muji | March 2026
Here's the thing... MU is not a company I'd normally be drawn to. No SaaS flywheel. No NRR. No platform optionality in the way I define it. Historically, memory is the anti-platform — commodity pricing, brutal cyclicality, COGS-heavy, capital-intensive.
But the AI supercycle has done something structurally different to this business, and I want to work through whether it's a durable shift or a cycle dressed up as a structural story.
NUANCE: What makes MU interesting right now is not the commodity DRAM business. It's that HBM has created something this industry has never had: qualification moats + pricing power + supply constraints simultaneously. That's a picks-and-shovels trifecta.
Let me just lay out what's happened here:
Revenue trajectory:
Q2 FY25: $8.1B ████████░░░░░░░░░░░░░░░░░░
Q3 FY25: $9.3B █████████░░░░░░░░░░░░░░░░░
Q4 FY25: $11.3B ███████████░░░░░░░░░░░░░░░
Q1 FY26: $13.6B █████████████░░░░░░░░░░░░░
Q2 FY26: $18.7B+ ██████████████████░░░░░░░░ (guide)+38% → +37% → +46% → +57% YoY ^^
That's accelerating YoY growth at $13.6B scale. DING DING DING.
Margin trajectory:
Q1 FY24: -0.7% GM (trough)
Q2 FY24: 18.5%
Q3 FY24: 26.9%
Q4 FY24: 35.3%
Q1 FY25: 38.4%
Q2 FY25: 36.8%
Q3 FY25: 37.7%
Q4 FY25: 44.7%
Q1 FY26: 56.0% !!!
Q2 FY26: 68.0% !!! (guided)From negative gross margin to 68% in seven quarters. I've covered hundreds of companies. I have never seen a margin inflection this dramatic at this scale. Not in SaaS. Not anywhere.
FCF trajectory:
Q1 FY25: $38M (0.4% margin)
Q2 FY25: $857M (10.6%)
Q3 FY25: $1.9B (21.0%)
Q4 FY25: $803M (7.1% — capex heavy)
Q1 FY26: $3.9B (28.6%) !!!$3.9 BILLION of free cash flow in a single quarter. At 29% FCF margin. For a memory company. Let that sink in.
Is MU a platform company? No. Not by my traditional definition.
Is MU a "building block" or "process tool"? BUILDING BLOCK — and here's why that matters.
HBM is not something you can substitute. It's physically stacked on the GPU/ASIC die. It's co-designed with Nvidia. It requires qualification that takes 6-12 months. Once you're in, you're in for that generation. Once you're out... well, ask Samsung about HBM3E.
This is the "building block" concept in hardware form. MU's HBM is literally embedded in the customer's product. It scales with the customer's success (more GPU shipments = more HBM). It has high switching costs. It's not a process tool that sits alongside and can be ripped out.
Technology leadership as moat:
NUANCE: The internal logic die for HBM4 is underappreciated. SK Hynix and Samsung outsource this to foundries (TSMC, Samsung Foundry). MU designs and builds it internally. That means faster iteration, tighter integration, and better power efficiency. In a market where every watt matters for AI clusters, this is a real differentiator.
| Business Unit | Q1 FY26 Rev | Mix | Signal |
|---|---|---|---|
| Cloud Memory (HBM + DC DRAM) | $5.28B | 39% | Record. AI-driven. HBM ~$2B+ of this |
| Mobile & Client | $4.26B | 31% | Stable. Benefiting from shortage pricing |
| Core DC (Enterprise SSD) | $2.38B | 17% | Record. +51% QoQ !! |
| Auto/Embedded/Industrial | $1.72B | 13% | Steady growth |
DRAM = 79% of revenue (10.8B).NAND = 202.7B).
The mix shift toward DC + HBM is the profitability driver. HBM margins are dramatically higher than commodity DRAM. As HBM grows from ~15% of DRAM revenue toward 25-30%, the blended margin story keeps getting better.
OK, here's where it gets spicy. There are conflicting reports and I want to be straight about what we know and don't know.
What we know:
The bear case (Enertuition/SemiAnalysis, Feb 2026):
The bull case (TechStock01/DigiTimes):
My read: The truth is probably in the middle. MU likely has HBM4 technical qualification but may not be in the initial production allocation for Vera Rubin. That's a meaningful distinction. SK Hynix has been Nvidia's #1 HBM partner for years and will get first allocation. But as Rubin volumes scale through H2 CY26 and into CY27, MU gets its share.
The March 18 earnings call will be the single most important data point. If Mehrotra confirms HBM4 volume shipments and Nvidia allocation, the stock re-rates. If he's vague or confirms initial exclusion, there's a pullback.
Let me walk through this systematically.
Revenue: Beat. Probably big.
Guide: 18.7B±400M. Consensus: 19.15B.MostAccurateEstimate: 19.5B.
Here's why I think the beat could be larger. TrendForce reported in February that Q1 2026 DRAM contract pricing gains exceeded 90-95% QoQ, vs. the ~30% Micron originally contemplated in guidance. NAND Flash contracts up 55-60% QoQ. Enterprise SSD pricing up 53-58% QoQ — record quarterly increases.
MU's guide was set in mid-December when these pricing negotiations were just beginning. The actual pricing environment has been significantly stronger than assumptions. Trail-four-quarter average beat: 14.4%.
My estimate: $19.5-20.5B. Possibly 20B + .Thiswouldbe 7B bigger than Q2 FY25 ($8.05B). +138-155% YoY. Absurd.
Gross Margin: Beat.
Guide: 68% ±100bps (Non-GAAP). Given the pricing environment, 69-71% is realistic. This would be the highest gross margin in Micron's history by a wide margin. The prior cycle peak was ~47% in Q2 FY22.
To put this in perspective: Micron is guiding for gross margins higher than most SaaS companies. A hardware memory company. With ~68-70% gross margins. That's the HBM effect.
EPS: Beat.
Guide: 8.42±0.20 (Non-GAAP). Consensus: $8.69. Most Accurate: $9.19.
If revenue beats by 1 − 2Bat689-11.**
FCF: Strong but capex-heavy.
FY26 CapEx guided at ~$20B. Q2 will likely be a heavy capex quarter (Idaho Fab ramp). FCF might compress to 15-20% margin despite revenue surge. Still $3-4B of FCF.
| Risk | Probability | Impact | Signal |
|---|---|---|---|
| HBM4 initial Rubin exclusion | Medium | High | March 18 call |
| Cyclical downturn in CY27 | Low-Medium | Very High | Memory pricing starts topping |
| Samsung HBM3E qualification | Confirmed (AMD) | Medium | Adds supply, moderate pricing pressure |
| Consumer DRAM/NAND weakness | Low (exited consumer) | Low | Already pivoted enterprise |
| CapEx overcommitment | Medium | High | $20B+ if demand slows |
| Tariff exposure | Low-Medium | Medium | Not in guidance |
The fundamental risk nobody wants to talk about: Gross margins of 68% in memory are not normal. Historical peak was ~47%. The current level requires sustained supply shortage + HBM pricing power + mix shift all simultaneously. Any one of these easing returns margins to 45-55%. All three easing returns margins to 30-40%.
IMHO, the market is debating whether this is "normal" now. It's not. But the "new normal" floor is probably 35-45%, up from the old floor of 20-30%. That's the structural improvement.
At ~$280/share:
If Q2 delivers 20Bat7010 EPS:
But — and this is critical — you cannot annualize a cyclical peak. The market knows Q2 FY26 is probably near-peak margins. What matters is the through-cycle earnings power, which I'd estimate at $15-20 EPS. At $280, that's 14-19x through-cycle P/E. Reasonable, not cheap.
This is tricky for my framework because MU doesn't fit the traditional SaaS tier structure.
Revenue growth: +57% YoY → Tier 1 territory by the numbers Platform: Building block, not SaaS platform → Tier 2 treatment (no recurring dynamics) Cyclicality discount: Memory company with -33% margins two years ago → Tier 3 weight cap
Classification: Tier 2 — 8-12% allocation. Growth deserves Tier 1 weight, but cyclicality and non-recurring nature cap it. The current wsm007 allocation of ~10.4% is appropriate.
Intact — Strengthening on execution, cautious on cycle duration.
Here's the thing... MU is executing at a level I don't think the memory industry has ever seen. Revenue acceleration at $13.6B scale. Margins at SaaS-like levels. FCF generation that makes enterprise software companies jealous. Management beating guidance by double-digit percentages consistently. The HBM thesis — picks and shovels for AI — is playing out even better than the bull case.
But I refuse to ignore the elephant. This is still a memory company. The cycle giveth and the cycle taketh away. -33% gross margins in FY23. The investor who thinks "this time is different" at peak margins is the investor who gets hurt the most.
What I'd do:
For the March 18 report specifically: expect a monster beat ($19.5-20.5B revenue, 69-71% GM, $9-11 EPS). The pricing data is unambiguous. The real question is the guide and the HBM4 commentary. That's what will move the stock.
long MU ~10%
-muji
Sources: