MU Q3 FY26: Records Continue, the Second Derivative Has Rolled Over

Micron reported fiscal Q3 2026 on Wednesday, June 24, after the close, and the headline is the kind that gets repeated in every market summary for a week:

• Revenue $41.5B, up +345.7% year over year and +73.7% sequentially
• Non-GAAP gross margin 84.9%
• Non-GAAP EPS $25.11
• Adjusted free cash flow $18.3B in a single quarter

It is the highest revenue, highest margin, highest cash quarter the company has ever produced. Memory stocks rallied after the print. The "AI memory supercycle" narrative is now fully consensus.

This piece is about why the rate of change inside those numbers — the part that usually matters more than the headline — has quietly turned.

Records and inflection are not the same thing

Two things can be true at once: a business can be printing all-time records and its rate of improvement can have already peaked. Memory cycles are the textbook example. The headline is always strongest at the top, because the headline is the cumulative result of everything that has already happened. The interesting question is what the next-quarter rate of change looks like.

Stack Micron's last four quarters side by side and the answer is uncomfortable:

Read the bottom rows, not the top.

• Sequential revenue growth went from a peak of +74.9% (Q2) to a guided +20.6% (Q4). That is not a deceleration of the level — it is a deceleration of the rate, which is what cycles do at the top.
• Year-over-year revenue growth rolls over from +345.7% (Q3) to a guided +341.9% (Q4). The YoY line, which usually keeps climbing through a cycle's middle, has effectively flatlined.
• Gross-margin expansion is the cleanest tell. It went +11.1pp → +18.1pp → +10.0pp → guided +1.1pp. The margin number is at a record. The expansion of the margin number is collapsing toward zero. Once that hits zero, the next inflection is downward.
• Operating-margin expansion peaked at +22pp in Q2 and is already halved.

Every second-derivative line that matters in a memory cycle has turned. The first derivative (the level) is still spectacular and will probably remain spectacular for at least the September quarter. Records will continue. But records and inflection are different questions.

Where the growth is actually coming from

Inside the +73.7% sequential revenue jump, management disclosed roughly what the mix of price and volume was:

• DRAM bit shipments grew "low-single-digit" QoQ. ASPs grew "low-60s%" QoQ.
• NAND bit shipments grew "mid-single-digit" QoQ. ASPs grew "mid-80s%" QoQ.

Translation: somewhere on the order of 95% of the sequential revenue gain is price, not volume. The factories are running at capacity, supply is constrained, and customers are paying whatever it takes to get bits. That is what a memory pricing spike looks like.

It is also, historically, the most volatile component of memory revenue. Volume growth on a capacity expansion is durable. Price growth on a supply shortage reverses the moment supply catches up — and Micron is currently bringing new fabs online in Idaho, New York, Japan, Singapore, and Tongluo, Taiwan. Management used the phrase "meaningful moderation in the rate of price increases" on the call. That phrase has never appeared on a memory call near a cycle bottom.

The narrative being built on top

The most interesting part of the Q3 report is not the financials. It is the disclosure changes.

Micron used the quarter to formally introduce a new commercial architecture: 16 signed Strategic Customer Agreements (SCAs) — multi-year, take-or-pay contracts that lock in committed volumes. On the largest of them, price is bounded by a floor and a ceiling set near current-quarter market levels; a handful of smaller agreements are fixed-price or carry no band at all. Alongside them came a metric memory investors have never had to think about: Remaining Performance Obligations (RPO). The figure formally on the books at quarter-end was just over $5B — but the number management led with was ~$100B, the minimum-price RPO across 14 of the 16 agreements signed so far (several inked after the quarter closed), and a floor management was quick to say real revenue should comfortably exceed. The contracts are backed by roughly $22B of customer deposits and related commitments, about $18B of it cash. And the framing was unmistakable: once the full slate is signed, management expects half or more of company revenue to sit under SCAs — though today's signed deals cover closer to a quarter — all in the name of "durability and predictability."

Read that again. A commodity-cyclical memory manufacturer is introducing a multi-year backlog disclosure metric — the kind of disclosure usually associated with software ARR or aerospace book-to-bill — at the moment its sequential growth rate is peaking and its margin expansion is collapsing.

This is the late-cycle move every commodity story makes when management knows the equity needs to be re-rated away from its historical cyclicality multiple. It worked for the steel companies in 2007. It worked for the shipping companies in 2008. It worked for the lithium miners in 2022. In each case the underlying business cycle eventually did what cycles do, and the "structural" framing did not change the destination — only the path the stock took to get there.

Micron's specifics may or may not rhyme with those. Micron's HBM leadership — HBM4 already ramping faster than its prior generation, with over $1B shipped in the quarter — is genuinely scarce in a way that commodity steel is not. The Cloud Memory Business Unit (CMBU) — Micron's hyperscaler-and-HBM segment — is the one place sequential growth is still accelerating (QoQ went +16% → +47% → +78%). That part of the business is closer to a long-lived ramp than a commodity cycle.

But CMBU is roughly a third of revenue. The other two-thirds — Core Data Center DRAM, mobile/PC, automotive — are pure pricing-on-volume-constrained-supply right now. Management itself volunteered on the call that "incremental price yields less in gross margin expansion" — a quiet hedge that the ceiling is approaching.

What we are actually watching for

The Q3 print does not invalidate the AI memory thesis. It tightens the question to a specific empirical test: does the FQ4 guidance get beaten the way the prior three quarters did?

• Q1 FY26: revenue beat guidance midpoint by +9.1%
• Q2 FY26: beat by +27.6%
• Q3 FY26: beat by +23.7%

Q4 FY26 guidance midpoint is $50B. A continuation of the acceleration pattern would put the actual print at $55B or higher. A roll-over would put it closer to the guide. The size of the beat — not whether there is one — is the diagnostic.

Same test on gross margin: the GM beat versus guidance has been +5.3pp → +6.9pp → +3.9pp. Shrinking. If FQ4 GM comes in below 89% (versus the ~86% guide), the topping is confirmed in real time. If it pushes past 89% the pricing power is still surging and this analysis is wrong.

Records will likely continue in September. They will likely be re-labeled as proof that "this cycle is structurally different." Both can be true. The question for an investor is whether the second derivative — the part that the headline does not show you — has turned. The Q3 numbers say it has.