The memory supercycle is now signed in contracts: inside Micron's $100B take-or-pay floor

The most famous short seller of the 2008 era just bet against the strongest income statement in the S&P 500. Michael Burry disclosed added short positions in Micron $MU and NVIDIA $NVDA this week, and X has spent days litigating it with price charts and deal headlines.

Almost nobody in that debate is quoting the two documents that actually decide it: Micron's fiscal Q3 2026 press release and earnings call, filed June 24, 2026. We pulled both through Aether, our SEC search engine. Here is what is actually signed, sworn and filed — on both sides of the argument.

If you hold $MU through this run, or you are waiting on the sidelines for the crash everyone keeps promising, this is the paper trail to size your view against. We never say buy or sell. We read filings.

The quarter the bears have to explain

Micron's fiscal Q3 2026 (quarter ended May 28, 2026, reported June 24):

• Revenue $41.46 billion — versus $23.86 billion the prior quarter and $9.30 billion in the same quarter last year. That is a 4.5x in twelve months.
• GAAP net income $28.24 billion, or $24.67 per diluted share.
• Operating cash flow $25.39 billion, versus $4.61 billion a year ago. Adjusted free cash flow $18.3 billion. Cash and investments $30.2 billion.
• Guidance for fiscal Q4: revenue $50.0 billion ± $1.0 billion at a gross margin of approximately 86%.

Sanjay Mehrotra's words in the release: "Micron's record fiscal Q3 financial results and even stronger outlook for Q4 reflect the strategic value of memory in the AI era."

For scale, three months earlier Micron guided fiscal Q3 to $33.5 billion ± $750 million. It printed $41.46 billion — roughly $8 billion above its own guide. The 10-K arc underneath: DRAM revenue was $10.98 billion in FY2023, $17.60 billion in FY2024, $28.58 billion in FY2025. The current single-quarter run rate now exceeds entire recent fiscal years.

Every business unit is printing margins that memory has never seen:

Core Data Center grew 7.5x year over year at an 87% gross margin. Commodity businesses do not print 87% gross margins. Which is the whole point: the filings say memory has stopped trading like a commodity.

The part that changes the cycle argument: take-or-pay

The classic memory-bear thesis — the one Burry's 2008 framing leans on — is that euphoric pricing always collapses when supply catches up, because nothing binds customers to keep buying. That thesis had a factual basis in every prior cycle. Micron's June 24 call describes the mechanism that breaks it.

Micron announced it has completed 16 Strategic Customer Agreements (SCAs) — a structure it calls a pioneer move for the industry:

• Structured as take-or-pay: binding commitments to purchase specific volumes over multi-year terms.
• Terms generally run five years, calendar 2026 through the end of calendar 2030 (automotive agreements about three years).
• The 16 signed agreements cover roughly 20% of Micron's DRAM volume and a third of its NAND volume over the period.
• 14 of the 16 carry cumulative revenue at minimum price of approximately $100 billion over the remaining term.
• Micron projects $22 billion of cash deposits and related financial commitments from customers under the agreements signed so far.
• The largest agreements carry a price ceiling at current market prices and a floor through the term. When all planned SCAs are executed, agreements with fixed prices or ceilings near current levels are expected to be about 40% of revenue — and half or more of total company revenue is expected to sit under SCAs.

Read that as a memory bear and the problem is obvious: customers have signed binding minimum-volume commitments, with deposits, at floors, for five years. The demand side of the next downcycle is contractually pre-sold. The floor is filed. As Micron put it, the SCAs provide "a very robust gross margin ... well above our peak quarterly margins in any past cycle."

Supply: "no line of sight"

The other half of the equation, in Micron's own words from the call: "we currently do not have line of sight as to when memory supply will be able to catch up with increasing demand." Tightness in both DRAM and NAND is expected to persist beyond calendar 2027.

The constraints listed are physical, not financial: greenfield fabs take years to build, skilled-trade workers are short, permitting is slow, energy infrastructure has to be built, and each new HBM generation consumes more wafer capacity per bit (the "trade ratio"), squeezing non-HBM supply. Meanwhile HBM4 12-high is ramping about twice as fast as HBM3E 12-high did, and Micron has already shipped over $1 billion of HBM4 revenue.

Demand context from the same call: industry data-center DRAM and NAND bit shipments in calendar 2026 are expected to more than double from two years ago; server units are now expected to grow high-teens percent this year; an L2+ vehicle carries over five times the memory of an average car, and a humanoid robot carries roughly ten times the memory of an L2+ vehicle.

This is not one company talking its book. Intel's $INTC Q2 2026 earnings release (July 23) flags "current industry-wide substrate and memory shortages" as a live supply-chain risk. Penguin Solutions' $PENG CEO said in its July 7 release that "memory is increasingly becoming one of the primary performance and scalability bottlenecks" of AI workloads. Applied Materials $AMAT disclosed side-by-side development programs with both SK hynix and Micron for next-generation DRAM and HBM. Four different filers, one filed bottleneck.

What is announced but not yet filed

Discipline requires the line: the NVIDIA and SK Group initiative reported at more than $500 billion (AI factories plus next-generation memory, including a 2-gigawatt data center), and the reported $200 billion Samsung–Broadcom agreement, are announcements, not SEC filings. They are consistent with everything above, but we treat them as context until they show up in documents. Burry's short, likewise, is a disclosure of position, not an argument — the argument has to beat the contracts.

The scoreboard to watch

The bull case and the bear case now have the same referee: filings. Watch three things from here. First, whether the remaining planned SCAs get signed and whether deposits actually land ($22 billion projected so far). Second, the fiscal Q4 print against the $50 billion guide and the 86% margin. Third, any softening in the "no line of sight" supply language — that phrase changing is the earliest filed signal the cycle is turning.

Headlines move prices. Filings size the exposure. If you take one small step from this piece, read the Strategic Customer Agreement section of Micron's June 24 press release yourself — it is two paragraphs — or run the search on Aether, our SEC search engine, and check any number in this article against the source documents.

Sources: MU fiscal Q3 2026 press release and earnings call, June 24, 2026 (acc 0000723125-26-000013, sec.gov); MU fiscal Q2 2026 press release, March 18, 2026 (acc 0000723125-26-000004); MU 10-K filed October 3, 2025 (acc 0000723125-25-000028); INTC Q2 2026 earnings release, July 23, 2026; PENG Q3 FY2026 release, July 7, 2026; AMAT Q2 FY2026 release, May 14, 2026. All retrieved and cited via Aether.

Research, not investment advice.