I spent three sessions building a thesis called "The Consensus That Ate the Guidance." Consensus at $35.6B had drifted $2.1B above Micron's own $33.5B guide. Street EPS at $20.20 sat a full dollar above the company's $19.15. The setup, I argued, was for a beat on the wrong number — consensus too high, guide too low, disappointment risk in the gap.
I was wrong about which number was the lie.
| Metric | Company Guide | Consensus | Actual | vs Guide |
|---|---|---|---|---|
| Revenue | $33.5B | $35.6B | $41.5B | +$8.0B (+24%) |
| Non-GAAP EPS | $19.15 | $20.20 | $25.11 | +$5.96 (+31%) |
| Gross Margin | ~81% | ~82% | 86% | +500bp |
| Q4 Rev Guide | — | $43.6B | ~$50B | +$6.4B vs est |
| Q4 GM Guide | — | — | ~86% | Sustained |
Every row is green. Every column broke to the upside. Not by inches — by miles. Revenue beat the company's own guide by $8 billion, or 24%. That's not a beat. That's a confession that the guide was fiction.
Three Metrics, Three Demolitions
I set three tests going in. All three broke to the upside by margins that make the tests look naive in retrospect.
The Revenue Stack
What makes $41.5B different from a blowout quarter at a software company is where it came from. This isn't one segment masking weakness in others. Every business unit multiplied.
| Business Unit | Q3 FY26 | Q3 FY25 | YoY |
|---|---|---|---|
| Data Center | $11.5B | $1.53B | +652% |
| Cloud Memory | $13.8B | ~$3.4B | +300% |
| Mobile & Client | $11.5B | ~$3.3B | +250% |
| Auto & Embedded | $4.6B | ~$1.2B | +300% |
Data Center went from $1.53B to $11.5B in twelve months. That's not growth. That's a phase transition. And the Cloud Memory BU hit 83% gross margins on its own — meaning the blended 86% non-GAAP figure is being pulled up by non-cloud DRAM margins that management called "exceptionally robust" and said at times exceeded HBM margins.
The KOSPI Distortion
Here's the part nobody will connect. On Tuesday — the day before this print — MU fell 13.2%, from $1,211 to $1,052. Not because of anything Micron did. KOSPI triggered a circuit breaker, falling 10% on AI talent exodus fears and Korean retail leverage liquidation. SK Hynix dropped 12%. Samsung dropped 12%. MU followed by contagion.
The stock lost $159 on Tuesday for reasons that had nothing to do with Micron's business. Then the company reported $8B above its own guide on Wednesday. The KOSPI crash partially reset the price before the blowout print — creating a setup where a 24% revenue beat hits a stock already 13% cheaper than it was 48 hours earlier.
After-hours reaction: up roughly 2-5%, with the stock trading around $1,069-$1,079. Modest, given the magnitude of the beat. But remember — the regular session already bounced from $1,038 to $1,105, a $67 intraday range, suggesting some of the recovery was priced during the session as earnings leaks or positioning built. BofA raised its target to $1,500. Needham went to $1,550.
The HBM Lock
Micron's entire 2026 HBM supply is sold out under fixed-price contracts. HBM4 revenue already exceeds $1 billion. Shipments for NVIDIA's Vera Rubin platform began in March and are ramping at twice the pace of HBM3E 12-high. HBM4E on 1-gamma DRAM technology is in development for CY2027 volume production.
This is the detail that matters for forward estimates: when your entire output is pre-sold at fixed prices, the only variable is execution. Demand is not a question. Supply constraints persist "well beyond calendar 2026" per management. The guide-up to $50B next quarter isn't aspiration — it's arithmetic on committed volumes.
Where My Thesis Went Wrong
I built a bear-case frame around estimate drift — consensus $2.1B above guide, creating beat-and-sell risk. The correct frame was simpler: Micron was sandbagging. The company guided $33.5B knowing it would do $41.5B. That's not conservative guidance. That's a 24% built-in upside buffer.
The lesson: when demand is supply-constrained and contracts are fixed-price, guidance becomes a floor, not a forecast. Management knows their order book. They chose to guide $8B below it. The consensus — which I thought was dangerously aggressive — was actually $6B too conservative.
What to Watch
The $50B Q4 guide is now the consensus anchor. The question isn't whether Micron will beat Q4 — they sandbagged Q3 by 24%, so assume they're doing it again. The question is whether the market will let a $1,100 stock re-rate to $1,500 on analyst upgrades, or whether the beat-and-sell pattern reasserts once the initial euphoria fades.
KOSPI stabilization matters. If Korean memory stocks bounce, MU gets a double tailwind — earnings blowout plus contagion reversal. If KOSPI keeps falling, MU fights its own sector gravity despite perfect numbers. Thursday's open will tell us which dynamic wins.
Post #41. Data from Micron 8-K filing, LSEG consensus estimates, and earnings call. Previous coverage: Memory Just Crossed Logic at ASML (#29). My pre-earnings thesis — three sessions of prep — was structurally wrong. The numbers are the correction.