Nvidia’s compute dominance wanes as memory prices surge and custom silicon expands

Nvidia’s stock price has declined 15% since its May peak even as revenue projections continue to rise, making the company cheaper than the S&P 500 average on a valuation basis. Analysts note investors are paying less per projected dollar of profit than for the typical large U.S. company.

At the same time, memory chip suppliers have benefited from a surge in demand for high-bandwidth memory (HBM) used in data centers. Micron, one of the largest DRAM manufacturers, has seen its valuation nearly triple over the same period, as spot prices for DRAM have increased tenfold in the past year due to supply constraints and unmet demand.

The shift reflects a broader rebalancing in AI infrastructure economics. While Nvidia’s GPUs remain central to many AI workloads, the easing of GPU shortages has reduced the scarcity premium on compute time. Spot prices for an hour of Nvidia H100 GPU compute peaked around $3.20 in May and have since declined.

Cloud providers and AI labs are accelerating efforts to reduce dependence on Nvidia by developing custom processors. Companies including Google, Amazon, Microsoft, and OpenAI have launched or expanded their own silicon offerings, which, while not matching Nvidia’s latest performance, are sufficient to lower the market price of compute.

Industry observers attribute the divergence to supply and demand dynamics. Wayne Nelms, co-founder and CTO of Ornn, said more accelerator players are entering the market, increasing competition for compute, while no comparable new entrants are emerging in memory. He added that until a major technological breakthrough occurs in HBM or a new supplier enters the memory market, current conditions are likely to persist.