Anthropic offers chip researcher to teach Claude to design processors

Anthropic posted a job for a researcher with experience in chip development, tasking the role with reinforcement-learning training of the language model Claude to design processors. The position is intended to feed engineering knowledge directly to Claude so the model can generate code that describes chip architecture, identify failures, and optimize the physical layout on silicon.

The initiative follows a broader hiring push. In June, Clive Chan, a senior AI-chip engineer who left OpenAI, joined Anthropic. Shortly thereafter the company advertised a silicon engineer role, suggesting the formation of an internal team that would cover the entire design chain—from architecture through verification to tape-out—with Claude embedded in the workflow.

In July Anthropic entered talks with Samsung to produce a custom 2-nanometer chip and approached SK Hynix for memory for an independent chip line. CEO Dario Amodei confirmed that agreements with both Korean firms have been finalized, but declined to disclose additional details such as timelines, production volumes or technical specifications.

The direction is clear: Anthropic is building a vertical capability where the model is not merely a compute consumer but a partner in the hardware it runs. It remains uncertain whether the resulting product will be a dedicated accelerator for training, inference, or both, and whether the company will sell the chip or retain it for internal use. What is certain is that the boundary between “open weights” and closed proprietary infrastructure is blurring as the model begins to design the silicon it executes on.