AI Pulse · Frontier Labs Watch

Anthropic has hired Amir Salek, the engineer who led Google’s TPU program through its first seven generations. He’s joining Anthropic’s compute team and will report to James Bradbury. On the surface, this is a senior hiring move. In practice, it’s a strong signal that Anthropic may want more control over the hardware used to train and run its AI models.

Today, Anthropic buys chips from Nvidia, Google and Amazon. Building its own silicon would be a much bigger step. Google and Amazon have both gone down that path, creating chips designed around their own AI workloads rather than relying only on general-purpose supplier hardware. Anthropic hasn’t announced a chip program, but hiring someone with Salek’s background is the first concrete indication it could be heading that way, according to the reported hiring details.

For business owners, this matters because the cost, availability and performance of AI tools are increasingly tied to compute infrastructure. If leading AI firms can control more of their chip supply, they may be better placed to manage capacity and reduce dependence on a small number of hardware vendors. That won’t change your AI budget next week. But it reinforces a simple point, the firms building AI are also competing to own the machinery underneath it.

This is the kind of thing we build into an AI command centre, tracking the infrastructure shifts that can affect the tools your business relies on.