Most of us rent an apartment a month at a time and still stress about the lease. Anthropic just signed up for six years of an entire power grid's worth of computing capacity, running on chips that haven't been manufactured. Different tax bracket, different relationship with certainty.
460 Megawatts, One British Startup, Zero Chill
Anthropic has agreed to a six-year, $45 billion deal with UK infrastructure firm Nscale for roughly 460 megawatts of compute capacity out of a flagship data center Nscale is building in West Virginia, according to Bloomberg and CNBC. The capacity will run on Nvidia's next-generation Vera Rubin chips — silicon that won't even start coming online until late 2027.
This isn't Anthropic's first massive infrastructure land grab this year, either. It follows a $10 billion deal signed earlier this month with Norway-based Volta for data center capacity, and it lands as Anthropic reportedly prepares for a public offering, having filed confidentially with the SEC back in June.
Buying Tomorrow's Chips With Today's Money
Locking in six years of compute before the chips generating it have even shipped is a bet that AI demand keeps compounding rather than just holding steady — and it's the kind of bet a company gets more comfortable making when it's trying to show future investors that capacity, not ambition, is already secured.
The real story here isn't really about Anthropic specifically, it's about what this says about the AI compute market as a whole. Model quality stopped being the bottleneck a while ago; raw power and silicon are the constraint now, and the companies racing to lock up gigawatts years in advance are the ones betting the shortage gets worse before it gets better. When the "cutting edge" purchase is measured in megawatts instead of GPUs, you know the game has changed shape.
Six years is a long time to bet on a chip you haven't seen yet — but at this scale, betting late is apparently the riskier move.
Enterprise AI adoption is happening at every scale right now, not just the multi-billion-dollar infrastructure tier — if you want help figuring out where AI actually earns its keep in your business, reach out and let's map it out.
Source: CNBC