Nvidia Is a Victim of the Compute Marketplace It Created

For years, Nvidia stood unchallenged as the undisputed leader of the AI revolution, its graphics processing units (GPUs) and software ecosystem forming the backbone of nearly every major advancement in machine learning. But over the past two months, the company has found itself in an unexpected and uncomfortable position: it is now facing pressure from the very market it helped build, while other players in the supply chain are reaping greater rewards.

A Shift in Investor Sentiment

The numbers tell a clear story. Since hitting its peak in May 2026, Nvidia’s stock price has dropped roughly 15%, even as the company continues to report rising revenue. Measured against projected earnings, Nvidia is now valued more cheaply than the average company in the S&P 500, meaning investors are paying less for every dollar of its expected future profit than they are for most other large U.S. corporations.

Meanwhile, money flowing into AI infrastructure stocks has shifted direction. Instead of going primarily to chip designers, capital is flooding into memory manufacturers. Over the same period, Micron Technology, one of the world’s largest producers of DRAM and high‑bandwidth memory (HBM), has nearly tripled in market value. The dynamic has flipped: the GPU shortage that dominated headlines and drove prices upward throughout 2025 has eased significantly, while memory has emerged as the new critical bottleneck for data centers and the hottest investment in the AI sector.

Two Very Different Business Models

The contrast between Nvidia and memory makers like Micron highlights a fundamental shift in how value is distributed across the AI hardware stack.

NVIDIA’s rise was built on decades of innovation. It developed CUDA, a powerful programming platform that made its GPUs accessible and efficient for complex mathematical calculations, effectively turning them into the standard engine for AI research and training. Its hardware represents some of the most advanced engineering in the world, combining billions of transistors, custom architectures, and precise manufacturing to deliver massive computing power.

For memory companies, the picture is far simpler. They produce DRAM and HBM chips components designed to move data rapidly between storage and processors. These technologies have improved incrementally over the past 20 years, with no single revolutionary breakthrough in recent times. Yet as data centers scale up to support larger AI models, demand for memory has exploded far faster than production capacity can keep pace. The result: spot prices for DRAM have risen roughly tenfold since August 2025, according to data from Datatrack.

“The industry simply underestimated how much memory would be required to run these massive models and systems,” explained one analyst. “There was no sudden invention, just a realization that every new GPU needs more memory than the previous generation, and the supply chain cannot expand fast enough.”

Falling Compute Prices, Rising Memory Costs

The divergence is even clearer when looking at real‑time market rates. Data from the Compute Marketplace Ornn shows that the spot price for one hour of access to an Nvidia H100 GPU peaked in May at around $3.20, before entering a steady decline. As more capacity comes online and competition grows, the cost of raw computing power continues to drop.

In contrast, DRAM and HBM prices have kept climbing. While Nvidia’s fortunes are directly tied to the price of compute, which is now under pressure, Micron and its peers benefit from a commodity whose value keeps increasing.

Wayne Nelms, co‑founder and CTO of Ornn, attributes this gap to structural differences in the market. “Large cloud providers like Google, Amazon, and Microsoft, as well as developers like OpenAI, are all designing their own custom processors to reduce reliance on Nvidia,” he said. “Even if those chips aren’t as powerful as Nvidia’s latest models, they are ‘good enough’ for many workloads, and that added competition naturally pushes prices down.”

But when it comes to memory, the landscape is entirely different. “Everyone wants to build their own silicon, but almost no one manufactures their own DRAM or HBM,” Nelms noted. “Until there is a major technological leap in memory technology, a massive expansion of production capacity, or new competitors enter the field, this imbalance will likely remain.”

The Cost of Success

This dynamic is, in many ways, a consequence of Nvidia’s own success. The company proved that AI compute was not just useful, but essential, and in doing so, it turned its own core product into a commodity that every major player wants to produce or source independently. As barriers to entry in processor design gradually fall, the market becomes more crowded, and pricing power weakens.

Meanwhile, memory remains a specialized, concentrated industry with high barriers to entry and limited room for rapid expansion. What was once a supporting component has become the limiting factor, allowing memory manufacturers to capture more of the economic value in the AI supply chain.

For Nvidia, the irony is hard to ignore: it created the blueprint for the modern AI infrastructure market, only to see that market evolve in a way that benefits others more than itself. As the industry matures, the question facing investors and executives alike is whether Nvidia can continue to innovate fast enough to stay ahead, or whether it will settle into a more stable but less profitable role as just one part of a larger, more complex ecosystem.