What happened
Anthropic signed a deal to rent about $45 billion in AI compute from Nscale, a British infrastructure company. The six-year agreement, reported by TechCrunch, will supply Anthropic with compute via Nvidia’s new Vera Rubin chip systems. The capacity is expected to power Anthropic's services starting in late 2027 from Nscale's data centre in West Virginia.
This follows a string of other large compute deals Anthropic has made this year with companies including Volta, AMD, SpaceX, Amazon, and Google.
How the room's reading it
The consensus is that this isn't a one-off deal — it's part of an aggressive compute land-grab. Frontier labs are all chasing as much AI horsepower as they can find. Anthropic watchers on X point to the string of recent partnerships as evidence of a clear strategy to keep pace with OpenAI's infrastructure. The sheer scale of these deals has infra teams talking about the future cost and availability of advanced models. Some builders are concerned this consolidation of compute could lock smaller players out of training, while others see it as a necessary step to push the boundaries of AI capability.
Sailfish's take
We don't see this as just another training run. This is about locking in inference capacity for the next six years. Frontier labs aren't just buying lottery tickets to find AGI — they're buying the factory to mass-produce whatever they build. For builders, this signals a future where the absolute top-tier models will carry a significant and persistent price premium. The real competition won't just be on model performance, but on who can afford to serve it at scale. We're advising teams to architect systems that can swap between models easily. Don't build your entire product on a single provider's most expensive API.