Nvidia faces a market it helped create, as memory makers reap the rewards

Nvidia, long the undisputed leader in AI hardware, is experiencing an unusual downturn. The company’s stock has fallen roughly 15% from its May peak, even as projected revenue continues to climb. According to Bloomberg data, Nvidia is now trading at a lower price-to-earnings ratio than the S&P 500 average — meaning investors are paying less per dollar of projected profit than they do for a typical large American company.

The cause is not a decline in demand for AI infrastructure. Instead, money is flowing into a different part of the supply chain: memory manufacturers. Micron, one of the world’s largest producers of DRAM and high-bandwidth memory (HBM), has nearly tripled in value over the same period. The GPU shortage that dominated headlines last year has eased, while demand for memory chips has surged, creating a new bottleneck in data center buildouts.

Why memory is the new hot commodity

High-bandwidth memory is a specialized component designed to move data in and out of processors as quickly as possible. For two decades, memory companies have been making incremental improvements to these chips. But the technology itself has not undergone a revolutionary leap. What changed is the market.

Data centers require enormous amounts of memory to feed the growing number of AI accelerators. As demand has outstripped supply, memory prices have risen dramatically. According to Datatrack, the spot price for DRAM — the price paid on the open market, as opposed to long-term contracts — has increased roughly tenfold over the past year.

In contrast, the spot price for an hour of compute time on an Nvidia H100 GPU has been steadily declining. Data from the compute marketplace Ornn shows prices peaked around $3.20 per hour in May and have since fallen. The value of Nvidia’s business is closely tied to the price of compute, and that price is dropping.

More players, same memory constraints

Several factors are driving GPU prices down. Google, Amazon, Microsoft, and even OpenAI have all launched or announced custom processors designed to reduce reliance on Nvidia. While these chips may not match Nvidia’s latest models in raw performance, they are good enough to increase supply and push prices lower.

Memory is a different story. As Ornn co-founder and CTO Wayne Nelms explained, "More GPU and accelerator players are entering the market. Everyone wants to make their own silicon, but no one is making their own DRAM." Until there is a major technological breakthrough in high-bandwidth memory, or a new entrant disrupts the supply side, Nelms expects the current dynamics to persist.

What this means for investors and the industry

For Nvidia, the situation is a direct consequence of its own success. The company proved how valuable AI compute can be, and now a market has emerged around it — one where simpler, less flashy technologies are capturing the upside. Memory companies, which have not changed their products dramatically, are benefiting from a demand surge they did not create.

This shift has implications beyond stock prices. It signals that the AI infrastructure buildout is entering a new phase, where the bottlenecks are no longer just about processing power. Memory supply constraints could slow data center expansion and affect the pace of AI development. For companies building AI applications, the cost of compute may continue to fall, but the cost of memory may remain elevated.

Conclusion

Nvidia remains a technological powerhouse, and its GPUs are among the most complex devices ever produced. But the market it helped create is evolving. The company now finds itself competing in a space where its own innovations have lowered barriers for rivals, while memory makers — producing technology that has improved incrementally for years — capture the scarcity premium. For investors and industry observers, the lesson is clear: in the AI infrastructure market, the most valuable position is not always the most advanced one.

FAQs

Q1: Why has Nvidia’s stock fallen if its revenue is still growing?
Investors are concerned about falling GPU prices and increased competition from custom processors by Google, Amazon, and others. While revenue grows, profit margins may compress, and the stock is being revalued accordingly.

Q2: What is high-bandwidth memory and why is it suddenly so valuable?
High-bandwidth memory (HBM) is a type of memory chip designed to move data quickly between processors and storage. AI data centers need enormous amounts of HBM, and supply has not kept up with demand, driving prices up sharply.

Q3: Could Nvidia regain its pricing power?
Nvidia could regain pricing power if its next-generation GPUs offer a significant performance leap over competitors, or if demand for compute accelerates faster than supply can grow. However, the trend of falling GPU prices may continue as more custom processors enter the market.