Anthropic is hiring a custom silicon team to design its own chips for running Claude, the company confirmed this week. The clearest public evidence sits on its own job board, where a new Silicon Engineer listing asks for people who have shipped silicon and opens searches across front-end design, pre-silicon verification, physical design, design for test, analog and mixed-signal, technology and foundry, and packaging and signal integrity.
A spokesperson confirmed the plans to TechCrunch and Business Insider on August 5, and told Ars Technica that Anthropic will keep a multi-chip approach, pairing its own designs with hardware from other companies as it scales. The Silicon Engineer listing advertises an annual salary of $320,000 to $485,000.
The goal is hardware and software co-design. Anthropic says its teams will design new hardware and models side by side, and the role is expected to partner with the company’s inference, performance, kernels, and infrastructure teams and to support first-silicon bring-up and debug. The listing describes the job as one of a small number of people covering a large surface, responsible for writing specifications, keeping interface definitions coherent, and making directional calls on technology and IP selection, including where to build versus buy versus license. It names outside partners such as ASIC houses, IP vendors, foundries, and test partners, and says Anthropic already works from the chip level up with its silicon partners. Anthropic has co-designed hardware with partners before; the shift is about bringing more of that expertise inside the company.
Anthropic is not alone in walking this path. OpenAI unveiled a custom inference chip called Jalapeño, built with Broadcom, in June. Google DeepMind has long run its models on Alphabet’s TPUs, Meta has designed and deployed its own MTIA accelerators, and Mistral is reportedly looking into the same move. Ars Technica points to two reasons behind the trend: the industry’s heavy reliance on Nvidia for the hardware that runs frontier models is a potential strategic vulnerability as compute demand outruns capacity, and chips designed for specific models can deliver better performance than a general-purpose part. That is the bet Anthropic is now making in-house.
Anthropic already secures compute from AWS, Google, Nvidia, and AMD. Last month The Information reported that Anthropic had been scouting Samsung as a potential manufacturing partner for a custom chip. Anthropic has not confirmed a foundry, a product name, or a timeline for first silicon.
If the effort delivers, the payoff is cost and control. Chips tuned to a specific model family can run it faster and more efficiently, and vertical integration eases the same supply pressure that has made GPUs scarce and expensive. It could also shape where frontier inference runs next: as developers explore cheaper, smaller, and open-weight models on their own hardware or on edge devices, a model maker with its own silicon has more freedom over that trajectory. The move tracks the wider industry argument that open weights and smaller models make their own hardware case.
For now Anthropic is still filling the team. Because the program depends on staffing, the company says benefits will take time to show up in Claude’s performance or pricing. The next milestone is the hiring round itself, and which of the listed silicon domains gets staffed first.
