Jassy Didn't Endorse Micron. He Named the Bottleneck That Makes It a Winner — and a Risk.

On Amazon's July 30 earnings call, Andy Jassy raised the company's 2026 capital expenditure forecast from $200 billion to $220 billion and pointed directly to the cause: higher memory costs. He added that even at that level, capacity may still be insufficient for 2026 and 2027 demand. Demand for 2028, he said, is "striking."

The market immediately rewrote this as a Micron buy signal. The competitor headlines — including one claiming Jassy gave investors "$220 billion reasons to buy Micron" — were out within hours.

I think the market is misreading what Jassy actually said, and the difference changes the allocation judgment.

Jassy didn't endorse Micron. He named the bottleneck in the AI infrastructure build-out — and that bottleneck is the reason Micron is posting numbers we haven't seen from memory chips since the dot-com era. But a bottleneck you're selling into at a $995 billion market cap is not the same as a bottleneck you're buying into for the first time.

The AI architecture bottleneck has moved to memory

High-bandwidth memory — the ultra-fast stacked DRAM that sits next to AI accelerators and feeds them data at speeds traditional memory can't match — has become the single most constraining input in the AI compute stack. HBM now accounts for roughly two-thirds of the total AI chip component cost, up from about 50% in early 2024. The price per stack has climbed from roughly $200 for HBM3 to an estimated $550 for HBM4 — a nearly 175% increase as stack heights and bandwidth requirements grow.

That's the architectural shift. The market spent 2023 and 2024 fixated on GPU supply. Nvidia's H100 and H200 were the choke points. But the GPU supply chain is maturing. TSMC is expanding 4-nanometer and 3-nanometer capacity. What's tightening now is the memory that has to move at the same speed.

Apple CEO Tim Cook confirmed this independently, warning of "exponential increases in memory prices" that he called a "100-year flood." Two of the three largest technology CEOs, in the same week, pointing at the same constraint. That's not a product cycle rumor — that's the supply chain telling you where the scarcity is.

GlobalData's analysts quantified the mechanism: roughly 55–70% of 2026 revenue growth at Samsung memory, SK Hynix, and Micron comes from price, compared with 15–25% at TSMC and the Japanese equipment makers. Memory isn't just in demand — it's in a pricing power environment we haven't seen since the last AI supercycle turned into a memory boom-bust in 2022.

Micron's financials are the numbers the bottleneck produces

Micron's fiscal 2026 second quarter — reported June 24 — delivered $23.86 billion in revenue against a consensus of $19.97 billion. EPS came in at $12.20 versus an expected $9.19. Revenue growth year-over-year was 167%. Sequential growth was 73.75%, up from $13.64 billion in the prior quarter.

The margins are what separate this cycle from a standard commodity upswing. Gross margin hit 72.57%. Operating margin reached 65.63%. Return on invested capital was 58.63%. Free cash flow for the trailing twelve months came in at $26.17 billion, up 1,291% year-over-year, on a free cash flow margin of 28.99%. The balance sheet carries $25 billion in cash against $33.4 billion in total debt — a net cash position of roughly $20.3 billion. Debt-to-equity is 5.68%.

Put plainly, this is not a company cyclical-rebounding. This is a company sitting on the narrowest supply bottleneck in the AI infrastructure stack and charging for it.

The strategic agreements are the real story

Here's the part of the story the Jassy headlines didn't cover, and the part I care about more. On June 24, the same day Micron reported earnings, management disclosed 16 strategic customer agreements carrying $22 billion in cash deposits and related financial commitments. These multi-year contracts include take-or-pay provisions (the buyer pays even if they don't take delivery), binding volume commitments, and pricing floors. The total remaining performance obligations across these deals represent approximately $100 billion.

For 47 years, Micron was a commodity business. Memory prices cycled. Manufacturers absorbed the downside. Customers walked away when prices peaked. These agreements fundamentally restructure that relationship. They're projected to cover roughly half of Micron's sales once fully ramped, shifting risk from the producer to the buyer.

What this means for the investor is that the boom-bust memory cycle — the pattern that has erased memory stock returns as predictably as a clock for three decades — is being structurally dampened. Not eliminated. About 60% of revenue will remain exposed to market pricing even after all planned agreements are executed. But half the business having a floor under it is a change in the risk profile that hasn't existed in the memory industry's history.

But Micron is not the HBM leader

This is where the $220 billion headline starts to break down as a Micron bull case.

SK Hynix leads the HBM market with 50–55% share by revenue. Samsung holds 35–40%. Micron — despite its extraordinary financials — holds 5–10%. Micron entered the HBM3E market later than its Korean rivals. It was SK Hynix that first mass-produced HBM3E and has been Nvidia's primary memory supplier from the Blackwell generation backward. Samsung, despite falling to third place in recent quarters, is aggressively ramping and has won major allocations with AMD and Google's TPU supply chain.

All three vendors have now qualified for Nvidia's Vera Rubin platform using HBM4. Volume shipping has begun. Micron has demonstrated 16-die stacking capability and shipped HBM4 samples to customers. But the architecture qualification timeline — not the financial headline — is what determines who captures the next generation of demand. SK Hynix still holds the relationship advantage with Nvidia. Samsung has the scale advantage and is resuming construction of its Pyeongtaek P5 fab.

Micron is on the right side of the transition. But it's not leading it.

The price action already reflected most of the good news

Micron's stock is up 605% over the rolling year and 208% year-to-date. It's currently trading around $881, carrying a $995 billion market cap. The trailing P/E is 19.7 — which looks cheap until you look at the forward P/E of roughly 140x. That divergence exists because consensus expects earnings to compress as the cycle eventually turns.

On July 30, the day of Amazon's earnings, Micron rallied 18.3% on Samsung's earlier announcement that memory shortages would persist through 2028. The next day, it dipped 5.9%. The Jassy headline didn't move the stock because the information was already digested. The market had already front-run the supply shortage narrative.

What the trailing P/E of 19.7 doesn't tell you is how much future margin compression the forward multiple is pricing in. The SCA structure caps upside pricing to April–June market levels while locking in a floor. That protects the downside but also limits the upside — which is exactly the signal that management expects a turn, even if that turn is years away.

Where this sits in the allocation framework

I believe Micron's position in the AI memory bottleneck is real and durable through at least 2027. The strategic customer agreements represent a genuine structural change in how memory risk is allocated between producer and buyer. The financials are exceptional.

But the question isn't whether Micron is a good company. The question is whether a $995 billion market cap, up 605% in a year, with a forward P/E of 140x, is better deployed than the alternatives in the AI trade.

Micron holds 5–10% of the HBM market. SK Hynix holds 50–55%. Samsung holds 35–40%. Both Korean rivals are public in their home markets and trade at valuations that reflect their scale advantage. Nvidia, which is building the demand that's pricing Micron at nearly a trillion dollars, trades at a forward P/E fraction of Micron's. AMD, at roughly 124x trailing earnings and a $799 billion market cap, is the dark horse competitor trying to take GPU share from Nvidia — and also a major HBM customer.

The debate is not whether Micron benefits from the AI memory shortage. It is whether the return profile at this price — after a six-fold gain in 12 months, with the bottleneck already priced into the stock and the cycle turn already partially hedged by management through SCAs — is as compelling as what can be found elsewhere.

I think Micron still has room to run if demand stays tight through 2028 as Jassy suggested, and if HBM4 qualification goes smoothly. But much of the asymmetric return is likely back-half weighted. A position here makes sense if you're holding for the 2027–2028 inflection when SCAs start delivering predictable recurring revenue. It's less compelling as an entry point for someone trying to catch the AI infrastructure wave — because the wave is already two-thirds up the beach.

What would change my view: HBM4 qualification setbacks for Micron while SK Hynix or Samsung pull ahead architecturally. A material renegotiation of SCAs if memory pricing reverses earlier than 2028. Or, conversely, if Micron gains meaningful HBM share — moving from 5–10% toward 15%+ — the current market cap would no longer reflect a laggard pricing multiple.

Where capital goes: I'd prefer to wait for a pullback toward $700–$750 before establishing a new position, or allocate to the companies building the demand that's making Micron's numbers this good in the first place.