Nvidia, a long-time leader in advanced computing, has recently hit an unexpected patch of turbulence. Despite its consistent trajectory of innovation and growth, the company’s stock price has notably dipped, signaling a shift in the artificial intelligence (AI) infrastructure market.
Over the past few months, Nvidia’s stock has fallen 15% from its peak in May, even as revenue projections continue their upward climb. This unusual decoupling means investors are now paying less per dollar of Nvidia’s anticipated profit than for the average large American company, suggesting a recalibration of market expectations.
The Unexpected Shift in the AI Gold Rush
While investment continues to pour into AI infrastructure, the focus has surprisingly pivoted away from graphics processing units (GPUs). Instead, a significant portion of capital is now flowing into memory companies, highlighting a critical new bottleneck in data center development.
Take Micron, a global leader in DRAM manufacturing, for example. In the same period Nvidia’s stock dipped, Micron’s value nearly tripled, cementing memory as the new hot commodity in the AI trade. This dramatic shift indicates that while last year’s alarming GPU shortage has eased, data centers are now voraciously consuming all the memory they can acquire.
For those who have admired Nvidia’s groundbreaking technological journey, this turn of events might feel a little perplexing. The company’s rise has been fueled by genuinely impressive innovations, from the development of CUDA, its widely adopted programming platform, to pushing the very limits of GPU performance.
Nvidia’s GPUs rank among the most intricate devices ever created, residing at the cutting edge of human engineering. Their success has reshaped industries, setting new benchmarks for high-performance computing and AI research.
Unpacking the Compute vs. Memory Dynamic
The narrative for memory companies like Micron is remarkably different and, in many ways, much simpler. These firms specialize in building high-bandwidth memory (HBM) chips, components meticulously engineered to move data in and out of processors with blistering speed. For two decades, these chips have seen steady, incremental improvements.
Without radical changes to the chips or their manufacturers, the services they provide have become immensely valuable. With demand vastly outstripping the industry’s capacity, memory companies have increased prices tenfold over the past year, enjoying an unprecedented boom.
Indeed, a look at the spot price for DRAM—the open market rate for these chips—since 2023 reveals a dramatic surge, particularly from mid-2025 onwards. This isn’t due to a sudden technical breakthrough, but rather the industry’s collective underestimation of just how much memory would be required for the massive global data center buildout.
In stark contrast, the spot price for an hour on an Nvidia H100 GPU has mirrored Nvidia’s stock performance. After peaking around $3.20 an hour in May, the price has steadily declined. This illustrates that Nvidia’s valuation is intrinsically linked to the price of compute, which is now undeniably falling.
Conversely, Micron and its peers find their fortunes tied directly to the escalating price of DRAM. This divergence highlights a fundamental shift where the cost of processing power is decreasing, while the cost of feeding that processing power with data is soaring, creating a unique economic imbalance.
The Self-Inflicted Wound of Success
Wayne Nelms, co-founder and CTO of Ornn, a compute market analysis platform, attributes this disparity to a straightforward issue of supply and demand. Major tech players like Google, Amazon, Microsoft, and OpenAI have all introduced their own custom processors.
While these proprietary chips may not always match Nvidia’s latest offerings, they are “good enough” to significantly increase overall compute supply and drive down market prices. This influx of alternative accelerators dilutes the premium once commanded solely by Nvidia’s technology.
As Nelms aptly puts it, “More GPU and accelerator players are entering the market. Everyone wants to make their own silicon, but no one is making their own DRAM.” This statement encapsulates the core of the problem for Nvidia and the boon for memory providers.
Until there’s a significant technological leap in HBM, a fundamental shift in its supply-demand dynamics, or new players enter the memory market, Nelms expects current trends to persist. This creates a frustrating predicament for Nvidia, a scenario largely born from its own remarkable success.
By proving the immense value of advanced compute, Nvidia has inadvertently cultivated a fiercely competitive market that everyone now wants a piece of. Meanwhile, companies dealing in seemingly simpler, foundational technologies are quietly accumulating significant wealth, benefiting from the very ecosystem Nvidia helped create.
Source: TechCrunch – AI
