Micron's HBM output for calendar year 2026 is completely sold out.1 That single fact reframes the semiconductor investment thesis: the bottleneck is supply, not demand — and it stays that way through at least 2027.
Micron has begun shipping HBM for Nvidia's Vera Rubin platform, signaling the next generation of AI infrastructure is already ramping.1 Q3 FY2026 revenue guidance sits at ~$33.5B — a figure that would exceed Micron's revenue for any full prior fiscal year.1 The company has crossed $1 trillion in market capitalization.
Sandisk tells the same story from a different angle. The company's CEO identified inference and reasoning workloads as the fastest-growing segment in data center demand.1 Sandisk shares have surged over 700% year-to-date in 2026.1
The valuation setup is unusual. Both Micron and Sandisk trade at 10-11x forward earnings despite triple-digit stock gains.1 That compression signals earnings are growing faster than share prices — a condition that typically precedes further multiple expansion or sustained outperformance.
The structural driver is inference at scale. Unlike training runs — discrete, capacity-bound events — inference workloads are continuous. Every query processed by a deployed AI model generates persistent memory demand. As reasoning models grow more complex, per-query memory consumption rises. More models deployed means more bandwidth consumed around the clock.
HBM (High Bandwidth Memory) sits at the center of this. It is the memory architecture of choice for AI accelerators, and Micron is one of a handful of suppliers globally capable of producing it at volume. With 2026 allocation already committed, new capacity coming online in 2027 and beyond will determine pricing power and margin trajectory.
The test for this thesis is straightforward: track HBM average selling prices and gross margins over the next four quarters. If ASPs hold while supply remains constrained, margin expansion continues. Watch also for Micron capacity announcements and whether hyperscaler supply contracts are renewed or extended.
At forward P/E multiples of 10-11x, the market is not pricing in a supercycle. That gap between earnings trajectory and valuation is the investment thesis.
Sources:
1 Via News Market Intelligence — AI Hardware Supercycle Signal Report, June 19, 2026
