---
title: Anthropic Is Designing Its Own AI Chips and Training Claude to Design Them Too
description: Anthropic says it has taken design authority over the chips its models run on. Its own job listings show a program running from hiring to tapeout.
author: Darie Nani (Editor-in-Chief)
date: 2026-08-06T18:01:00.297Z
updated: 2026-08-06T18:30:40.777Z
canonical: https://www.sovereignmagazine.com/article/anthropic-custom-silicon-team-claude-chip-design
image: https://cdn.nanimediahouse.com/anthropic-custom-silicon-team-112789.webp
categories: Artificial Intelligence
content_type: News
region: Global
publication: Sovereign Magazine
schema_type: Article
---

Anthropic has taken design authority over the chips its models run on, and it is now hiring the team to use it. Between 3 and 6 August the company posted a cluster of roles on its own job board that set out what the program consists of: a silicon engineering search running across eight disciplines at once, a program manager whose milestones end at tapeout, and a hardware lab in the Bay Area being stood up to receive the machines.

Anthropic [trains and serves Claude on other companies' accelerators](https://www.sovereignmagazine.com/article/anthropic-volta-bitdeer-norway-compute) today, across what its listings call multiple hardware platforms. Amazon designs Trainium, which it calls a purpose-built AI chip aimed at the best economics for high performance AI training and inference at scale. Google designs Tensor Processing Units, which it describes as custom accelerators co-designed with open software to power the entire AI lifecycle. Anthropic is going the same way, and its listings are unusually specific about how.

## A Small Senior Team, With Most of the Work Pushed Out to Partners

The Silicon Engineer listing, which pays $320,000 to $485,000, is open across front-end design, pre-silicon verification, physical design, design for test, analog and mixed-signal, technology and foundry, CAD and design infrastructure, and packaging and signal integrity. Anthropic describes the hire as one of a small number of people covering a large surface, someone who has shipped silicon before and is comfortable making consequential calls without a large organization behind them. Among the calls listed: where the company designs, buys or licenses technology, and where it keeps design authority rather than leaving it with a partner.

The [program manager listing](https://job-boards.greenhouse.io/anthropic/jobs/5380377008), posted on 6 August, is blunter about the stage the effort has reached. There is no process to inherit, it says, and the postholder will make the execution machinery the program runs on. It says the effort is partner-heavy by intent, built on a portfolio of outside ASIC design services, IP vendors, foundry and packaging relationships rather than a chip division.

> "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."
> — Anthropic job listing, Technical Program Manager, Silicon

## The Listed Milestones End at Tapeout and First Silicon

That listing sets out the sequence Anthropic expects to manage: architecture closure, RTL freeze, design verification, IP delivery, physical design closure, design for test, tapeout, the point where a finished design goes to a foundry to be made, and post-silicon bring-up.

The physical side is being assembled alongside it. A hardware systems architect role covers everything from the package boundary out through boards, racks, interconnect, power, cooling and the data center interface. A second program manager, on what its listing calls a new hardware team, handles contract manufacturers and optics vendors through the build stages and into production. The earliest of the group, posted on 3 August, is for a hardware lab manager, described as the first dedicated one, running racking, cabling, power distribution, cooling and the bring-up of new accelerator platforms in Anthropic's Bay Area labs.

## The Role That Teaches Claude to Design Chips Went Up First

On 13 July, three weeks before any of the engineering roles, Anthropic posted a [research engineer position](https://job-boards.greenhouse.io/anthropic/jobs/5231612008) on its reinforcement learning team whose stated job is to advance the models' ability to design silicon. Hardware design is difficult and unforgiving, the listing says, exactly the sort of domain the company wants Claude to excel at. The work is building reinforcement learning environments for [agentic RTL generation](https://www.sovereignmagazine.com/article/chipmind-rtl-canvas-ai-chip-design), formal and design verification, and physical design optimization, along with EDA tool latency and proxy rewards.

It pays $500,000 to $850,000, a ceiling $365,000 above the silicon engineering roles that would do the adjacent work by hand. The same expectation runs through the rest of the postings. Anthropic tells the hardware systems team it expects them to push on what AI-assisted hardware design and review can look like, and tells the silicon program managers the same about program management.

## FAQ

**Q: Does Anthropic make its own chips?**
Not yet. Its listings describe a program that has only just started hiring, with milestones from architecture closure through tapeout and first-silicon bring-up still ahead of it.

**Q: Which chips does Claude run on now?**
Anthropic says it runs its workloads across multiple hardware platforms and works from the chip level up with its silicon partners, which the listings do not name.

**Q: Can AI design chips?**
Anthropic is researching it. Its Chip Design RL listing describes building reinforcement learning environments for agentic RTL generation, verification and physical design optimization, with the stated aim of advancing its models' ability to design silicon. The listing does not claim the work is finished.

**Q: What is tapeout?**
The point at which a finished chip design is handed to a foundry for manufacture. It is the milestone Anthropic's silicon program manager listing is written to reach.

**Q: Why do AI companies design their own chips?**
Amazon designs Trainium and Google designs Tensor Processing Units for the same work, large-scale AI training and inference, and both describe the aim as better economics and performance at scale. Anthropic has now decided to take design authority over its own hardware rather than rely only on partners.