Micron Technology (MU) is no longer just the challenger in high-bandwidth memory. With more than $1 billion in HBM4 revenue already shipped and a contract book approaching $100 billion, the Boise-based company is mounting the most credible threat to SK Hynix’s dominance in the memory segment that powers AI accelerators.
The stakes are rising just as the supply chain shows signs of strain. Nvidia (NVDA) is reportedly testing lower-memory configurations of its next-generation Rubin Ultra accelerator, evaluating versions with as little as 192GB of HBM compared with a planned 1TB. The move, detailed in a BofA Global Research note, stems from HBM4e supply constraints and qualification delays rather than any shift in what AI workloads ultimately require.
The despec risk is real for memory suppliers. Every accelerator shipped with less HBM translates into fewer memory bits sold. Yet BofA’s analysis indicates GPU performance deteriorates sharply below 500GB, and the industry’s next-generation HBM4e and HBM5 stacks are already being designed around 12-high and 16-high configurations supporting roughly 500GB to 1TB per accelerator. That suggests the current testing represents a temporary engineering compromise, not a permanent demand reset.
Micron’s fiscal third quarter underscored how dramatically the competitive landscape has shifted. Revenue reached $41.456 billion, up 345.7% year over year, with GAAP gross margin expanding to 84.6%. Cloud Memory alone contributed $13.769 billion. CEO Sanjay Mehrotra told analysts that the HBM4 12-high volume ramp is tracking twice as fast as the prior HBM3E generation, with over $1 billion in HBM4 revenue already booked.
The company’s 16 take-or-pay agreements represent the structural difference in this cycle. Mehrotra said that even at floor prices, margins are expected to come in significantly above prior peak levels. Those contracts will eventually cover roughly half of company revenue, creating an earnings floor that earlier memory cycles never had.
SK Hynix, meanwhile, posted its own record quarter with revenue of 79.3 trillion won and operating profit of 60.5 trillion won. The Korean company made its NASDAQ debut in July 2026 through a $26.5 billion foreign offering, giving U.S. investors direct access to the HBM incumbent. Analysts maintained unanimous Buy ratings with 12-month targets averaging $245, though a modest earnings miss triggered an initial 10% selloff in the ADR.
Micron carries a heavier analyst consensus: 40 Buys, 5 Holds, and an average target of $1,501.98. The company guided for fiscal fourth-quarter revenue of approximately $50 billion with gross margin near 86%, and Mehrotra indicated tight supply conditions should persist beyond calendar 2027.
The valuation gap is striking. Micron trades at a forward P/E of roughly 6 against a contract book that locks in cumulative floor-price revenue near $100 billion. The stock sits near $972, valuing the company around $1.1 trillion, or about 22 times trailing earnings.
History offers a cautionary note. Micron earned $8.7 billion in fiscal 2022, then lost $5.8 billion in fiscal 2023 as revenue nearly halved to $15.5 billion and gross margin collapsed from 45% to negative 9%. The company has recorded annual losses three times in the past 15 years. The current boom dwarfs prior cycles, with trailing 12-month net income of $50.5 billion, but memory remains a commodity business where high prices have historically financed the new supply that eventually ends the boom.
The strategic customer agreements are untested in a downturn. They may shape how a future correction arrives without preventing one. For now, the key variables are whether SK Hynix keeps HBM4 qualification on pace at Nvidia, whether Micron’s HBM4E on 1-gamma DRAM stays on track for volume production in calendar 2027, and whether AI hyperscaler capex guidance holds. A rollover there would reset the setup for both stocks, regardless of contract structures.
Nvidia’s Rubin Ultra testing adds near-term uncertainty. If constrained HBM4e availability forces lower-memory configurations during the initial ramp, both Micron and SK Hynix could see temporary volume headwinds. But the broader trajectory points in one direction: AI workloads are becoming more memory-intensive, not less, and the industry is building toward 500GB-to-1TB accelerators as the standard.