The question in every portfolio manager’s mind this morning is not whether to own Nvidia. It is whether the market has been measuring AI scarcity in the wrong place.
On Monday, AMD touched $615.89 a share and briefly crossed a $1 trillion market value for the first time, while Arm closed up 17.16% and Intel gained 12.14%. Nvidia gained 2.33% over the same session. That gap is not noise. When the market bids conventional server processors up five times as fast as the dominant GPU maker, it is expressing a view about where the bottleneck now sits.
Why Wall Street Cares
The session did not happen in isolation. On September 17, cloud provider Nebius announced it would raise pay-as-you-go prices from October 1, including CPU-only instances rising 25% and memory offerings rising roughly 41%. The second increase in three months suggests scarce accelerator capacity is giving specialist cloud providers more pricing power. Professionals watching the Nebius announcement noted something the headlines missed: the CPU increase was proportionally larger than the GPU increase. Nvidia H100 instances will rise roughly 17%, while AMD EPYC Genoa CPU rates rise 25%. That differential matters. It implies the processors managing inference orchestration are tighter than the GPUs running the actual model.
The Bull Case
Meta’s new Muse personal AI agent became the most downloaded free iPhone app in the U.S. for three consecutive days, according to Sensor Tower, stoking investor excitement about a boom in AI agent-fueled CPU demand. Agents run continuously, responding to requests around the clock. Every live interaction passes through a CPU before it ever touches a GPU. The more agents proliferate, the more that orchestration layer becomes a governor on the entire system.
AMD’s own numbers support the trade. The company has argued that 2026 is the first year the world uses more AI compute to run models than to train them. CEO Lisa Su said server CPU revenue is expected to increase more than 80% in the second half of 2026 compared to a year earlier, and projected server CPU sales to grow by more than 70% again in 2027. AMD has raised its growth forecast for the server CPU market from 18% to 35% annually, with expectations that it will exceed $120 billion by 2030.
The Bear Case
The skeptics in the investment committee have a real argument. Chip stocks sold off earlier in September after Anthropic chief executive Dario Amodei called for a slowdown in frontier AI development. AMD’s current P/E ratio stands at 157.95, considerably higher than its five-year median of 102.02. One session of CPU outperformance does not rewrite the infrastructure reality that Nvidia controls the most critical AI workloads. And Monday’s move may partly reflect short-covering rather than fresh institutional conviction.
The Mavens’ View
The professionals most worth watching here are not extrapolating from Monday’s price action alone. They are connecting three separate signals: Nebius raising CPU prices faster than GPU prices, Meta’s inference-heavy agent platform topping app charts, and AMD’s Lisa Su guiding to more than 80% server CPU growth in the back half of this year. Each individually is interesting. Together they form a coherent argument that inference at scale is CPU-constrained, not GPU-constrained.
Arm’s position is structurally distinct from AMD’s. CEO Rene Haas highlighted a path to more than $2 billion in customer demand for Arm’s AGI CPU data center chip across fiscal years 2027 and 2028.
What Investors Are Missing
The conversation fixates on whether CPUs are replacing GPUs. They are not. The overlooked consequence is that a boom in inference demand raises GPU revenue and CPU revenue simultaneously, but with different supply elasticity. Nvidia can add GPU supply faster than AMD and Intel can add server CPU wafer capacity. Constraint plus rising demand equals pricing power, which is exactly what Nebius confirmed on September 17. The companies that own that constrained layer in a world of proliferating agents may compound quietly while the GPU debate absorbs all the attention.
Stocks to Watch
AMD carries the broadest exposure: server CPUs growing above 80% this half, GPU accelerators ramping through Helios deployments with OpenAI and Meta, and a data center segment that more than doubled in Q2 2026. The trillion-dollar valuation is demanding, but the earnings trajectory is real.
Arm Holdings benefits from every server CPU sold regardless of brand, through its licensing model. Piper Sandler initiated coverage with an Overweight rating and a $320 price target, citing strong momentum in server CPU design wins and broad hyperscaler adoption of in-house Arm-based server chips.
Intel is the most complicated call. Monday’s 12% gain reflects the same inference-CPU thesis, but Intel is a company still rebuilding its manufacturing credibility. The upside is larger if the thesis holds; the downside is also larger if execution slips.
Nebius is the most direct read on whether the pricing holds. In its most recent quarter, Nebius said it closed four landmark deals averaging more than $1 billion each. Fourth-quarter results will show whether the price increases reach the income statement, or whether Monday bought demand that has not yet arrived.
