充分肯定多头理由之后得出的结论，不是无视它。 52 周区间 $106.75 → $1,255 → $892：距高点 −28.9%，但距低点 +736%。
🚩 最要命的数字：2021-2025 财年累计自由现金流只有 $1.22B，而累计收入是 $136.5B — 五年自由现金流利润率 0.90%。 2025 财年是创纪录的盈利年，也只产生了 $1.67B 现金。中位投入资本回报率约 4%；2023 财年毛利润是 −$1.42B（低于现金成本销售）。
🚩 5.74x 前瞻市盈率不是便宜，而是市场在给崩塌定价 — 而且定得还不够：它只折价了 62% 的从峰值到常态的回落，而美光历史上的实际回落是 90-100%。
🚩 多头逻辑通不过它自己最有力的检验 — 即使再给三年当前的经营水平并且从 12/09 起 100% 返还现金，你拿回市值的 22-31%，剩下的部分仍相当于 21-34 倍的穿越周期盈利。
🚩 CXMT 到 2026 年底约 350K 片/月（美光约 375K）→ 2028 年market share 17%，瞄准的正是没有合约保护的那 60-80% 出货量；而全行业位元供给每年只增长 6-8%。
🚩 资本开支 2026 财年 $27B → 2027 财年 $40B+，相当于常态收入的 40-50%，产能在 2028-30 年落地 — 恰好是 SCA 合约到期的时候。
🚩 内部人卖出 约 $233M，公开市场买入为零；CEO 在创纪录财报公布两天后于 $1,128-1,192 卖出 $46.3M。
🚩 $18.6B 的"净现金"很大程度上是客户的钱 — 非债务负债增加 $15.5B，而现金增加 $14.8B；$18B 的客户押金是可退还的，且集中在 2029-30 年退回。
✅ 确实存在并已计入的利好：16 份到 2030 年的照付不议长约，约 $100B 合约收入，$22B 抵押品 — 而且价格上限不包括 HBM/DDR6/LPDDR6，所以 AI 业务线并没有被封顶。✅ 债务从 $16.1B 降到 $6.4B。✅ DRAM 份额 25%，距 SK 海力士只差 1 个百分点。
❌ 但美光在 HBM4/Vera Rubin 上只拿到 个位数到十几个百分点，而 SK 海力士是 60-70% — 而且 SK 海力士在 7 月彻底取消了自己的价格上限。美光买下了地板，SK 海力士留住了天花板。
常态化每股收益 约 $20（是 HBM 之前那十年平均值的 5.1 倍）；合理估值 $280-560（中值约 $400）— 当前股价约为其 2.2 倍。
⚠️ 这里刻意不设倍数形式的减持线 — 在 5.74x 前瞻市盈率下，Trim 8x fwd 会算出 $1,244，高于历史最高价。周期股在盈利顶部需要的是明确指令，不是倍数。清仓，或至少减持 — 目前仍是亏损，所以卖出税负便宜，但随着股价回升，可收割的税务价值已经缩水了 $577。
⚠️ 时机风险真实存在：46 家评级机构、0 家卖出、平均目标价 $1,508，而且 FQ4 指引 $50B/86%/$31.00 很可能被超越。A2026-08-04
Historical results历史业绩 past fiscal years — these values do not change过去财年——这些数值不会变动
Profit margins利润率 the share of revenue that stays as profit营收中留存为利润的比例
of every $1 of revenue每 $1 营收中的占比
- Return on equity净资产收益率
- 67%
- Revenue growth营收增长
- +346%
- Dividend yield股息率
- 0.06%
Competitive advantages竞争优势 from the thesis and the reports来自投资逻辑和报告
- A company with a cost moat does not sell below cash cost in a downturn.
- Moat sources, rated Source Rating Evidence --- --- --- Efficient scale Present (strong) — but industry-owned Three-player DRAM oligopoly, ~93% combined share; $20–30B and 4+ years to add a fab.
- This is the real moat, and it belongs to the industry, not to Micron — which is the smallest of the three holders.
- Switching costs Weak — protecting the wrong thing HBM qualification is 9–18 months and design-locked (Samsung's 18-month HBM3E failure proves the cost is real).
- Memory sells on spec and price; no brand pricing power.