Micron: The Market Is Selling Sentiment, Not Fundamentals

Micron shares have pulled back on sentiment spillover from SK Hynix's earnings miss, and the tape is treating a demand story as if it were cracking. It isn't. What's actually changed over the last week is that the evidence for the thesis got stronger, not weaker — and there's now a hard date on the calendar that resolves the single biggest capital-allocation constraint on the stock.

SK Hynix posted operating profit up more than 550% year-over-year and record absolute profit, and the stock still sold off because the print landed below consensus. Management's own explanation matters here: the shortfall wasn't a demand problem, it was a mix problem — more HBM volume moving onto long-term contracts constrains near-term profit flexibility versus spot pricing. That's a company locking in years of forward revenue at the expense of this quarter's beat. Read correctly, it confirms the structural pricing environment rather than undermining it. DRAM and HBM pricing across the industry remains elevated; the miss was about contract timing, not softening orders.

Micron's own setup looks, if anything, cleaner. Fiscal Q3 2026 revenue hit $41.46B (+346% YoY) at 84.9% gross margin, with Q4 guided to $49–51B and non-GAAP gross margin near 86%. That guide implies operating income in the low-$40B range on an operating margin approaching 83% — ahead of SK Hynix's just-reported 76.3% — despite Micron holding meaningfully less DRAM share. The mix, not the market-share ranking, is doing the work.

Micron's entire HBM output is locked in under fixed-price contracts through 2026 and, per multiple sell-side channel checks including Morgan Stanley's Joseph Moore, well into 2027. New capacity doesn't change that picture soon: Idaho 1 doesn't contribute meaningful volume until H2/mid-2027, the Taiwan (Tongluo) acquisition doesn't add wafer output until H2 2027, and the New York complex isn't online until 2030. For the next several quarters, Micron is selling out of a fixed asset base into a demand curve that keeps re-rating higher — Moore's latest checks point to a further 25%+ pricing step-up between Q2 and Q3 calendar 2026, with 2027–2028 conditions expected to tighten further as AI workloads keep scaling.

On Tesla's Q2 2026 earnings call, Elon Musk thanked Micron by name — twice — for a "very significant allocation" of memory on "reasonable terms" at a moment he described current memory pricing as "pretty insane." This is not something Musk does. He was candid that Micron had to make "very tough decisions on memory allocation" to make room for Tesla, and circled back later in the call to thank Micron again alongside TSMC and Samsung. Context matters: this comes as Tesla is simultaneously funding its own $55B Terafab project specifically so it isn't dependent on outside memory supply going forward — which is itself the tell. The world's most vertically-integrated manufacturer, with every incentive to internalize supply, is publicly grateful for an allocation today and building around the constraint for tomorrow. That's a scarcity signal from the most demanding customer in the chain, not a one-off pleasantry.

Tesla's allocation lands alongside a run of new Strategic Customer Agreements that extend Micron's demand visibility well outside data centers. General Motors signed an SCA on July 1st and Ford followed on July 6th — both securing committed, long-term supply of LPDRAM, NOR, and UFS NAND for next-generation vehicle platforms, with GM's deal specifically covering ADAS and AI-enabled in-cabin systems. Both were named as part of the 16 SCAs Micron disclosed on its fiscal Q3 call, a roster that already includes Anthropic on the AI-compute side. The pattern across these deals is consistent: multi-year, take-or-pay commitments that lock in both price and volume, backed by Micron's own capacity investment (the $2B Manassas upgrade underpinning the automotive agreements specifically). Every new signature narrows the pool of supply available at spot and extends the visibility window on the revenue base — this is the SCA/LTA structural pillar of the thesis playing out in real time, not a hypothetical.

If the SCA book confirms broadening demand outside data centers, this week's Big Tech earnings confirm the core data center demand isn't slowing either — it's accelerating, and memory is now an explicit line item in why.

Meta reported Q2 2026 revenue of $60.8B, beat on the top line, missed EPS on the spend, and raised full-year 2026 capex guidance to $130–145B (from $125–145B) — the third upward revision this year. Management's own language on the driver is the relevant part for this thesis: Meta said the increase reflects expectations for higher component pricing this year, in addition to data center buildout, and it committed to a new $14B Texas facility with BlackRock alongside the raise. That's a hyperscaler telling investors, in its own earnings release, that memory cost inflation is now material enough to move full-year guidance.

Microsoft closed fiscal 2026 with Q4 revenue of $90.0B (+18% YoY) and Azure surpassing $100B in annual revenue for the first time, growing 43% in the quarter on capacity that customer demand keeps outrunning. The number that matters most here: Microsoft guided FY2027 capex to $255–260B, up roughly 35% from FY2026's $190B, with quarterly capex set to top $50B next quarter and management describing plans to double total data center footprint within two years. This isn't a company pulling back — it's accelerating the buildout precisely as AI compute demand outpaces available capacity.

Two of the largest hyperscalers on earth just told the market, within 24 hours of each other, that they're spending more, not less, and that component/memory pricing is now explicitly part of the reason why. Every dollar of that incremental capex converts into DRAM, HBM, and NAND content at a wafer intensity ratio that only rises as the industry shifts toward AI-optimized silicon. This is the demand side of the Micron thesis showing up in the exact place skeptics said to watch — the hyperscalers' own capex lines — in the same week the stock sold off on an unrelated Korean earnings miss.

None of the above changes the near-term capital allocation picture — that's mechanical, not sentiment-driven. Micron's CHIPS Act award (Dec. 9, 2024) carries a two-year restriction on buybacks beyond offsetting stock-comp dilution, expiring December 9, 2026. Sell-side models for what comes after are aggressive: UBS models buybacks approaching 40% of shares outstanding by 2028 on $400B+ of cumulative projected free cash flow through that year; BofA models $31.7B in FY2027 buybacks alone, with net cash reaching $140B by the end of fiscal 2027. In the interim, Micron isn't sitting idle — debt paydown continues, and cash is stockpiling rather than being trapped.

The market is treating Micron as though demand is deteriorating. The evidence released over the past week suggests the opposite. HBM remains sold out, hyperscalers continue raising AI infrastructure spending, strategic customer agreements are extending revenue visibility beyond the data center, and the final capital-allocation constraint expires in just over four months. If those conditions persist, the next major change in Micron's story is unlikely to come from demand—it will come from what management does with one of the largest free cash flow streams in corporate America.

This is not investment advice. Position sized per CCF framework; see prior MU issue for full thesis construction.