BOISE — Sanjay Mehrotra said something unusual at Micron Technology’s last earnings call: his company cannot build memory chips fast enough to meet demand. Not close. He estimated Micron can supply roughly 60% of current HBM orders. The other 40% goes unmet.
That sounds like a problem. In the AI chip industry right now, it’s the opposite. The companies that cannot keep up with demand are the companies that matter.
Micron shares traded around $922 Friday, down from Thursday’s $935.39 close, as the broader Nasdaq fell 0.5% following Federal Reserve Chair Kevin Warsh’s hawkish Jackson Hole address. The stock has fallen more than $300 from its 52-week high of $1,257, reached in early 2026 when HBM supply constraints first became visible to the market. The 52-week low is $114.25. The distance between those two numbers describes what two years of AI infrastructure spending has done to memory chip valuations.
The 60% supply figure matters because it explains why Micron’s HBM pricing has held even as the broader memory market cycles. DRAM and NAND — Micron’s commodity businesses — move with the classic semiconductor cycle: oversupply drives prices down, constrained supply drives them up. HBM doesn’t work that way. The customers buying HBM are building AI data centers on multi-year timelines. Micron’s entire 2026 HBM supply is sold out. The question is not whether Micron will sell the memory it makes. It’s whether Micron can make enough of it.
The memory wall is the structural reason this isn’t a temporary shortage. AI compute performance has been doubling roughly every eighteen months, a pace that has only accelerated since the original generation of large language model deployments. High Bandwidth Memory — the stacked DRAM that feeds AI accelerators with the data they need — has been improving more slowly. Bandwidth roughly doubles every two years. The gap between what AI chips can compute and what memory systems can deliver to them has been widening, not narrowing. Mehrotra called it the defining constraint on AI scaling at Micron’s last public presentation, and the math behind his concern is straightforward enough that it should probably worry the engineers designing the next AI chip generation even more than it worries Micron’s investors.
HBM4, which Micron began shipping in commercial quantities in the second quarter of 2026, doubles the bandwidth of HBM3E. That sounds like a matching move. The problem is that the AI accelerators HBM4 will feed have also made a generational jump in raw compute. The memory wall doesn’t close at the step change; it narrows temporarily and then the compute lead reasserts itself. This dynamic is why Micron’s position in the AI supply chain is structurally stronger than a drawdown from $1,257 to $922 suggests.
Micron is one of only three companies in the world — alongside Samsung and SK Hynix — capable of producing HBM at commercial volume. Its market share in HBM has been growing. The September 30 earnings report will be the next data point investors use to calibrate how much of the AI-driven demand Micron can actually convert into revenue. The consensus analyst target for MU sits at $1,515. The current price is $922. That $593 gap reflects what the market thinks Micron’s HBM trajectory is worth in a few years if demand holds and supply scales. Whether that projection survives contact with the earnings call depends on what Mehrotra says about the 60% figure — specifically, whether it’s moving toward 70% or retreating toward 50%.
What Warsh said Friday complicated the near-term setup. A higher-for-longer Federal Reserve rate environment affects capital spending decisions across the technology sector. Data center build-outs — the primary driver of Micron’s HBM demand — are not rate-sensitive in the immediate term because the commitments have already been made. Amazon, Microsoft, and Alphabet have disclosed AI infrastructure spending that runs through 2027 and 2028. But a sustained high-rate environment affects the next wave of commitments, and those are the ones that will determine Micron’s revenue profile in 2028 and beyond.
The $1 trillion memory chip market projection that Micron’s management cited when the company briefly crossed $1 trillion in market capitalization in May 2026 — the first time a memory chipmaker had done that — was premised on that next wave materializing. Micron’s market cap has since pulled back. The premise hasn’t changed. The timeline has become less certain, which is a different kind of risk than demand disappearing.
What hasn’t changed is the immediate constraint: Mehrotra cannot fill his order book. The 40% of demand Micron cannot supply is going somewhere — to Samsung, to SK Hynix, or in some cases it’s simply going unfulfilled because no one has enough HBM to satisfy the AI buildout at its current pace. Industry reporting from earlier this year noted that AI accelerator deployments at several large data centers were delayed in the first half of 2026 specifically because HBM supply could not match GPU shipment schedules. The memory is the bottleneck. That is a terrible problem for the data center operators and a remarkable competitive position for the memory makers.
For Micron investors who bought above $1,000 and are sitting on a meaningful drawdown, Friday’s session — down roughly 1.4% as of midday, alongside the chip sector broadly — is a reminder that the distance between a supply constraint that is a competitive moat and one that is simply a capacity failure depends almost entirely on how quickly Micron can build. The September 30 earnings call will be the next place to measure that. Whether the market gets what it needs before then — a production milestone, an HBM4 contract announcement, an updated supply estimate from Mehrotra that closes the 60% gap — is the uncertainty that lives in the roughly $600 spread between the current price and where analysts think Micron should be.
Chip sector context: Intel’s foundry turnaround story, covered separately in Friday’s INTC analysis, showed the same hawkish Fed pressure dragging down a different kind of semiconductor ambition. At Micron, the issue isn’t ambition. It’s throughput.
Related Posts
