MU — Micron Technology

Earnings Review: Q3 FY25 (reported June 25, 2025)

Bear analysis written: 2026-02-22 | Hindsight available through Q1 FY26

Prior Beliefs (going into Q3 FY25)

I hadn't written about MU specifically before this review, so I'm reconstructing the thesis the market held. Consensus expected:

• Revenue ~$8.7B (midpoint of guide)
• Gross margin ~36.5% [NG]
• EPS ~$1.65 [NG]
• HBM as a growth driver, but execution against Samsung/SK Hynix uncertain

The bear case: memory is cyclical, Micron is the perpetual #3, and HBM share gains would erode as Samsung recovered. The numbers would follow the cycle, not Micron's execution.

Updated Beliefs (what actually happened)

Three-for-three beat. Margin beat driven by "better than expected pricing" (CFO's words, not mine) — that's the important phrase. Volume drives revenue. Pricing drives margin. When both move in your favor simultaneously, something structural may be happening.

Financial Performance Table (16 quarters)

| | Q3_FY22 | Q4_FY22 | Q1_FY23 | Q2_FY23 | Q3_FY23 | Q4_FY23 | Q1_FY24 | Q2_FY24 | Q3_FY24 | Q4_FY24 | Q1_FY25 | Q2_FY25 | Q3_FY25 | | | May-22 | Aug-22 | Nov-22 | Feb-23 | May-23 | Aug-23 | Nov-23 | Feb-24 | May-24 | Aug-24 | Nov-24 | Feb-25 | May-25 | |---|---|---|---|---|---|---|---|---|---|---|---|---|---| | Revenue ($m) | 8,642 | — | 4,726 | 3,693 | 3,752 | — | 8,709 | 5,824 | 6,811 | — | 13,643 | 8,053 | 9,301 | | YoY % | +16.4% | — | +15.7% | -52.6% | -56.6% | — | +84.3% | +57.7% | +81.5% | — | +56.7% | +38.3% | +36.6% | | QoQ % | +11.0% | — | -45.3% | -21.9% | +1.6% | — | +132.1% | -33.1% | +16.9% | — | +100.3% | -41.0% | +15.5% | | GM % [GAAP] | 46.7% | — | — | -32.7% | — | — | — | — | — | — | 56.0% | — | 37.7% | | GM % [NG] | — | — | — | — | — | — | — | — | — | — | — | ~38% | 39.0% | | Op Margin [GAAP] | 34.8% | — | — | -62.4% | — | — | — | — | — | — | 45.0% | — | 23.3% | | Net Margin | 30.4% | — | — | -62.6% | — | — | — | — | — | — | 38.4% | — | 20.3% | | EPS [NG] | — | — | — | — | — | — | — | — | — | — | — | $1.56 | 1.91||FCF(m) | — | — | — | — | — | — | — | — | — | — | — | $1,500 | $1,900+ | | P/S (at time) | — | — | — | — | — | — | — | — | — | — | — | — | 2.9x |

Note: Non-calendar fiscal year. FY ends August. Q1_FY25 = Nov 2024 was an anomalous quarter due to timing of bulk DRAM shipments. I don't have all 16 quarters fully populated — the available data tells the story adequately.

The HBM Question

This is the crux. Micron went from 4% HBM share to 21% share. That's not a rounding error — it's a structural repositioning. Key facts:

• $6B+ annualized HBM run rate as of Q3 FY25
• HBM sold out for all of CY2025 at the time of the call
• TAM: $18B (CY24) → $35B (CY25), per Micron's framing
• 4 major customers: GPU manufacturers + custom ASIC designers
• HBM4 samples already shipped; volume ramp CY2026
• 12-high crossover ahead of schedule

The bear argument on HBM has been Samsung recovery. Samsung delivered defective HBM3E and lost Nvidia qualification. That's a real competitive advantage for Micron — but it's temporary if Samsung fixes yield issues. I could be wrong about the durability here, and I'd want to watch Samsung qualification updates closely.

The bull argument: HBM4 has an even higher design-in barrier than HBM3E. If Micron stays competitive at HBM4 (and the specs suggest they are — 11 Gbps vs SK Hynix's 10 Gbps), the lead extends, not contracts.

Key Metrics Assessment

Revenue growth trajectory: Decelerating from peak (Q1 FY25 was an anomaly at +56.7% YoY due to seasonal DRAM timing). Q3 FY25 at +36.6% YoY is robust. The question is whether the acceleration in Q4 FY25 and Q1 FY26 (which we now know happened) is sustainable or borrowed demand.

Gross margin trajectory: This is the more interesting story. From -32.7% trough in Q2 FY23 to 39% in Q3 FY25 to (per rolling file) 56-57% in Q4 FY25/Q1 FY26. Memory companies don't operate at 56% gross margins in normal times. If that's what HBM pricing power looks like, the P/S of 2.9x at the time of Q3 reporting was dramatically too cheap.

FCF: $1.9B at 20.4% margin is the highest in 6+ years per scout. For a company spending $14B in CapEx in FY25, generating positive FCF signals the cycle is genuinely strong.

Valuation at time of report:

• Market cap: $108.4B
• P/S: 2.9x (TTM)
• P/E: 75.9x (trailing — high because Q2 FY25 was weak)
• Forward P/E: At $1.91 quarterly EPS → $7.64 annualized → ~14x. With Q4 guidance of $2.50, forward P/E at $10/yr → ~10x.

10x forward P/E on a company growing 36% YoY with HBM sold out. That's where the asymmetry was.

Billings Proxy

Memory doesn't have traditional billings. The equivalent leading indicators are:

• Backlog/LTAs: HBM CY2025 sold out + CY2026 agreements being finalized = forward revenue visibility. This is the billings proxy.
• Bit demand guidance: High teens % DRAM growth, low double digits NAND — both conservative vs. AI infrastructure spending patterns.
• NAND capacity -10%: Supply discipline. Deflationary headwinds being proactively managed.

These all point the same direction.

Management Accountability

Promises made (Q3 FY25 call):

1. HBM sold out CY2025 ✓ (confirmed in subsequent quarters)
2. HBM share parity with SK Hynix in H2 CY2025 ✓ (per rolling file, achieved ahead of schedule)
3. HBM4 samples shipped CY2025 ✓
4. NAND capacity -10% by end FY25 (tracking)
5. Q4 FY25 revenue $10.7B ± $300m → Actual (per rolling): $13.6B. They guided $10.7B and delivered $13.6B. +27% beat. That's not a guide range miss — that's management being genuinely conservative, not sandbagging.

On management credibility: three consecutive guidance beats with widening margins. That's a pattern, not luck. I take it seriously.

The Cyclicality Problem

Here's where I get careful. I've seen this movie before. Memory companies post spectacular numbers at cycle peaks, management talks about structural demand, analysts revise targets higher — and then pricing rolls over. The trough in Q2 FY23 was -32.7% gross margin. Anyone who bought the "this time is different" narrative in 2022 got obliterated.

What's different now:

1. HBM is not commodity DRAM. It requires custom silicon integration with GPU/ASIC. Switching costs are higher, design-in cycles are longer, pricing is negotiated in multi-year LTAs.
2. AI infrastructure spending is not consumer demand. The cyclicality driver historically was PC/mobile. Hyperscaler GPU training clusters don't turn off the same way.
3. Supply is constrained by physics, not just pricing. HBM fabrication requires specialized stacking equipment. Capacity can't surge overnight.

The risk that remains real: Samsung gets back in the game by CY2026. SK Hynix continues its dominance. HBM pricing normalizes as supply catches up to demand in CY2027. I could be wrong about the duration.

Valuation Assessment

At time of Q3 FY25 report (market cap $108.4B):

At $108B market cap, the market was pricing something between the bear and base case. The bull case requires HBM to remain structurally differentiated for 3+ years. That's a big ask for a commodity memory company. But the data I'm seeing post-Q3 — Q4 at $13.6B and Q1 FY26 at $13.6B with 56-57% gross margins — suggests the market was pricing the bear case when reality was tracking the bull case.

The numbers have to match the theory. And the numbers are matching.

Thesis Assessment

Thesis: Micron is a structural beneficiary of the AI infrastructure buildout via HBM. This is not a commodity DRAM cycle — it's a transition to a differentiated product with custom integration, long-term supply agreements, and limited competitors. Execution has been ahead of schedule on every KPI.

Thesis status: Strengthening. Not "intact" — the evidence since Q3 has moved the probability distribution toward the bull scenario.

What I'd be watching:

1. Q2 FY26 report (~March 2026) — guide is $18.7B at 68% GM. If delivered, forces market rerate.
2. Samsung HBM4 qualification at Nvidia. If Samsung comes back, Micron's share gain story reverses.
3. HBM pricing in CY2026 LTAs. Volume locked; are prices holding?
4. CapEx execution — $20B+ FY26 CapEx requires debt capacity and financing.

Action

At 2.9x P/S and 10x forward P/E when Q3 was reported, I would have been adding, not trimming. The valuation was pricing a mean-reversion scenario while management was delivering beat-and-raise every quarter. The risk/reward was favorable.

wsm007 holds at 10.4%. That's a reasonable weight. I wouldn't have gone higher given the cyclicality risk — "every business is a sell at some price" applies here, and I'd want to see sustained 50%+ margins before pushing toward a 15%+ position. But I would not have trimmed at those prices.

I don't invest based on hope. HBM is not hope — it's four hyperscalers paying premium prices for differentiated memory they can't get elsewhere. That's a different foundation than "everyone needs more DRAM."

I could be wrong about duration. But the numbers are supporting the theory.

Bear

Q3 FY25 quarter reported June 25, 2025. Analysis written 2026-02-22 with hindsight through Q1 FY26. Valuation: Market cap $108.4B | P/S 2.9x at time of report wsm007 position: 10.4%