source: techcrunch ai: nvidia is a victim of the compute marketplace it created
level: business
nvidia's stock has dropped 15% since its may peak, even as projected revenue grows. the company is now cheaper than the s&p average relative to expected earnings. investors are paying less per dollar of nvidia's projected profit than for typical large american companies. money is still flowing into ai infrastructure, but mostly into memory companies. micron, a major dram maker, has nearly tripled in value over the same period. memory has become the new bottleneck for data centers and the hot ai trade.
the gpu shortage that seemed alarming last year has eased. data centers need all the memory they can get. nvidia's technological achievements are impressive, from cuda to rapid gpu development. but memory companies like micron have a simpler story. they make high-bandwidth memory chips that have improved incrementally for 20 years. demand for these chips has grown faster than supply, allowing prices to increase tenfold over the past year. dram spot prices have soared since 2023, while the spot price for an hour on an nvidia h100 gpu has fallen steadily since may.
the disparity comes down to supply and demand. major cloud providers and even openai have launched custom processors, reducing reliance on nvidia. these chips may not match nvidia's latest, but they are good enough to drive down compute prices. in contrast, no one is making their own dram. until there is a breakthrough in high-bandwidth memory or new market entrants, the situation will likely persist. nvidia's success in proving compute's value has created a market everyone wants to enter, while simpler memory technologies reap the rewards.
why it matters: shows how ai infrastructure demand is shifting from compute to memory, affecting investment and hardware strategy.
source: techcrunch ai: nvidia is a victim of the compute marketplace it created