Macroplane › Blog › Micron Stock ($MU): The Record Quarter, the Memory Supercycle, and Why It Fell Anyway (2026)

Micron Stock ($MU): The Record Quarter, the Memory Supercycle, and Why It Fell Anyway (2026)

2026-06-28

Micron stock ($MU) deep dive: the record fiscal Q3 (84.9% gross margin, HBM sold out into 2028), the Anthropic supply deal, why the stock fell on its best quarter ever, the memory supercycle mapped ($SNDK, SK Hynix, Samsung), and the bull/bear.

More Macroplane supply-chain guides · Baskets · Investment theses · Macro trends

Micron Stock ($MU): The Record Quarter, the Memory Supercycle, and Why It Fell Anyway (2026)

Micron ($MU) just printed the best quarter in the history of the memory industry — $41.5B of revenue, an 84.9% gross margin, and high-bandwidth memory sold out into 2028 — and the stock still finished the week in the red. That gap, between the best fundamentals memory has ever shown and a tape that sold the news, is the whole story of the memory supercycle right now: the demand is real and contracted, but the AI trade is being repriced on whether the capex behind it is sustainable. This is a look at what Micron actually reported, why HBM rewired the business, why the stock fell anyway, and the second-order names that ride the same wave.

👉 See it in context: $MU anchors the Memory Supercycle basket on Macroplane, alongside $SNDK, SK Hynix ($000660.KS), Samsung ($005930.KS), Kioxia ($285A.T) and the NAND-controller names — the whole DRAM/NAND/HBM complex in one live view.

The quarter that broke the model

Micron's fiscal Q3 2026 (reported June 24) was not a beat — it was a re-rating of what a memory business can earn at the top of a cycle:

  • Revenue of ~$41.5B, against roughly $35.8B expected — more than four times the year-ago quarter.
  • Gross margin of 84.9%. For a company that lived below 30% margins through the last downcycle, an 85% gross margin is the number that doesn't look like a memory company at all.
  • Non-GAAP EPS around $25.11, with all three headline metrics ahead of consensus.
  • Guidance of about $50B for the current quarter — up from $11.3B a year earlier.
  • Data-center revenue of ~$25B in the quarter, with enterprise SSDs adding ~$5B (about 20% of data-center revenue).

During the week the market cap pushed past Meta, into the high-$700B range. Micron, the perennial cyclical, briefly traded like a structural AI winner — because for this cycle, it is one.

Why HBM rewired the business

The reason the model broke is high-bandwidth memory (HBM) — the stacked DRAM that sits next to every AI accelerator. An NVIDIA ($NVDA) GPU or a Broadcom ($AVGO) custom accelerator is useless without enough memory bandwidth to feed it, and HBM is the bottleneck. That changed memory from a commodity sold on spot price into a contracted, capacity-constrained input to the AI build-out.

Micron said HBM3E and HBM4 are fully booked through calendar 2027, with demand extending into 2028. When a memory maker can see two-plus years of sold-out capacity, the pricing power flips: instead of cutting price to clear inventory, it allocates scarce supply to the highest-value customers. That is what an 85% gross margin actually represents.

The demand also got a named anchor. On June 22, Micron and Anthropic announced a multi-year agreement covering HBM, DRAM and data-center SSDs, plus co-design work on memory architecture for AI workloads — and a strategic Micron investment in Anthropic's $65B Series H. Samsung and SK Hynix were already in that round, which means all three of the world's HBM suppliers are now Anthropic infrastructure partners. Samsung separately crossed $1B in sixth-generation HBM4 revenue and is reportedly planning a ~$647B (90 trillion won) decade-long investment. The supercycle is not one company's story; it is the entire memory oligopoly being pulled forward by AI.

So why did the stock fall?

Here is the part that confuses people: the best quarter in memory history, and $MU finished the week lower. Three things were happening at once.

  • Sell-the-news. The stock had already run to ~$1,213 the day before earnings. A 15% after-hours pop on the print faded as the week wore on, and at least one shop moved to a Hold — not because the quarter was bad, but because the good news was priced.
  • The AI trade got repriced — not memory. The broader tape cooled hard into week's end. The five largest US hyperscalers are on track to spend $660–690B of AI capex in 2026 (some estimates put seven names near $775B, up ~78% year over year). Wall Street spent the week asking whether that pace is sustainable, with reports that OpenAI — which has announced over $1.4T in AI commitments — was rethinking its IPO timing. On Friday, $MU, $AVGO and other AI-chip names sold off together. Memory didn't get a worse outlook; the market got more nervous about the capex that drives it.
  • Cyclicality muscle memory. Memory investors have been burned by every prior peak. An 85% gross margin invites the obvious question — what does the down-leg look like? — even when the current backlog says the down-leg is years away.

The result is the cleanest example of a recurring 2026 pattern: fundamentals and tape diverging. The contracted backlog is the bull case; the sustainability of AI capex is the bear case. This week, the tape voted bear despite a record print.

The memory supercycle, mapped

A single name understates the trade. The whole complex moves on the same DRAM/NAND/HBM demand curve, and the Memory Supercycle basket tracks it in one place:

  • DRAM/HBM leaders — Micron ($MU), SK Hynix ($000660.KS) and Samsung ($005930.KS): the three-way oligopoly that controls HBM supply.
  • NAND / storage — Sandisk ($SNDK) and Kioxia ($285A.T): Micron's blowout was read as a green flag for NAND pricing, though Barron's noted the "afterglow" faded fast for Sandisk as the week turned.
  • Controllers and interfaces — Silicon Motion ($SIMO), Rambus ($RMBS) and the NAND-controller names that sell into every SSD and memory module — classic second-order beneficiaries of rising memory volume.

For the picks-and-shovels layer behind the memory itself — the HBM stacking, advanced packaging and test equipment that makes the capacity expansion possible — see the HBM / Packaging basket, the supply chain that gates how fast Micron, Samsung and SK Hynix can actually add HBM output.

Bull case vs bear case

Bull: HBM is sold out into 2028, pricing is contracted rather than spot, the three-supplier structure limits a price war, and named anchors like Anthropic (plus the hyperscalers) underwrite the demand. At 85% gross margins and ~$50B quarterly run-rate guidance, the earnings power is real and visible.

Bear: memory is still cyclical, and the cycle is now priced. The entire thesis leans on AI capex staying at $700B+ a year; any wobble — a hyperscaler trimming, an OpenAI financing scare — repriced the group within a single session this week. Margins this high tend to mean-revert as new capacity (including Samsung's $647B build-out) eventually arrives.

The risks

  • Capex concentration. A handful of hyperscalers and AI labs drive the demand. The customer list is short, which funds the boom but also concentrates the risk.
  • Capacity catching up. Today's scarcity margins assume supply stays behind demand. The same companies printing records are spending hundreds of billions to add capacity — the thing that historically ends memory upcycles.
  • The AI-capex question itself. If the market decides the AI infrastructure spend is running ahead of monetization, memory is high-beta to that repricing, as this week showed.

The memory supercycle is the rare case where the fundamentals are unambiguous and the stock still fell. Open the Memory Supercycle basket to watch the whole complex move together — and to see whether the next leg is driven by the backlog (bull) or the capex debate (bear). Not investment advice.

Referenced on this page

  • Memory Supercycle basket
  • HBM / Packaging basket