Micron's Q3 earnings blew past every estimate on record, briefly pushing its market cap above $1.4 trillion and forcing a rethink of who actually wins when AI infrastructure spending accelerates.
On June 25, Micron Technology's stock touched $1,255 before settling around $1,236, a move that briefly lifted the chipmaker's market capitalization to roughly $1.418 trillion, clearing both Tesla at $1.403 trillion and Meta at $1.384 trillion. It was the first time Micron had ever topped either company, and it didn't happen because of a narrative or a product launch. It happened because the numbers were simply that large.
Fiscal Q3 revenue came in at $41.46 billion, up 346% from $9.3 billion a year earlier, according to the company's earnings release. The Wall Street consensus was $35.3 billion. Non-GAAP earnings per share of $25.11 crushed the $20.28 estimate. Gross margins crossed 81% for the first time in company history, driven by high-bandwidth memory chips that are now effectively sold out through the end of 2026. Customers have committed $22 billion in upfront cash deposits to lock in supply across contracts running three to five years, and 14 of Micron's 16 major supply agreements carry cumulative minimum revenue commitments totaling roughly $100 billion, as the company disclosed in its earnings call prepared remarks.
For anyone still mapping AI infrastructure investment as a straight line from chips to Nvidia and nothing else, that quarterly print is the correction. Memory has quietly become as load-bearing as compute, and Micron is the company that makes it. HBM4 chips supply the bandwidth that makes Nvidia's GPUs useful at scale: without fast, dense memory sitting close to the processor, the math of a large model doesn't work. Micron disclosed that HBM4 revenue has already crossed $1 billion, and the entire 2026 production run is gone.
Three days before earnings, on June 22, Micron announced a strategic agreement with Anthropic covering multi-year supply of HBM, DRAM, and SSDs, co-design work on memory and storage architecture for AI data centers, internal deployment of Claude across Micron's own operations, and a strategic investment in Anthropic's Series H round. That round raised $65 billion at a post-money valuation of $965 billion, with Samsung, SK Hynix, Altimeter Capital, Sequoia, and Amazon already in. With Micron's participation, as The Next Web noted, all three global HBM suppliers are now Anthropic infrastructure partners.
That's not a coincidence. Anthropic is building the kind of training and inference infrastructure that consumes memory at a scale that makes supply security an existential question. Locking in Micron before the supply crunch deepens is exactly the kind of deal that looks obvious in retrospect. For Micron, it's validation that the company's biggest customers aren't just buying chips: they're building long-term architecture relationships with whoever can actually deliver.
Reuters reported that Micron projected fiscal Q4 revenue of approximately $50 billion, against a year-ago figure of just over $11 billion and analyst expectations of around $44 billion. The stock is up roughly 700% over the past year. For context, that's the kind of return that rewires how investors think about the entire stack.
Here's the thing most commentary on the AI trade misses: Nvidia gets the attention because it's the most visible constraint. But memory is the quieter one, and in some ways the harder one to solve. You can design a new GPU architecture in a few years. Building HBM fabrication capacity takes longer, costs more, and involves a much smaller set of companies capable of doing it. Micron, Samsung, and SK Hynix are essentially it. That oligopoly structure, combined with demand that is doubling and then doubling again, is what produces 346% revenue growth and $22 billion in prepayments from customers who cannot afford to be wrong.
For founders and VCs evaluating where the real infrastructure plays are, Micron's quarter makes the argument plainly. The Nvidia-only framing of AI infrastructure investment was always too narrow. Memory, power, cooling, networking: each layer has its own supply constraint, and the companies that own those constraints are compounding in ways the market is only beginning to price. Micron briefly being worth more than Meta isn't a footnote. It's a signal about where the leverage in this cycle actually sits.
Also read: Samsung and SK Hynix pledge over 1,000 trillion won to dominate the AI memory race • Baidu's Kunlunxin is chasing a $50 billion Hong Kong IPO with a condition investors have rarely seen • Google DeepMind is losing the researchers who built its best work and the damage is already showing