The AI trade is no longer just Nvidia — the power stack is where the next winners sit
AI demand is still real, but this week’s tape says the market is starting to care less about who sells the chips and more about who can power, cool, and house them. After Nvidia’s strong-but-muted print, the sharper signal came from infrastructure deals tied to electricity, grid capacity, and data-center buildout.
The market is still talking about AI like it begins and ends with GPUs. We think that framing is getting stale. Nvidia’s May 20 results were strong enough to confirm that AI demand remains intact, yet the stock reaction was muted relative to the scale of the beat — a sign that the obvious part of the trade is now crowded. The more interesting move this week came from the physical layer beneath AI: power, cooling, utility capacity, and the infrastructure that makes high-density compute usable at all.
That matters because AI is no longer constrained only by access to chips. It is constrained by whether those chips can be deployed at scale without running into power shortages, thermal limits, or grid bottlenecks. Nvidia itself helped make that case. Around the same stretch as its earnings, the company backed IREN in a deal tied to as much as 5 gigawatts of infrastructure and up to $2.1 billion of investment. That is the market’s tell: even the company at the center of the AI boom is leaning into capacity, not just compute.
The valuation setup reinforces the shift. NVDA still looks exceptional on growth, with revenue growth of 65.5%, EPS growth of 66.0%, and a 63.0% net margin, all while trading at 33.57x earnings. But that is exactly why the reaction function is changing: the business is so dominant, and expectations are so high, that another huge quarter no longer guarantees a fresh rerating. By contrast, the market is increasingly willing to pay up for the bottlenecks around AI deployment because those bottlenecks are becoming more visible in real time.
