Micron: When Memory Became the Constraint

The quarter that broke the model

Micron sells memory: DRAM, the working memory that sits beside a processor, and NAND, the flash that stores data at rest. For most of its history that made it the textbook cyclical — a maker of an undifferentiated commodity whose price collapses whenever the industry over-builds. The fiscal third quarter it reported on 24 June 2026 does not fit that description. Revenue was $41.46 billion, against $23.86 billion the prior quarter and $9.30 billion a year earlier.¹ A single quarter now out-earns the entire prior fiscal year, when Micron booked $37.4 billion across twelve months.⁷

The question this dossier tries to equip you to answer is not whether the numbers are large — they are — but what kind of business produced them. A cyclical peak looks like this right before it breaks. A structural re-rating looks like this and holds. The difference lives in two things a reader can actually check: where the margins come from, and how much of the revenue is contracted rather than spot.

Read the margin, not the revenue

Revenue scale is easy to mistake for strength; margin is where a memory maker's position shows. In fiscal 2025 Micron's GAAP gross margin was 39.8 per cent.⁷ In the quarter just reported it was 84.6 per cent, with a non-GAAP figure of 84.9 per cent, and management guided the current quarter to roughly 86 per cent.¹ Margins of that order are not what a commodity produces when supply is ample. They are what a scarce input produces when buyers compete for allocation.

The composition confirms it. Micron reports four business units, and the quarter's operating profit is concentrated in the two that serve the data centre. The Cloud Memory unit — which houses high-bandwidth memory for every data-centre customer — turned $13.77 billion of revenue into $10.79 billion of operating income; the Core Data Center unit turned $11.52 billion into $9.52 billion.³ The Core Data Center line is the clearest tell of the cycle's shape: it was $1.53 billion in the year-ago quarter and $5.69 billion two quarters ago before reaching $11.52 billion, at an 87 per cent gross margin.¹

The reason margin can move this far is that memory pricing is non-linear. Fabs run at high fixed cost and near-flat marginal cost, so once demand outruns installed capacity, incremental bits sell at whatever the marginal buyer will pay. In a glut that mechanism destroys Micron; in a shortage it works in reverse, which is the regime the data-centre numbers describe.

Why Micron, and not just anyone

A shortage only rewards the incumbents if newcomers cannot flood in to arbitrage it. Advanced DRAM is one of the few markets where that holds. Micron names its principal competitors as Samsung and SK Hynix among the leaders, with Kioxia and SanDisk in NAND and the Chinese entrants CXMT and YMTC behind them.⁶ Leading-edge DRAM is effectively a three-company business, and the barrier is not secrecy but capital and time: a new fab is a multi-year, multi-billion-dollar project, and Micron's own filings frame semiconductor manufacturing as capital-intensive work demanding specialised expertise that cannot be stood up quickly.⁶

Within that structure Micron argues a process lead. It says it began shipping the industry's first 1-gamma (1γ) DRAM production node in 2025, using EUV lithography for gains in power, performance and bit density, and that its HBM4 — built on the prior 1-beta node — is now in high-volume shipment for its lead customer's platform, with qualification samples out to further end-customers.¹ Node leadership matters here because HBM is bit-hungry: better density means more saleable HBM per wafer in a market where wafers are the binding constraint.

The concentration cuts both ways, and the bull case should say so. Micron notes that roughly half its annual revenue comes from its ten largest customers.⁶ The same handful of hyperscalers and accelerator vendors that are paying today's prices are the ones with the leverage to renegotiate them, and the ones whose capital-expenditure plans, if trimmed, would show up in Micron's order book first.

The case that this time is contracted

What separates this cycle from prior Micron peaks, on the company's telling, is that a growing share of the demand is now written down. Management points to what it calls multi-year Strategic Customer Agreements, which its quarterly filing describes as binding commitments for specific volumes across multiple years, and argues these will "significantly enhance the durability and predictability" of results.¹ If that holds, it changes the security's character: a spot-priced cyclical re-rates toward something closer to a contracted supplier, because the revenue is booked before the spot market can turn.

The honest reading is that the contracts move the question rather than answer it. They convert "will the spot price hold?" into "how much of the volume is locked, at what price, and for how long?" — a question the agreements' headline exists to raise but whose terms the filings do not fully disclose. A reader watching this name should track the share of revenue under those agreements as the single most important disclosure, above any given quarter's beat.

The tape

Walled off from the business, and secondary to it: the equity has re-rated alongside the memory cycle, and the current quote and its 52-week range are worth checking against the fundamentals above rather than in isolation.⁸ With the current quarter guided to about $50 billion in revenue at roughly 86 per cent gross margin, the market is not being asked to price growth so much as durability — whether margins of this order persist once the contracted volumes are delivered and new capacity, Micron's included, comes online.¹ The business is the story; the tape is the footnote.

Go deeper

Start with the primary segment note in the fiscal Q3 2026 Form 10-Q: the four business units, their revenue and operating income, and the language on Strategic Customer Agreements are where the durability case is either supported or not.³ Then read the competition and cyclicality risk factors in the fiscal 2025 Form 10-K, which describe the oversupply mechanism in Micron's own words and are the sharpest available statement of the bear case.⁶ To judge whether the momentum is real or a peak, watch three numbers in the next report: the current-quarter revenue against the $50 billion guide, gross margin against the ~86 per cent guide, and any figure that quantifies HBM volume or the contracted share of revenue.¹ Compare the Q2 and Q3 fiscal 2026 releases side by side to see how fast the data-centre units are compounding.⁵