Analysis
Anthropic confirmed on August 5 that it is building an in-house silicon team to design custom chips for Claude, the first time the company has publicly acknowledged the effort, according to TechCrunch. The confirmation came alongside a Silicon Engineer job listing on Anthropic's careers site, seeking engineers across front-end design, pre-silicon verification, physical design, design-for-test, analog and mixed-signal work, and packaging, with salaries ranging from $320,000 to $485,000.
Anthropic said the push is driven by the arithmetic of serving billions of tokens a day at a roughly $30 billion revenue run-rate -- co-designing hardware and models can cut cost per query in ways that buying off-the-shelf chips cannot. The company was explicit that this isn't an "escape from Nvidia": it plans to keep using Nvidia, AMD, Google TPUs and Amazon Trainium chips alongside whatever it eventually builds. By July 2, TechCrunch had already reported Anthropic in talks with Samsung as a potential manufacturing partner, though the chip's specifications and purpose haven't been decided.
“Google, Amazon and OpenAI have each spent years and billions building or backing custom silicon programs; Meta has done the same for inference.”
Google, Amazon and OpenAI have each spent years and billions building or backing custom silicon programs; Meta has done the same for inference. Anthropic is the last of the major labs to confirm one publicly, and a from-scratch chip program typically takes 18-24 months from design to working silicon even with committed funding -- this is a multi-year bet on 2028 economics, not a near-term Nvidia bill reduction. However, the risk case is straightforward: if Anthropic's tapeout slips past 2028 or the chip underperforms merchant silicon on cost-per-token, the company will have spent two-plus years of senior engineering time on a program that a bigger Nvidia discount could have replaced.