The biggest AI story this week wasn't a model. It was a building
Anthropic has signed a six-year deal worth around $45bn with the UK infrastructure firm Nscale, roughly 460 megawatts at a West Virginia data centre, running Nvidia's next-generation Vera Rubin chips. The site doesn't switch on until the end of 2027.
I read that as an architect, which may be why it lands differently.
$45bn doesn't buy a product. It buys future electricity, in a place that doesn't exist yet. Somebody has committed to six years in a building that is currently a construction site.
For two years the AI conversation has been about which model is cleverest and which benchmark moved. The bottleneck now isn't the algorithm. It's power and land. 460MW is roughly the draw of a small city, and that problem doesn't get solved on a GPU. It gets solved in planning consent, grid connections, and on site.
So when I talk to my children about AI, it isn't only about chatbots.
Some of the most sought-after people of the next decade will be electrical engineers, grid specialists, and the people who actually build. None of which any school currently files under "AI".
We keep asking which jobs AI will replace. The better question might be which jobs AI needs.
