Date: 2026-02-22 Quarter Reviewed: Q1 FY26 (Nov 2025, reported Dec 2025) Context: Scout brief covers Q3 FY25 transcript; _ROLLING.md updated through Q1 FY26 Market cap ~$108B at time of analysis
Let me start with the numbers that matter most. Revenue:
| | Q2 FY25 | Q3 FY25 | Q4 FY25 | Q1 FY26 | Q2 FY26E | | | Feb-25 | May-25 | Aug-25 | Nov-25 | Feb-26 | |---|---|---|---|---|---| | Revenue ($m) | 8,053 | 9,301 | 13,600 | 13,600 | 18,700 | | YoY % | +38% | +37% | ~+56% | ~+56% | ~+80%+ | | QoQ % | -41% | +15% | +46% | flat | +37% | | Gross Margin % [GAAP] | 36.8% | 37.7% | 56% | 57% | 68% (guide) | | GM % [Non-GAAP] | ~38% | 39% | 56% | 57% | 68% (guide) | | Op Margin % | 22% | 23% | 45% | 47% | ~57% (guide) | | EPS [Non-GAAP] | $1.56 | $1.91 | $4.50 | 4.78| 6.50+ (implied) | | FCF ($m) | 1,500 | 1,900 | 3,000 | 3,900 | — |
Now let me tell you what I see when I look at those numbers. Gross margin went from 37% to 57% in two quarters. Free cash flow went from $1.9B to $3.9B in two quarters. EPS went from $1.91 to $4.78. And the Q2 FY26 guidance is for 68% gross margins! That is such a preposterous result that I had to look at it twice!
This is not a SaaS company — I know that! Micron makes memory chips. But I follow the money. And what the money is telling me is that something fundamental has changed at this company, not something cyclical.
Here is what happened. Artificial intelligence requires enormous amounts of memory bandwidth. Every GPU, every AI accelerator, every custom ASIC needs High Bandwidth Memory (HBM). And Micron — which spent years as the also-ran to Samsung and SK Hynix — has vaulted to the #2 HBM position with 21% market share, up from a rounding error a year ago.
The numbers tell the story:
CEO Sanjay Mehrotra on the Q3 FY25 call: "We are now shipping HBM in high volume to four customers spanning both GPU and ASIC platforms." Four customers — likely the hyperscalers. Concentated? Yes. But these are the most durable customers in the world right now.
And the key promise tracker — every promise management made has been delivered or is on track:
Management said they would do things, and they did them. That matters enormously to me.
Look at this guidance history:
| Quarter | Revenue Guide | Actual | Beat |
|---|---|---|---|
| Q3 FY25 | $8.7B | $9.3B | +7% |
| Q4 FY25 | $10.7B | $13.6B | +27%!!! |
| Q1 FY26 | ~$11.2B implied | $13.6B | +21%!!! |
That Q4 FY25 beat of 27% is extraordinary. Management guided $10.7B and delivered $13.6B. When management beats guidance by 27% and then does it again at +21%, one of two things is happening: either they are sandbagging dramatically, or demand is running so far ahead of their own expectations that even they can't keep up with it. Either answer is bullish!
Now let me be honest with you, because I always try to tell you both sides. Memory is one of the most cyclical businesses in the world. Look at what happened to MU in FY23: gross margin went from +47% to -33%. Revenue fell from $8.6B to $3.7B in two quarters. The company lost billions of dollars. I have seen this movie before.
So when I see 57% gross margins today, the question I have to ask is: will these margins hold, or are we at a cycle peak?
Here is why I think this time is different — or at least more durable than prior cycles:
HBM is not commodity DRAM. Standard DRAM is a commodity — Samsung can ramp capacity and crush pricing in 12-18 months. HBM requires advanced packaging (through-silicon vias, stacking), proprietary yields, and thermal management. Samsung actually failed Nvidia's HBM qualification! This is a technology barrier, not just a price barrier.
Supply cannot ramp quickly. Even if every memory maker wanted to build HBM capacity tomorrow, the equipment lead times, clean room construction, and yield learning curves take years. Idaho Fab 1 won't produce its first wafer until mid-2027. Micron itself says it can only meet 50-67% of customer demand.
Demand is secular, not cyclical. AI compute is not a fad. Every hyperscaler — Google, Microsoft, Amazon, Meta — is spending tens of billions on AI infrastructure. That spend is not discretionary. And memory bandwidth is the bottleneck. You cannot train or run large AI models without enormous amounts of HBM.
The contracts are changing. Management said they are negotiating multiyear LTAs (long-term agreements) with "specific commitments and much stronger contract structure." In prior cycles, memory was pure spot pricing. LTAs reduce the cyclical whipsaw.
But — and this is important — I am not saying the cycle is dead. Consumer electronics is weak. PC DRAM is weak. NAND is weak (Micron has already cut NAND capacity 10%). If AI CapEx spending slows meaningfully, HBM demand could soften. That risk is real and I hold it in mind.
26 out of 28 analysts have Strong Buy ratings. Average price target $405, with Rosenblatt at $500. Morgan Stanley called Q1 FY26 results "the best in the history of US semis."
I pay no attention to analyst estimates for earnings — who cares what some analyst thinks EPS will be? But analyst qualitative assessments about competitive positioning and product roadmaps are worth noting. When 26 out of 28 say strong buy, that is not contrarian alpha. But it is confirmation of what the numbers are showing.
SK Hynix leads HBM at 62% share. Micron is #2 at 21%, up from ~4% a year ago. Samsung fell to 17% after failing Nvidia specifications.
That Samsung failure is crucial. Samsung has historically been the most aggressive capacity builder in memory — when they ramp, everyone's margins get crushed. But if they cannot get HBM4 yields right, their comeback is delayed. Micron shipped HBM4 samples at 11 Gbps vs. SK Hynix at 10 Gbps — Micron is leading on next-gen specifications. That is not where I expected them to be.
Micron also achieved #2 market share in data center SSDs for the third consecutive quarter. The data center memory franchise is expanding beyond just DRAM.
You know I don't usually focus on valuation for growth companies — I focus on the business quality and let the results take care of the price. But here the valuation is so obviously cheap that I have to mention it.
Annualizing Q1 FY26 EPS of 4.78: 19 run rate. At $108B market cap with ~1.1B shares, the stock is roughly $100/share, implying P/E of about 5x. FIVE TIMES EARNINGS! On a company growing revenues 56% YoY with 57% gross margins and a guide for 68% next quarter!
Now — the market is not stupid. The market is pricing in significant margin mean-reversion. It is saying: these 57% gross margins will not last. And historically it would be right. But the historical cycles did not have HBM at the center of a structural AI infrastructure buildout. If margins stay above 50%, MU is dramatically undervalued at 5x earnings.
I have been thinking carefully about whether MU belongs in a growth portfolio. It is not SaaS. It is not recurring revenue. It is a semiconductor company in what is historically the most cyclical business on earth.
But I follow the money. And the money says:
The HBM opportunity is real. The technology leadership is real. The customer commitments are real. The promise tracker shows 100% delivery. And the valuation is absurdly cheap.
This is a Hold/Add for me. The thesis is Intact and Strengthening. At 10.4% of portfolio, wsm007 has a meaningful position. I would not trim this. If we get a pullback related to macro or tariff fears that temporarily clouds the AI capex outlook, that is an opportunity to add.
The main risk I am watching: Samsung qualification recovery in HBM4. If Samsung gets its yields sorted and ramps HBM4 capacity into CY2027, Micron's pricing power could compress. That is the scenario that breaks the thesis. Watch it carefully.
Next quarter (Q2 FY26, expected March 2026): the guide is $18.7B revenue with 68% gross margins. If delivered, it would be the most profitable quarter in Micron's history. I will be watching closely.
Best, Saul