Micron Technology’s stock has climbed almost 9x in the past 12 months, pushing the company’s market value above $800 billion as investors pile into a memory maker suddenly tied to the AI boom. The mu stock price rally is being driven by surging demand for high-bandwidth memory, a product so tight that Micron says it has already sold out its entire HBM output through 2026 under binding contracts.
The move has been powered by the scale of AI spending now expected from hyperscalers. Microsoft, Alphabet and Meta are projected to spend roughly $700 billion on AI infrastructure this year, and that wave of capital is flowing directly into the hardware that keeps advanced systems running. Micron is trading at 14x estimated FY '26 earnings and 8x next year’s earnings, a valuation that looks modest only if today’s shortages last as long as bulls expect.
That expectation rests on a market with a long memory of its own. The memory industry has spent the past fifteen years moving from boom to bust and back again, with new DRAM capacity taking two to three years to build and costing tens of billions of dollars. That delay has repeatedly let supply run ahead of demand, then crash back into it. The 2022 to 2023 downturn was one of the most severe in financial terms. Supplier inventories reached 31 weeks by early 2023, Micron reported a GAAP net loss of $2.31 billion in a single quarter, cut its workforce by 10%, and sharply reduced its capital budget. Its stock fell roughly 50% from early 2022 levels after trading at about 11x forward earnings at the cycle’s peak.
It was hardly the first time. During 2018 and 2019, cloud operators over-ordered memory through 2017 and then steadily reduced purchases as inventories swelled, sending NAND prices down roughly 60% and DRAM prices about 40%. Micron peaked near $64 in May 2018 and fell to around $28 by year-end, a decline of about 57%, after trading at just 4.5x forward earnings. In the 2014 to 2016 crash, DRAM capacity had expanded ahead of PC demand that never arrived as consumers shifted to mobile devices, and Micron’s stock dropped about 70% from around $37 in late summer 2014 to under $10 by February 2016.
The difference now is that HBM is no longer being treated like a commodity with a spot price. It was initially used mainly for training large language models, but in 2026 the focus has shifted toward inference, where ultra-low latency matters more. Applications such as real-time video generation and advanced AI agents require memory performance that only HBM can deliver, and the product is increasingly sold through long-term agreements with hyperscalers rather than the spot market. Nvidia’s H100 used 80GB of HBM, while its Rubin Ultra successor targets 512GB per GPU module, underscoring how quickly demand is rising. For Micron, the question is not whether AI memory is hot today. It is whether a business famous for violent cycles can stay this tight long enough to justify a stock that has already run almost ninefold.
