Daily Pulse · · 09:00 CET · framework · MU
Memory dipped into Micron's print — the stock had sold off roughly 13% off its 22 June high going in — and then ripped about 15% on the beat. Trade the move if you like. The move is the least interesting thing that happened. The interesting thing is that Micron's customers just agreed to underwrite its future supply.
The 16 take-or-pay deals are the structural game-changer
Alongside the quarter, Micron announced 16 Strategic Customer Agreements (SCAs) that management said it expects to “fundamentally transform” the business model. They span data-center, consumer and auto customers, typically run five years (2026–2030), and together cover roughly 20% of Micron's DRAM volume and one-third of its NAND volume over the period.
The two words that matter: take-or-pay. Customers commit to buy specific volumes over a multi-year term. That is the opposite of the old memory model, where buyers pulled back the moment demand softened and suppliers were left with inventory, falling prices and collapsing margins. Old Micron sold into a volatile spot/contract market. New Micron reserves scarce memory capacity for strategic customers under multi-year commitments.
| What Micron disclosed | Figure | Why it matters |
|---|---|---|
| Strategic Customer Agreements | 16 | Broad validation, not one-off demand |
| Typical term | 2026–2030 | Multi-year visibility |
| DRAM / NAND volume covered | ~20% / ~1⁄3 | Core business, not just HBM |
| Minimum contracted revenue | ~$100B (14 of 16) | A revenue floor, conservatively struck |
| Deposits & commitments | ~$22B (~$18B cash) | Customers funding capacity access |
| Margin floor | > any past peak quarter | Floor prices lock a robust gross margin |
Why this floors the bear case
The old memory bear case was always one sentence: make huge profits at the top, then oversupply arrives, pricing collapses, and earnings disappear. The SCAs attack it four ways at once. They lock in demand (customers are contractually reserving, not saying “we may need memory”). They improve pricing visibility — the largest agreements carry ceiling prices near current Q2 levels and floor prices through the term, and Micron says the floor enables a gross margin well above peak quarterly margins in any past cycle. They fund capacity — about $22B of deposits and commitments lets Micron invest in fabs, tools and packaging with more confidence. And they change customer behaviour: AI labs, clouds, auto OEMs and device makers are no longer treating memory as interchangeable commodity, but as strategic supply.
The AI angle: memory becomes a platform constraint
Micron's management made the thesis explicit: AI systems may use GPUs, ASICs and CPUs from many suppliers, but they all share one thing — system performance is architecturally dependent on memory subsystem performance and capacity. That, they said, has elevated memory into a strategic asset. In the old cycle memory was a component; in the AI cycle it becomes a platform constraint, and customers sign because their own AI roadmaps depend on access to HBM, high-capacity DRAM and the broader memory/storage hierarchy.
The Anthropic agreement makes it concrete: a strategic partnership spanning memory-and-storage architecture design and supply, Claude adoption inside Micron, and a strategic investment — with Anthropic's compute leadership saying memory and storage are central to how efficiently it can train and serve Claude. That is the market repricing a memory maker from supplier to AI-infrastructure partner.
The quarter was spectacular; the deals matter more
The print itself was strong — revenue of $41.46B, gross margin around 84.6%, and Q4 revenue guided to $50B ± $1B at roughly 86% margin. But the quarter shows the cycle is strong; the SCAs suggest the cycle itself is changing. When customers commit ~$100B of minimum revenue and put $22B of deposits behind supply access, they are telling you they cannot build their AI, cloud, automotive and agentic roadmaps without guaranteed memory. That is a different business than the one the bear case was written against.
The Closelook read — how we play the dip. Our Generation Rotation Framework has us in Rubin Early Ramp — the phase where Memory, Packaging and Substrates lead — and Memory is the standout: the single strongest sub-index of the Rubin Build-Out 100, +383.65% year to date, ahead of all seventeen other layers, with the strongest trend state on our rotation board. The SCAs are why that leadership has a longer runway than a single ramp: a take-or-pay floor under the old oversupply bear case changes the shape of the downside. So we do not read memory's wobble into the print as the cycle topping — we read it as the market handing the structural owners of the memory wall (Micron, SK hynix, Samsung and the HBM supply chain) a discount on a re-rating that, in our view, is only part-finished. In the reference portfolio that reads as an accumulation window rather than a warning — a sensitivity, not a certainty. Common vehicles for the exposure run from the makers themselves to the broad memory and semiconductor ETFs.
Takeaway: Earnings told you the cycle is strong. The 16 take-or-pay deals tell you the cycle is changing. Memory is no longer bought like a commodity — it is being reserved like strategic AI capacity.