The pickaxes are selling faster than the gold.
The Signal
Nscale, a U.K. cloud provider you've probably never heard of, is negotiating to buy a massive AI data center site in West Virginia. Their customer list reads like a who's who of AI's deep pockets: OpenAI, Microsoft. And they're not alone at the table. Amazon, Meta, and a Google cloud partner are all circling the same property.
This is infrastructure poker at scale. The bet isn't on who builds the best model. It's on who controls the physical layer where models get trained. Nscale positions itself as Nvidia's cloud ally, which means they're probably getting preferential access to H100s and whatever comes next. In the GPU scarcity game, that's worth more than brand recognition.
West Virginia makes sense if you squint. Cheap power, land that doesn't cost Bay Area money, and a state government hungry for the kind of economic development that doesn't involve coal. The competition for this site tells you everything about where the real constraint is. It's not talent anymore. It's not even capital. It's power and cooling and physical space to rack hundreds of thousands of chips that run hot enough to cook on.
The companies bidding here aren't building data centers for customers. They're building them because their own infrastructure roadmaps demand it. When your cloud allies are competing with your cloud customers for the same dirt, the supply-demand curve has gone vertical.
The Implication
Watch where the infrastructure money flows, not where the model hype goes. If you're building anything in the agent economy, your costs are about to get more predictable (good) but your vendor options more limited (bad). The companies that own the compute layer will decide who gets to play in Web4. Position accordingly.
Source: The Information