Deep Research

Micron Technology (MU)

4 min read

Snapshot

  • Ticker: MU (Nasdaq)
  • Bucket: HBM
  • Q1 2026 position: 760,426 shares, $256,902,320 — 5.14% of 13F, #5 position overall and the fund’s 4th-largest equity long (after the QQQ put hedge)
  • HQ: Boise, Idaho
  • What it does: One of three global DRAM manufacturers (plus NAND), and the only US-based maker of high-bandwidth memory (HBM) for AI accelerators.

Business Overview

Micron designs and manufactures DRAM (~3/4 of revenue) and NAND flash, selling into data center, mobile, PC, automotive and industrial markets. The business has historically been brutally cyclical — commodity pricing, capacity-driven busts — but AI demand for HBM and high-capacity server DRAM has transformed the model. HBM is sold under long-term agreements with locked pricing, consumes roughly 3x the wafer capacity per bit of standard DRAM (constraining industry supply), and carries structurally higher margins.

In fiscal 2025 (ended August 2025), data center reached a record 56% of company revenue. HBM crossed 9.30B total revenue, +37% y/y). Full-year FY2025 revenue was ~$37B.

Fiscal 2026 has been a step-change. Q1 FY2026 revenue was 23.86B with GAAP net income of 12.07/diluted share) — records across revenue, gross margin, EPS and free cash flow, on tight industry supply. Cloud memory revenue rose more than 160% y/y to 33.5B ± $0.75B — an extraordinary sequential ramp that reflects both pricing and HBM mix.

Financial Trajectory

PeriodRevenueNotes
FY2024 (Aug-end)~$25.1BRecovery year off 2023 memory bust
FY2025~$37.4BData center 56% of revenue; HBM >$1B/qtr from Q2
Q1 FY2026$13.64BHBM4 volume production starts
Q2 FY2026$23.86BGAAP net income $13.79B; record margins
Q3 FY2026 (guide)~$33.5B±$0.75B

Why Atreides Owns It

Micron is Gavin Baker’s loudest public single-stock thesis and the purest expression of the “wafers” half of “watts and wafers.” His argument, made at Sohn Montreal 2025 and repeated on Invest Like the Best EP.473 (“Watts and Wafers,” May 2026): the market prices semicap equipment at ~40x earnings while pricing the memory makers those tools supply at mid-single-digit multiples — a relationship he called “internally contradictory.” Either the equipment makers’ implied growth is wrong, or memory is dramatically mispriced.

His resolution is that HBM structurally re-rates memory. Long-term supply agreements with pre-negotiated pricing, sold-out capacity through 2026, an effective three-player oligopoly (SK Hynix, Samsung, Micron), and HBM’s outsized wafer consumption all break the historical commodity-cycle pattern: memory now looks more like a contracted, visibility-rich AI infrastructure business than a spot-priced commodity. Micron at ~8x forward earnings against a HBM market growing toward ~$100B was, in Baker’s framing, the cheapest large-cap way to own the AI buildout.

The fund’s execution shows trading around a rising conviction: small in Q4 2024, scaled up in Q1 2025, slashed in Q2 2025, rebuilt in Q3 2025. Notably, in Q4 2025 Atreides held puts on 700,000 underlying MU shares ($199.8M notional) alongside its 738,916-share common position — effectively a collar/hedge on the common into the parabolic phase of the move — and closed the puts in Q1 2026 while letting the common run to a top-5 position.

Position History

QuarterTypeShares/NotionalValue% of 13F
Q4 2024Common240,589$20,247,9700.44%
Q1 2025Common889,842$77,318,3712.35%
Q2 2025Common70,478$8,686,4140.24%
Q3 2025Common724,957$121,299,8052.36%
Q4 2025Put700,000$199,787,0002.44%
Q4 2025Common738,916$210,894,0162.58%
Q1 2026Common760,426$256,902,3205.14%

The share count has been remarkably stable since Q3 2025 (~725-760K) while the position’s 13F weight doubled to 5.14% — the appreciation did the work. The Q2 2025 near-exit followed by a Q3 rebuild suggests a tactical de-risk rather than a thesis change; the Q4 2025 put overlay (nearly 1:1 against the common) is the clearest signal Baker wanted to hold the position through year-end volatility without taking the full drawdown risk.

Risks

  • Cycle risk in supercycle clothing: Q3 FY2026 guidance implies ~40% sequential growth; memory has ended every prior “this time is different” cycle with a capacity-driven bust. Non-HBM DRAM and NAND remain commodity-priced.
  • HBM competition: SK Hynix leads HBM share; Samsung is qualifying aggressively on HBM4. Three rational players can become three irrational players when capacity additions land in 2027.
  • Customer concentration: HBM demand is dominated by NVIDIA (and a handful of accelerator buyers); a slowdown in AI capex transmits directly.
  • China supply: CXMT and other Chinese DRAM entrants are adding conventional DRAM capacity, pressuring the non-HBM business, plus ongoing China market-access restrictions on Micron.
  • Re-rating already underway: the multiple-expansion leg of Baker’s thesis has partially played out; from here the position needs earnings delivery, not just a valuation argument.

Sources

Backlinks

Graph View