FQ3-26 revenue of $41.46B and non-GAAP EPS of $25.11 both crushed guidance and consensus. Sixteen take-or-pay contracts turned 'storage as a structural asset' from narrative into booked cash flow. After-hours stock rose 13%, nearing all-time highs, while the rest of the AI complex quietly pulled back.
Micron's FQ3-26 (ended May 28) delivered a report that beat guidance and consensus on nearly every metric. Revenue of $41.456B, up 74% QoQ and 346% YoY, far exceeded its own guidance midpoint of $33.5B and consensus of ~$35.8B. Non-GAAP gross margin of 84.9%—up another 10 points from last quarter's 74.9%—set a company record. Non-GAAP diluted EPS of $25.11 beat consensus of ~$20.3 by nearly 24%. More importantly, the next-quarter guidance: FQ4 revenue pegged at $50B ± $1B, gross margin ~86%, EPS $31 ± $1—this single guide blew past the market's prior expectation of ~$42.9B for next quarter. After-hours shares rose about 13% to ~$1,186, approaching the 52-week high of $1,213.56.
Before earnings, I gave you a three-scenario framework: a beat was already priced in; the true determinant was the guidance, not the numbers themselves. The combination was either 'beat + cautious guide → sell the news' or 'beat + raised guide → narrative continues.' This quarter landed on the extreme version of the latter—not just a beat, but revenue nearly $8B above its own guidance midpoint, with next-quarter guidance taking another leg up. The market answered with a 13% after-hours gain: no sell-the-news this time. But more important than the stock move is what this report also disclosed.
The answer: 16 strategic customer agreements (SCAs). For the past year, the bull-bear debate on memory stocks—UBS's $1,625 price target based on 'cyclical stock turning structural asset' versus Goldman Sachs's $400 target with 'memory is always cyclical, current margins are peak, not floor'—has remained a narrative and model exercise, unfalsifiable by either side. This quarter, take-or-pay contract structures dragged that debate into real-world adjudication: roughly 40% of revenue is now locked into floor-to-ceiling pricing bands, with ~$10B in minimum-price contract revenue and $22B in customer deposits and commitments written into contracts. The structural camp has booked cash flows for the first time, not just a PowerPoint slide. But buried in those same contracts is a detail neither bulls nor bears have priced in yet.
数字：指引与共识的双重碾压
This quarter was 'beat++' among the three pre-earnings scenarios. Revenue exceeded its own guidance midpoint by nearly $8B, and the next-quarter guide itself constitutes a beat. DRAM and NAND ASPs rose in tandem; data center annualized revenue topped $100B.
Price, not volume, drove virtually all the growth. DRAM revenue hit a record $31.3B, 76% of total, up 343% YoY, with QoQ ASP up in the low-60% range while shipments rose only low single digits. NAND revenue of $9.9B, 24% of total, up 361% YoY, with QoQ ASP up in the high-80% range. Together they accounted for all revenue growth. This was a pure price-driven quarter, with shipments barely moving—that's both its strength and the risk I'll discuss later.
Structurally, data center revenue crossed $25B for the quarter, annualizing above $100B. Core data center revenue was up 653% YoY; data center SSD revenue broke $5B, doubling QoQ. This line has moved from 'storage supplier' to 'standard component of the AI rack.' All four business units set records, gross margins all above 79%: Cloud Storage (CMBU) $13.8B (83% margin), Core Data Center (CDBU) $11.5B (87%), Mobile & Client (MCBU) $11.5B (87%), Automotive & Embedded (AEBU) $4.6B (79%). Operating profit of $33.68B, operating margin of 81.2%, up 54 points YoY. Operating cash flow of $25.4B and free cash flow of $18.3B were both records. Net cash at quarter end was $22.4B. All three major rating agencies upgraded Micron to BBB+ during the year.
| Metric (Non-GAAP) | FQ3-26 Actual | Company Guidance Midpoint | Consensus |
|---|---|---|---|
| Revenue | $41.46B | $33.5B | ~$35.8B |
| Gross Margin | 84.9% | ~81% | — |
| Diluted EPS | $25.11 | $19.15 | ~$20.3 |
| FQ4 Revenue Guide | $50B ± $1B | — | Prior ~$42.9B |
真正的事件：16份长协把"结构"做实
This is the part of the report most worth analyzing. Take-or-pay, five-year terms, ~$10B in minimum-price contract revenue, $22B in deposits and commitments—the UBS 'long-term contract thesis' that supported its $1,625 target went from model to binding contract for the first time.
2.1 条款本身
Sixteen SCAs are typically five years, covering about 20% of DRAM shipments and about one-third of NAND shipments. They include four very large customers, three mid-sized customers, and several automotive small customers. Once all planned SCAs are in place, the company expects roughly half or more of revenue to be under agreements. The structure is take-or-pay—customers make binding purchase commitments for specific volumes. As of the end of FQ3, remaining performance obligations (RPO) exceeded $5B; after executing post-quarter agreements, that figure is ~$10B. Of these, 14 contracts have cumulative minimum-price revenue of ~$10B. The company expects to receive customer deposits and related financial commitments of $22B, of which ~$18B is cash deposits, mainly reflected on the balance sheet in FQ4.
2.2 为什么这是结构派论据的实物兑现
The key provision is the pricing band: the largest agreements set a ceiling at current CQ2 market prices for existing products and a floor for the entire term. Management explicitly stated that for SCAs with pricing bands, the floor price can support 'gross margins far above any historical cyclical peak quarterly margin.' This directly addresses Goldman Sachs's core objection—'current margins are peak, not floor.' When roughly 40% of revenue is propped up by a contractual floor, the logic that 'memory will inevitably fall back to commodity pricing' loses its most critical transmission mechanism: spot price declines can no longer proportionally smash contract revenue. The structural camp didn't win on narrative this round—they won on contract terms.
结语：四个可验证的节点
This quarter adjudicated the 'cycle vs. structure' debate with real-world evidence: structure is real. But the real test isn't whether structure is true—it's whether it's locked in near the top. Four things to watch.
First, the actual FQ4 numbers (September)—can the $50B revenue and 86% gross margin be delivered, or will 'slowing price hikes' spread from ASPs to shipments? Second, the magnitude of the capital return increase after December 9—that's management's true read on reinvestment runway. Third, the position of spot DRAM/NAND prices relative to the SCA ceiling—if they stay above the ceiling, the premium goes to customers, and Micron gives up upside; if they fall toward the floor, the structural band faces its first stress test. Fourth, whether SK Hynix in Q2 (end-July) discloses equivalent take-or-pay structures—if yes, the 'pure memory beta' valuation anchor shift will complete across both Korean and US markets; if not, Micron's contract visibility becomes its only—and hardest—premium over Hynix.
Risk disclosure: The views expressed in this article are for reference only and do not represent any investment advice. Markets are risky; invest with caution.
On investment analysis, market insight, and asset allocation — not chasing short-term noise, but understanding what truly drives long-term returns.
