AI TRENDS | Anthropic Pitches Claude’s Chip-Design Capabilities to IPO Investors

Anthropic is showcasing Claude’s chip-design capabilities to investors ahead of a planned IPO later this month. According to Sina Finance, four people familiar with the matter said the offering could value the company at $1.5 trillion, and Anthropic told investors that its new 15-person chip-design team is using Claude to engineer and design custom chips. Some Anthropic engineers say Claude could significantly shorten chip-design timelines compared with traditional workflows at major designers such as Nvidia and AMD. Anthropic began assembling an internal team in August to design custom chips for training and running Claude. The company has not disclosed plans to commercialize its chip-design capabilities, and it remains unclear how either Anthropic or OpenAI will turn advances in the field into customer products. Anthropic has also held talks with Samsung Electronics about a potential manufacturing partnership. OpenAI is also pursuing chip design as a potential business. Its CFO, Sarah Friar, said at a Goldman Sachs conference in September that chip design was among the company’s target areas, alongside financial services and life sciences. OpenAI says it designed its custom Jalapeño chip with its own advanced models and Broadcom in nine months, and plans to begin using it at scale next year. In September, the company also announced a multiyear agreement to develop models for operating Synopsys semiconductor design tools. Synopsys and Cadence Design Systems sell software and services used to design, simulate, and test semiconductors, circuit boards, and electronic systems before manufacturing. Their combined revenue is expected to reach $16 billion this fiscal year. Anthropic and OpenAI’s chip efforts could also create conflicts with supporters that supply chips through cloud contracts and for data centers. Amazon, Google, Nvidia, and AMD have committed a combined approximately $90 billion in investment to Anthropic, which currently uses chips from those companies. ChipAgents and Nvidia released a benchmark last month to assess AI models’ chip-design capabilities. Claude Fable 5.1 scored highest, with 16.6%. Anthropic has also asked some electronic design automation companies about buying or otherwise accessing chip-design data, according to an executive at an AI chip-design tools startup. Executives at startups in the field say AI could reduce the time needed to design a fully functional chip to months, but sensitive chip companies may prefer running open-source models on their own hardware over using external models.