Report: Anthropic Uses Claude to Help Design Custom AI Chips

On November 4, according to The Information, Anthropic is attempting to use its own Claude large model to directly participate in the full-process design of custom AI chips.

According to four people familiar with the matter, Anthropic plans to officially launch its initial public offering (IPO) later in November 2026, after which the company's valuation is expected to reach $1.5 trillion. In recent closed-door meetings held for institutional investors, Anthropic disclosed for the first time that its newly formed 15-person chip design team has fully integrated the Claude model to carry out engineering design work on custom chips.

Multiple people close to the project said that in testing, Anthropic's internal engineers found that with the assistance of the Claude model, the overall chip design cycle has been significantly shortened compared with the conventional workflows of traditional leading chip companies such as Nvidia and AMD, greatly reducing the time cost from architecture validation to front-end design. This progress also makes Anthropic the world's first general AI company to publicly deploy its own large model deeply into the chip design process.

In the same arena, Anthropic's direct competitor OpenAI is also making chip design a core expansion direction. OpenAI Chief Financial Officer Sarah Friar stated clearly at a Goldman Sachs industry conference in September this year that chip design is a new business track the company is focusing on, with its other layout directions including the two major fields of financial services and life sciences. She revealed that OpenAI is currently advancing the research and development and implementation of its custom chip Jalapeño based on its own frontier large model capabilities, and plans to achieve large-scale deployment and use of the chip starting in 2027.

Before this strategic shift, the hardware-related work of Anthropic and OpenAI mainly focused on optimizing the operating efficiency of large models on general-purpose chips and improving the resource utilization of their own AI model training processes. As of now, neither company has disclosed a complete commercialization path for chip design, and Anthropic has not yet publicly announced specific product plans for exporting the relevant capabilities.

Public information shows that Anthropic's custom chip layout can be traced back as early as August 2026, when the company first confirmed externally that it had formed an internal semiconductor team, with the goal of designing dedicated training and inference chips for the Claude series models. Since then, multiple industry sources have disclosed that Anthropic has held preliminary negotiations with Samsung Electronics on a potential chip foundry partnership, and has also approached a number of EDA startups to inquire about the possibility of purchasing or licensing relevant chip design datasets, so as to further train and strengthen Claude's professional chip design capabilities.