← Back to feed

Micron Technology reported Q3 revenue of $41.46 billion on June 25, with adjusted gross margin reaching 84.6% — up from 39% the same quarter a year earlier. Q4 guidance landed at $49 to $51 billion. The company’s market capitalization briefly crossed $1.27 trillion, and Wall Street is now openly framing the question as whether Micron can do for AI memory what Nvidia did for AI compute.

The comparison has structural backing. Every AI training rack requires high-bandwidth memory stacked directly beside the GPU. Nvidia’s H200 carries 141GB of HBM2e. Blackwell uses 192GB class parts. The GB300 Blackwell Ultra ships with 288GB of HBM3e per chip. As GPU density at the rack scale climbs, HBM demand climbs with it — and Micron is the only US-based manufacturer in that market.

The Supply Agreements That Change the Calculus

Memory chips have historically been a boom-bust business. Factories take years and billions to build. When multiple manufacturers expand simultaneously, the new supply lands together, pricing collapses, and margins reverse. Micron is attempting to break that cycle with a different commercial model: 16 strategic customer agreements worth $22 billion, built around deposits, pricing floors, and take-or-pay terms that make demand harder to cancel.

If those agreements hold, Micron has partially decoupled its revenue from spot market volatility. A customer who has committed to take a fixed volume at a floor price cannot simply step back when a new GPU generation requires less HBM per chip. The question is whether the agreements have enough teeth to survive a genuine demand reversal — and whether Micron can maintain pricing power against Samsung and SK Hynix as those companies invest in their own capacity.

Where Micron Sits in the HBM Stack

The global HBM market is effectively a three-player oligopoly. SK Hynix led with 58% share in Q1 2026. Samsung and Micron each held 21%. Micron’s 21% share understates its strategic position: as the only US manufacturer, it is insulated from the export restriction risk that could strand Asian-made chips in US AI infrastructure.

That geopolitical moat matters more than the market share number. AI data centers built for US government and defense customers cannot source HBM from Chinese manufacturers. CXMT, China’s state-backed memory push, sits on the Pentagon’s Chinese Military Company list. Samsung and SK Hynix are South Korean, not subject to current restrictions — but the last two years of supply chain policy have made procurement teams think more carefully about single-source dependencies.

Standard DRAM Is a Different Story

HBM is a specialty submarket inside DRAM. Standard DRAM — the working memory inside consumer devices, servers, and network equipment — is where the boom-bust logic is most likely to reassert itself.

AI server demand for HBM is pulling memory manufacturers toward high-margin, high-complexity production. That leaves standard DRAM capacity tighter than normal, which is part of what has driven DDR5 prices up sharply in 2026. Apple has reportedly sought permission to source standard DRAM from CXMT specifically because the shortage is raising cost of goods sold on iPhones and Macs. The AI buildout is cannibalizing consumer memory supply while simultaneously creating the HBM scarcity that benefits Micron.

If AI capital expenditure plateaus or a new GPU architecture reduces HBM requirements, the dynamics reverse: HBM oversupply could emerge at the same time standard DRAM supply recovers, and the pricing power Micron is extracting today becomes a prior quarter’s anomaly.

Key Data

• Q3 revenue: $41.46B
• Adjusted gross margin: 84.6% (vs 39% a year prior)
• Q4 guidance: $49–51B
• Long-term supply agreements: 16 agreements, $22B total, with pricing floors and take-or-pay terms
• HBM market share: Micron 21%, Samsung 21%, SK Hynix 58% (Q1 2026)
• Market cap: ~$1.27T at time of results