For two years, Wall Street tracked the artificial intelligence boom through one metric: who controls the most chips. That mental model is cracking, according to recent reporting.
The bottleneck is increasingly electricity. Electric utilities across the United States are now fielding interconnection requests from AI data center operators seeking 80 megawatts or more of dedicated capacity for a single facility — a load that dwarfs the 5 to 10 megawatts a conventional data center requires, according to MSN reporting. That means a single AI data center can consume more than double, and in some cases far more, the power of an ordinary one.
The shift reframes where the real constraints in AI development lie. As one analysis framed it, the next AI infrastructure race has nothing to do with chips — Wall Street has spent years pricing AI through compute, but power access, grid interconnection, and physical infrastructure are emerging as the binding limits on how fast the industry can grow.
Building a data center is relatively straightforward compared to securing a dedicated high-capacity grid connection, which can take years of permitting, utility negotiation, and transmission upgrades. That means companies that locked in power agreements early — or that own land near substations with spare capacity — may hold advantages that no amount of chip-buying can replicate.
This matters because it signals that the AI race is no longer purely a technology competition. It is increasingly a race for physical infrastructure — power lines, cooling systems, and grid access — which moves far more slowly than semiconductor manufacturing and is much harder to scale on demand.