Mizuho's Vijay Rakesh Walks Micron to $740 on the Agentic AI Memory Call

The $740 Print

The catalyst is concrete. Mizuho's Vijay Rakesh raised his Micron price target to $740 from $545 while reiterating Outperform, citing "agentic AI driving memory demand higher." That is a 36% target raise in a single revision — not a tweak around the edges of an existing thesis. Rakesh is one of the most-followed semis analysts on the sell side, and the Street tends to mark to his model on memory; this is the kind of note that moves consensus rather than just adds to it.

The specific argument is that autonomous, multi-step AI agents consume meaningfully more memory bandwidth and capacity per query than chat-style inference, and that the Street has been modeling memory demand off the wrong workload. If you accept that framing, the implied 2027–2028 demand curve is steeper than what's already in HBM4 supply commitments.

What the New Numbers Imply

Rakesh's revised model is what makes this more than an analyst headline. He took FY2026 revenue to $109B, FY2027 to $181B, FY2028 to $179B, with EPS at $58.16, $104.74, and $94.40 across those years. FY2027 revenue at $181B implies roughly +66% YoY on top of an FY2026 base that already reflects the AI ramp.

For reference, Micron's most recent quarter delivered $23.9B in revenue with a 75% gross margin and adjusted EPS of $12.20, so Rakesh is essentially modeling that the magnitude of the current quarter persists and accelerates rather than mean-reverts. That is the variant view: the bear case has always been that memory is cyclical and 75% gross margins don't last. Rakesh is saying agentic AI changes the cycle's shape because demand is being structurally repriced, not temporarily squeezed.

Why It Stuck

The note didn't drop into a vacuum. TD Cowen had already confirmed earlier this week that Micron's entire 2026 HBM book — HBM4 included — is locked on pricing and volume, with multi-year customer commitments replacing the usual quarterly negotiation. Bernstein, separately, has DRAM spot up 57% and NAND up 65–70% in April versus Q1 averages, so the spot tape is already validating the pricing assumptions baked into the new estimates.

The HIP-3 perp on Hyperliquid is trading at a notable premium to spot MU here, which is a familiar signature for stock perps when the underlying gaps into a fresh narrative on a Friday — leveraged longs front-run the next session's cash open. The book on a $58.6M-volume perp will not absorb size the way the listed name does, so the path back to convergence runs through either continuation in cash on Monday or a basis unwind. The catalyst itself, though, is not in dispute.