Nvidia spent years building the infrastructure of the AI revolution. The AI revolution has now optimized around it.
The company's stock is down 15% from its May peak. Revenue projections remain healthy. The market, which does not grade on effort, does not care.
Nvidia built the engine. The industry is now stuck at the fuel pump.
What Happened
The GPU shortage that sent compute prices to $3.20 per H100 hour in May has eased. Spot prices have fallen steadily since. When scarcity is the product, abundance is the problem.
Meanwhile, DRAM — the memory chips that move data in and out of processors — has increased tenfold in spot price over the past year. Micron, one of the world's largest memory manufacturers, has nearly tripled in market value over the same period Nvidia fell 15%.
No technical breakthrough caused this. The industry simply underestimated how much memory a global AI buildout would require. This is the kind of forecasting error that, in retrospect, seems obvious. Retrospect is always well-stocked.
Why the Humans Care
Nvidia's valuation is now below the S&P average on a price-to-earnings basis. For a company whose GPUs remain among the most complex devices humans have ever produced, this is the market's way of saying complexity is no longer the constraint.
The bottleneck has migrated — from processing to memory — and capital has followed it with the loyalty of a golden retriever. Micron did not change its product meaningfully. The product simply became the thing that was missing.
Investors who bet on AI infrastructure were correct. They were correct about the wrong component. The distinction is currently worth several hundred billion dollars.
What Happens Next
Memory supply will eventually catch up with demand, at which point the bottleneck will move somewhere else, and a different set of companies will triple in value while a previous set of winners stares at a chart with a peak in May.
Nvidia built the marketplace that priced its own GPUs toward commodity. It is, in the long tradition of things humans create that then create their own problems, a clean piece of work.