Anthropic is getting into the chip business — sort of.
The company behind the Claude AI assistant has announced that it is assembling a team to co-design custom ASICs, or application-specific integrated circuits, aimed at AI "inferencing" workloads, according to Tom's Hardware. Samsung has been reported as a manufacturing partner for the effort.
A quick translation: inference is the everyday work of running an AI model — the part that happens each time you type a question and Claude answers. That's distinct from training, the expensive up-front process of building the model in the first place. An ASIC is a chip built to do one narrow job extremely well, rather than a general-purpose processor that can do many things adequately.
Tom's Hardware frames the move as a way for Anthropic to sidestep costly Nvidia GPUs, which are the default hardware for AI work and are in famously short supply. The publication reports that designing its own accelerator would give Anthropic greater control over its compute buildout, and let it tune its AI models to run more efficiently on hardware it helped design — a two-way optimization that isn't possible when you're renting someone else's silicon.
The news was picked up across outlets including Fstoppers, which described it as Anthropic building a custom chip team for Claude, and eGamers.io, which characterized it as the company starting work on in-house silicon.
The available reporting is thin on specifics — no timeline, performance targets, or deployment date has been detailed in these accounts.
Why it matters: the cost of running AI models, not just building them, is becoming the industry's central expense, and control over the underlying chips is fast becoming the thing that separates AI companies from one another.