HBM Prices Keep Rising as Qualcomm, NVIDIA, and Cerebras Explore Memory-Reduced Architectures
N.R. Finch
Micron expects this quarter's operating profit to exceed the highest full-year revenue in its history; 15 new five-year supply deals stretch the tight-supply outlook past 2027 — yet those very prices are pushing AI customers to systematically explore ways around HBM.
How extreme is Micron's profitability right now?
Micron expects one quarter's operating profit to surpass the highest full-year revenue the company has ever posted. This means → it will earn more in three months than it used to sell in its best twelve.
The company has signed 15 new five-year supply agreements with price floors — even in a downturn, prices will stay well above past-cycle lows.
In plain terms = Micron has locked its worst-case scenario above its former best-case ceiling.
How far has the price surge traveled down the supply chain?
Apple this week raised prices on multiple Mac and iPad models between product-launch cycles, explicitly blaming memory-chip cost increases.
Analyst Rolf Bulk notes that even at the floor price of Micron's long-term contracts, profitability still exceeds prior-cycle peaks.
This reflects something broader: memory inflation is no longer an intra-chip-industry issue — it is reaching consumer end-products.
Who is trying to use less — or zero — HBM?
Qualcomm this week spotlighted its "High-Bandwidth Compute" architecture, designed to cut AI workloads' dependence on HBM through new system-level design.
Nvidia is reportedly adjusting parts of its next-generation Vera Rubin platform to reduce overall memory requirements.
Cerebras markets "no HBM" as a core selling point. CEO Andrew Feldman put it bluntly: "HBM is scarce and expensive — and we don't use it at all."
Can algorithms help close the gap?
Google published its TurboQuant research in March — a model-compression method that significantly cuts AI-model memory usage with near-zero performance loss.
After the announcement, Micron's stock dropped nearly a third before rebounding more than twofold. This means → capital markets price any technology that could weaken HBM demand as an existential threat to the memory sector.
High prices as both moat and catalyst — how does this contradiction resolve?
Supply is unlikely to improve materially for two to three years; customers have almost no bargaining power in the near term.
But a five-year contract window also gives the world's largest AI buyers enough time — and a powerful economic incentive — to invest in new chip architectures, software algorithms, and system designs.
Put simply = Micron's high prices are both the foundation of this super-cycle and potentially the strongest catalyst pushing customers to de-HBM. Which force wins first — around the 2027 supply-demand inflection — will define the memory industry's long-term demand structure.
Content is for reference only, not financial advice.