MU — Micron
The numbers look cyclical. The contracts are trying to make the cycle less cyclical.
Micron sells DRAM, NAND and HBM into a memory market that just got even harder to call a normal cycle. Q4 revenue reached $54.2B, Q1 FY2027 is guided to $61.5B, management now expects 2027 and 2028 to be tighter than 2026, and contractual customer commitments extend well beyond the current shortage. I am more convinced the earnings floor is higher. I am still not calling 87% gross margin normal.
Micron just reported $33.42 of non-GAAP EPS in one quarter and guided the next one to $38.15.
Why I care about this one
Q4 revenue was $54.23B, up 31% sequentially. Non-GAAP gross margin hit 87% and adjusted free cash flow was $33.2B.
The mix underneath that was not exactly weak either.
DRAM revenue reached $39.8B. NAND revenue reached $14.1B. Data-center SSD revenue was nearly $10B, more than ten times the year-ago quarter, and Core Data Center revenue jumped 56% sequentially to $18B. (Q4 results)
Then Q1 FY2027 guidance came in at $61.5B of revenue, ~86.25% gross margin and $38.15 of EPS.
Management also said Q1 should be the gross-margin floor for FY2027.
Yeah.
This is the point where a memory model can get very stupid very quickly.
Before earnings, I was already convinced the 2027 setup looked unusually tight. The part I was less willing to underwrite was duration.
That got better.
Micron says DRAM and NAND should remain supply constrained through both 2027 and 2028. HBM bit demand should keep growing faster than conventional DRAM through 2028. Even with new clean-room space coming online, management says it still cannot see when supply and demand get back into balance. (Q4 call transcript)
I do not take management’s cycle forecast as gospel. That would be a fun way to learn the same lesson memory investors have learned for forty years.
But the evidence behind it is better than a normal peak-cycle pitch.
Idaho 1 only starts wafer output in mid-2027. Idaho 2 and the Japan expansion do not start until late 2028. New NAND capacity in Singapore is also a second-half 2028 story. And first wafer output is not the same thing as full production.
The capacity response is coming.
It just has a long commute.
Fiscal Q4 2026 + Q1 FY2027 guide + Sep. 30 earnings call
Primary sources: Fiscal Q4 / FY2026 results ↗ · Micron Sep. 30 investor events / webcasts ↗ · Fiscal Q4 earnings call transcript — fallback ↗ · Sep. 29 HBM shortage / 2027 pricing — TrendForce ↗
Q4 revenue reached $54.23B with 87% non-GAAP gross margin and $33.42 of EPS, while Q1 FY2027 is guided to $61.5B of revenue and $38.15 of EPS. More important for the normalized model, Micron now sees 2027 and 2028 tighter than 2026, has more than 75% of 2027 output committed, 26 SCAs covering more than 35% of estimated revenue through 2030, roughly $150B of RPO and $32B of customer commitments. FY2029 normalized EPS moves from about $80 / $150 / $205 to $90 / $160 / $215, or roughly $810 / $2,240 / $3,440 at the same 9x / 14x / 16x multiples. Capital Allocation moves from 8.25 to 8.75 as the balance sheet and capital-return framework improve.
My position
What I actually bought and sold.
| Date | Type | Account | Dollars | Shares | Price | Return now* |
|---|---|---|---|---|---|---|
| Dec 12, 2025 | Buy | Taxable | $4,802 | 18.93 | $254 | +319.8% |
| Jan 14, 2026 | Reinvest | Taxable | $2.18 | 0.006 | $363 | +193.1% |
| Apr 15, 2026 | Reinvest | Taxable | $2.84 | 0.006 | $473 | +125.0% |
| Jul 20, 2026 | Buy | Taxable | $5,000 | 5.657 | $884 | +20.5% |
| Jul 21, 2026 | Reinvest | Taxable | $2.84 | 0.003 | $947 | +12.5% |
*Return now uses the latest available quote for buys and reinvestments. Sales and fees show “—”.
Valuation & scenario math
The HBM story did what I wanted.
Micron has agreements for the vast majority of its calendar 2027 HBM bit supply, with significantly higher pricing than 2026. HBM4 is ramping and HBM4E comes next.
What I find more interesting is everything around it.
Conventional DRAM pricing rose high teens sequentially in Q4. NAND pricing rose about 30%. Data-center SSD is becoming a serious AI business in its own right, helped by KV-cache offload and the broader memory hierarchy.
That is why I do not really think of this as “the HBM trade” anymore.
The whole memory stack is being pulled harder by AI.
Normalized FY2029 earnings + HBM/SCA durability + position size
No static buy band. A lower quote helps the return math, obviously. Another dollar still has to survive the FY2029 shortage-to-capacity transition and justify pushing an already large position closer to the 10% hard cap.
Q4 made the near-term earnings story even stronger, but the bigger change is duration: management now sees 2027 and 2028 tighter than 2026, more than 75% of 2027 output is already committed, and 26 SCAs provide contractual visibility through 2030 and into 2031. I still normalize in FY2029 because late-2028 capacity is real. The model gets a higher floor, not permission to capitalize 87% gross margin forever.
| Normalized cycle EPS | ~$160 |
| Terminal P/E | 14x |
| Checkpoint | Sep 30, 2029 |
| Implied base price | ~$2,240/share |
This is a normalized-cycle checkpoint, not me compounding peak economics forever.
| Case | What has to be true | Target range | Annualized return from today |
|---|---|---|---|
| Bear | ~$90 normalized EPS · 9x P/E. FY2029 revenue around $225B with gross margin near 55% and diluted shares around 1.10B. The late-2028 capacity response catches demand, pricing normalizes hard and customer optimization matters. SCAs soften the cycle but do not repeal it. | $729–$891 center ~$810 | -8.8% |
| Base | ~$160 normalized EPS · 14x P/E. FY2029 revenue around $300B with gross margin near 68% and diluted shares around 1.02B. Tight conditions persist through 2028, then normalize as new capacity ramps; richer data-center mix, SCAs and sustained buybacks keep the earnings floor structurally above old memory cycles. | $2,016–$2,464 center ~$2,240 | +28.5% |
| Bull | ~$215 normalized EPS · 16x P/E. FY2029 revenue around $350B with gross margin near 75% and diluted shares around 0.98B. AI memory demand keeps outrunning supply, HBM4/HBM4E and data-center SSD mix stay unusually strong, SCAs preserve pricing power and excess-cash repurchases materially lift per-share earnings. | $3,096–$3,784 center ~$3,440 | +48.4% |
Returns use the current quote and the time remaining to Sep 30, 2029. The scenario targets stay fixed until the underwriting changes.
Capital allocation
The cash-return plan finally gives the balance sheet a job. The next test is whether giant capacity spending stays tied to contracted demand.
Micron is funding leading DRAM, HBM and SSD products while generating extraordinary free cash flow, holds $68.3B of net cash, and now has 26 SCAs plus more than 75% of 2027 output committed. I raise capital deployment because management intends to return excess cash primarily through repurchases starting in December. I still stop well short of a perfect score because FY27 capex is accelerating sharply and memory companies have a long history of discovering supply discipline right up until they do not.
Stage-relative, not some objective rating-agency grade. The number is there to force the judgment into the open. See the rubric →
What management is actually doing with the cash
The capex wave is getting much bigger, which would normally be where I start reaching for the red pen. The difference is Micron now has 26 take-or-pay SCAs, more than 75% of 2027 output committed, roughly $150B of RPO at minimum pricing and $68.3B of net cash. Management is also saying excess cash will increasingly come back to shareholders, primarily through repurchases, once the CHIPS restriction rolls off in December. That is a much better capital-allocation setup than 'memory is hot, build everything.' I still want the equipment spend to follow contracted demand instead of today's margins.
What I’m doing with the next dollar
This is where I refuse to turn the page into a victory lap.
Micron is going to spend a lot of money.
Q1 FY2027 capex is projected around $11.5B, first-half capex around $25B, and second-half spending should be higher. Construction is growing much faster than equipment spending because Micron is trying to bring clean-room space forward while keeping equipment additions tied to demand. (Q4 call transcript)
That sequencing matters.
Construction today creates the option to add equipment later. Equipment is where the bit supply really shows up.
If management keeps that discipline, great.
If today’s margins convince the industry that every possible tool needs to be installed immediately, I know how this movie ends.
What can break this
This was probably the most important thing on the call.
Micron now has 26 take-or-pay SCAs. Management estimates they cover more than 35% of company revenue through 2030, with three-quarters of that expected revenue carrying a defined pricing framework. The company disclosed roughly $150B of remaining performance obligations based on committed volumes and minimum pricing, plus $32B of customer financial commitments, mostly cash deposits. (Q4 call transcript)
Management also said the RPO is conservative because it uses minimum pricing, and that even SCA floor prices should produce margins meaningfully above prior cycle peaks.
That is a pretty big claim.
I am not putting it into the model at face value. The SCAs do not cover the whole business, market-priced revenue still matters and 2030 is a long time away.
This has moved beyond “customers really want memory right now.” Customers are putting real commitments behind the demand.
The FY2029 checkpoint still makes sense to me because it forces the model to live through the late-2028 capacity response.
My current cases are:
| FY2029 | Bear | Base | Bull |
|---|---|---|---|
| Revenue assumption | ~$225B | ~$300B | ~$350B |
| Gross margin | ~55% | ~68% | ~75% |
| Normalized EPS | ~$90 | ~$160 | ~$215 |
| Terminal P/E | 9x | 14x | 16x |
| Implied value | ~$810 | ~$2,240 | ~$3,440 |
The share-count assumptions also get better because the balance sheet is becoming ridiculous.
Micron ended Q4 with $68.3B of net cash. Management expects Q1 free cash flow to be significantly higher than Q4’s $33.2B and intends to step up capital returns from December 9, primarily through share repurchases, with a long-run goal of returning 100% of excess cash to shareholders. (Q4 call transcript)
So I am allowing some real share-count reduction in the base and bull cases.
Not heroic buybacks. Real ones.
The evidence I’m watching
SCAs move toward the roughly 50% long-run revenue coverage management still sees as possible, 2027–2028 HBM and conventional-memory shortages persist without material demand destruction, data-center SSD stays structurally important, HBM4/HBM4E wins broaden, and the December capital-return step-up actually shrinks the share count while capacity remains tied to contracted demand.
The late-2028 capacity wave catches demand faster than management expects, customers materially de-spec or delay deployments, SCA floor economics prove weaker than the current RPO language suggests, HBM pricing loses its advantage, or management lets extraordinary current margins justify equipment spending that outruns the contracted demand behind it.