Anthropic is moving into silicon. According to a report carried by Cryptonews.net, the AI lab behind Claude has hired the founding architect of Google's TPU — the custom chip family Google built to train and run its own AI models — as part of a broader hardware push.
The hire lines up with a separate report from Geeky Gadgets, which says Anthropic is developing custom AI chips that could reduce the company's reliance on external hardware, and specifically on NVIDIA. That report also says Anthropic is reportedly assembling a silicon team, with salaries going as high as $485,000.
A bit of context helps here. Training large AI models today runs mostly on NVIDIA graphics processors, which are expensive and, at times, in short supply. Google's answer was the TPU: a chip designed from scratch for the specific math that AI models do, rather than a general-purpose part adapted for the job. Bringing in the person who helped originate that design is a strong signal that Anthropic wants to walk a similar path rather than simply buy more of what everyone else is buying.
It is worth being precise about what has and hasn't been established. Both items are reports, and the chip work is described as in development — not shipping. Neither source describes a timeline, a manufacturing partner, or how much of Anthropic's compute would eventually run on its own hardware.
Why it matters: if one of the largest AI labs can build its own chips, the economics of the entire industry — who pays whom, and how much it costs to run the models millions of people use — start to shift away from a single dominant supplier.