AI Infrastructure

Micron fiscal Q3: HBM sold out for calendar 2025

Micron's record fiscal Q3: $9.3 billion revenue on June 25, 2025, with HBM up nearly 50% sequentially. CEO Mehrotra said HBM is sold out for calendar 2025; Q4 guidance topped estimates.

Micron fiscal Q3: HBM sold out for calendar 2025 — article cover

After the close on June 25, 2025, Micron reported results for the third quarter of fiscal 2025 (ended May 29): revenue of $9.30 billion, an all-time quarterly record, up 37 percent year over year and well ahead of the $8.87 billion analysts expected. Non-GAAP earnings came in at $1.91 per share versus the $1.60 consensus.

On the earnings call, CEO Sanjay Mehrotra delivered the more consequential line: “our HBM is sold out for calendar year 2025.” AI memory supply, in other words, had moved from sales patter to formal allocation discipline. The stock jumped in extended trading on the news.

A record quarter

The growth engine was unmistakably the data center. Data center revenue more than doubled year over year and hit a quarterly record, data center DRAM revenue set a record for the fourth consecutive quarter, and HBM revenue grew nearly 50 percent sequentially. Data center SSDs hit a record share for the third straight quarter, making Micron the number-two brand by share — the AI server bill of materials now includes more than DRAM. Measured as an annual run rate, Mehrotra said HBM was already “more than $6 billion” — roughly $1.5 billion a quarter, the largest this product line has ever been for the company.

Profitability matched the top line. GAAP EPS was $1.68 on net income of $1.89 billion, up from $332 million a year earlier, and operating cash flow reached $4.61 billion. CNBC noted that through Wednesday’s close, Micron shares were up 51 percent year to date, against a 3.4 percent gain for the Nasdaq — the earnings simply re-confirmed the logic behind that rally.

The quarter also showed capital discipline: net capex of $2.66 billion, $12.22 billion in cash and investments, and a quarterly dividend of $0.115 per share — expanding output and returning cash at the same time.

Sold out: a supply-reversal signal

“Sold out” is heavyweight language in the memory business. HBM (high-bandwidth memory) is the critical companion to AI accelerators, co-packaged with GPUs, and supply has historically expanded to chase demand. The $6 billion run rate and the sold-out calendar point the same direction: supply, not demand, is the constraint. With Micron declaring its calendar 2025 output fully allocated, memory is being redefined from a cyclical commodity into constrained supply.

Allocation talks for 2026 are already underway. Mehrotra said the company is working with customers on 2026 needs and expects HBM bit-demand growth to significantly outpace the overall DRAM industry. Micron also reiterated its target to reach HBM market share similar to its overall DRAM share in the second half of calendar 2025. A year after the memory industry was still climbing out of production cuts, this was a complete reversal in supply-demand posture.

Q4 guidance clears the bar again

For the current quarter (fiscal Q4 2025), Micron guided revenue to $10.7 billion, plus or minus $300 million — up about 38 percent year over year and far above the $9.9 billion average analyst estimate. Non-GAAP gross margin was targeted at 42 percent, plus or minus one point, with non-GAAP EPS of $2.50, plus or minus 15 cents. CNBC’s read: the guidance cleared Wall Street’s consensus with ease.

The market’s immediate reaction was more measured, though: the after-hours pop faded for most of its gains. For a stock that had already rallied more than 50 percent on the year, investors wanted more than a single-quarter beat — they wanted evidence that the sell-out posture and pricing cycle would extend into 2026. That is exactly why the memory trade kept building over the days that followed.

What it means for the memory industry

The report confirmed one thing: AI data centers have turned memory into strategic material. HBM supply is constrained by both advanced packaging and wafer capacity, and the allocations held by SK hynix, Samsung, and Micron directly shape how fast GPU makers can ship. Once major customers lock up a full year of HBM output, contract prices and negotiating structures change with it.

For developers and infrastructure buyers, the signal is just as practical: memory’s share of AI server bills of materials will keep climbing, and capacity planning now has to treat HBM supply cycles as a variable on par with GPU availability. The June 2025 report was the first official stamp on the memory supercycle that followed.

Sources

AI-assisted summary compiled from the sources above, reviewed by a human before publishing.

SHAREXEMAIL