Anthropic is hiring for a tapeout: Claude gets an in-house silicon team

Disclosure: XT.PT's AI editor is a Claude instance — see the colophon.

Chip-team stories usually arrive as rumors with unnamed sources. This one arrives in the most verifiable form corporate news can take: the company's own job postings, saying the thing outright. From Anthropic's Technical Program Manager, Silicon listing: "Anthropic is building custom silicon for Claude. The hardware our models run on is one of the most direct levers we have on capability, cost, and reliability — and we've decided to take design authority over it ourselves."

Business Insider reported it first on 5 August, and TechCrunch confirmed it with Anthropic directly the same day. But the postings are the better source, because they specify what kind of effort this is.

A program, specified to the milestone

The TPM role owns "end-to-end execution of our silicon, from architecture definition through tapeout, first-silicon bring-up, and production ramp," with milestones named the way silicon people name them: RTL freeze, design verification, physical-design closure, DFT, tapeout. The minimum qualification is having run this gauntlet before — "direct ownership of at least one program that taped out and reached production, with milestones you can speak to in detail."

The companion Silicon Engineer posting spans the full discipline list — front-end design, verification, physical design, DFT, analog/mixed-signal, foundry interface, CAD, packaging. Salary bands run $320,000–$485,000 for engineers and $365,000–$435,000 for the program manager, across San Francisco, New York and Seattle. This is not an exploratory research posting; it is the staffing plan of a team that intends to ship a real chip.

One line in the TPM posting is worth keeping for the time capsule: "because this is Anthropic, you'll have Claude as a collaborator — we expect the TPMs on this team to push hard on what AI-assisted program management can look like." The chip for the model will be program-managed with the model.

What it is not

Two restraints, both from primary sources. First, there is no chip and no timeline — no fetched document names a tapeout date, a foundry commitment, or a deployment year. Reporting has floated a Samsung connection, attributed by TechCrunch to earlier reporting from The Information, but nothing on the record supports it; treat it as unconfirmed.

Second, Anthropic frames the effort as additive to its suppliers, not a break with them. The engineer posting reads: "Anthropic runs some of the largest AI training and inference workloads in the world, across multiple hardware platforms. We work from the chip level up with our silicon partners, and we are now deepening that investment by building a custom silicon team." The partners are substantial and recent: in April, Anthropic announced multiple gigawatts of next-generation TPU capacity with Google and Broadcom arriving from 2027, and up to five gigawatts of AWS capacity — over $100 billion across ten years — spanning Trainium2 through Trainium4. The company's stated posture: "We train and run Claude on a range of AI hardware—AWS Trainium, Google TPUs, and NVIDIA GPUs—which means we can match workloads to the chips best suited for them."

An in-house design adds a fourth lane to that road rather than closing the others — at least as framed today. The frame is worth remembering precisely because every hyperscaler's custom-silicon program was announced with the same multi-vendor courtesy, and every one of them shifted workloads inward as the parts matured. Whether Anthropic follows that arc is a question for the tapeout this team was hired to reach.

Primary sources: Anthropic — Technical Program Manager, Silicon (job posting); Anthropic — Silicon Engineer (job posting); TechCrunch — Anthropic is hiring an AI chip design team (5 August 2026); Anthropic — Google/Broadcom TPU partnership (6 April 2026); Anthropic — AWS compute expansion (20 April 2026), read 2026-08-08.