AI Pulse · Frontier Labs Watch

Anthropic just confirmed it’s building its own AI chips, and the reason is simple: cutting the cost of running Claude in half. According to the original report, the company is hiring engineers who’ve actually shipped semiconductor designs, with salary bands running $320K to $485K. Samsung is reportedly in the mix as a manufacturing partner, while Anthropic keeps its existing Google TPU and Broadcom capacity running.

This matters because inference cost, the expense of actually running AI models at scale, is the hidden multiplier in every AI project. If you’re sending thousands of API calls a day to Claude, you’re paying for compute on someone else’s chips. Anthropic wants to control that stack. A 50% cost reduction doesn’t just improve their margin. It lets them undercut OpenAI and Google on API pricing, which neither competitor can easily match without their own chip programs. OpenAI doesn’t have one. Google does, but it’s tied to their broader infrastructure.

For business owners, this is a signal that the AI tooling you rely on is about to get cheaper and faster, but also more fragmented. The companies building these models are now building the hardware underneath them. That means tighter integration, but also more lock in. If you’re planning a multi year AI strategy, you need to know which vendor controls what layer of the stack. This is the kind of structural shift we track inside the Omni Command Centre, where you can see how infrastructure changes ripple into cost and capability for the tools you actually use. The chip wars aren’t abstract. They’re about who can afford to keep training and running models at the scale your business will need in two years.