Anthropic is hiring engineers to build its own AI chips

Anthropic is building an in-house silicon team, a job posting and reporting from Business Insider show. The Claude developer is hiring chip engineers at salaries up to $485,000, and wants people who have taken semiconductors past research and into production.

The posting calls for candidates across chip design and verification, and requires "direct personal contribution" to the completion and shipment of semiconductor designs. The roles pay $320,000 to $485,000.

That salary range tells you what Anthropic thinks this problem is worth. Experienced chip engineers are scarce, and the companies that have them are not letting go easily.

The team sits inside what Anthropic calls a broader "multi-chip" strategy. "The new team will complement, not replace, our existing use of hardware from AWS, Google, Nvidia, and AMD," the company said in a statement reported by Business Insider.

Anthropic did not say when its first custom chip might arrive, or whether it plans to manufacture the processors itself. A June report from The Information said Anthropic was in talks with Samsung Electronics about building custom AI chips. The company had not confirmed that effort until now.

I think the "complement, not replace" framing matters more than the hiring news itself. Anthropic is not trying to exit the cloud. It is trying to change the terms of the conversation with its cloud partners, and designing its own silicon is the strongest negotiating position an AI lab can hold short of building a data center empire.

The strategy has precedent. Apple's transition to in-house silicon remade the Mac by building hardware around its own software. Anthropic's goals are different, but the principle is the same: hardware built for one workload beats hardware built for everyone.

The economics are the real driver. Training and inference costs scale with every generation of Claude, and every dollar spent on someone else's chip is a dollar without a margin attached.

Custom silicon can cut that cost while locking in capacity. The chip shortage is real enough that Moonshot AI paused registrations for Kimi K3 this week after demand surged past available infrastructure.

That is the failure mode Anthropic is trying to design out of its own future.

A few things give me pause. Chip design takes years, not quarters, and the best in-house silicon teams in the industry still depend on foundries they do not control.

Samsung talks are just talks until a wafer starts moving. The salary range shows Anthropic is serious, but the timeline says this is a long game.

The announcement redraws the map of who competes with whom. OpenAI has reportedly pursued its own chip efforts with Broadcom, Microsoft has Maia, and Google has TPU.

Now Anthropic has a team. The AI labs are no longer competing only on models, they are competing on the silicon under the models.

For companies buying AI capacity, the practical question is what this does to the supply picture over the next three years. If Anthropic ships its own chip at scale, inference pricing from every lab that does the same thing starts to compress. If the chips stay in-house and never reach the open market, the effect is smaller.

I am not sure which way it goes. The one thing I am confident about is that the era of every lab renting identical GPUs from the same three clouds is ending, and it is ending because the labs themselves decided it should.

Source