Nvidia Falls 15% as Memory Chips Become AI's New Bottleneck
Get the Finance newsletter
Daily finance — markets, central banks, M&A, the prints that move money. Free.
- Nvidia has seen its stock fall 15% from its May peak even as projected revenue continues to grow, leaving the company cheaper than the S&P average relative to expected earnings.
- Micron has nearly tripled in value over the same period as DRAM spot prices have risen roughly tenfold in the past year, establishing memory as the new bottleneck for data centers.
- Nvidia H100 GPU spot pricing peaked around $3.20 per hour in May and has steadily declined since, mirroring the stock's trajectory on the compute side.
- Google, Amazon, Microsoft, and OpenAI have all launched custom processors to reduce dependence on Nvidia, and even though those chips trail Nvidia's latest models, they're driving down compute prices.
- Ornn co-founder and CTO Wayne Nelms said no one is making their own DRAM, so the memory supply crunch will persist until a major HBM breakthrough, a supply/demand shift, or a new market entrant.
- The shift is largely a byproduct of Nvidia's own success in demonstrating how valuable compute can be — creating a market every hyperscaler now wants to enter.
Why it matters: For Nvidia, the 15% drop despite continued revenue growth marks a shift from scarcity-driven GPU pricing to a competitive compute market. For Micron, tenfold DRAM spot-price increases and a near-tripled stock reflect supply constraints no hyperscaler can engineer around. The AI infrastructure trade is splitting: chips commoditize, memory tightens.
