Exclusive / company strategy

Anthropic Touts Claude's Chip Design Skills to IPO Investors

Qianer Liu, Valida Pau and Phoebe Liu report, citing four people familiar with the discussions, that Anthropic has told investors in pre-IPO meetings that its new 15-person chip design team is already using Claude for engineering and design work on its custom chips, and that its engineers find Claude cuts design time versus their estimates of traditional workflows at Nvidia and AMD. Anthropic announced the in-house chip team in August and previously held talks with Samsung Electronics as a potential manufacturing partner. Rival OpenAI says it built its Jalapeño chip with Broadcom in nine months using its own frontier models, and CFO Sarah Friar has named chip design as a target sector. Claude Fable 5.1 topped a ChipAgents/Nvidia chip-design benchmark released last month, but scored only 16.6%. A startup executive said Anthropic has approached some EDA companies to buy or use their chip design data. Caveats: Anthropic hasn't disclosed any plan to commercialise AI for chip design, it isn't clear how either lab will turn these advances into products, and secretive chipmakers may prefer to run open-source models on their own hardware.

Why it matters

A company-strategy signal for the most-watched IPO of the month: a frontier lab moving into custom silicon and possibly EDA, which could put it in tension with backers Amazon, Google, Nvidia and AMD (about $90B committed between them) and threaten Cadence and Synopsys. It gives Dealroom a concrete update for Anthropic's profile and a theme for its AI chip design startup coverage (ChipAgents and peers).

Executive takeaways

  • Team: Anthropic's in-house chip design team is 15 people, announced in August; it uses Claude for engineering and design of custom training and inference chips.
  • Pitch: Anthropic is presenting this to IPO investors ahead of a listing later this month that could value it around $1.5T.
  • Rival: OpenAI built Jalapeño with Broadcom in nine months using its own models and plans significant deployment next year; Sarah Friar lists chip design alongside financial services and life sciences as target sectors; OpenAI also has a multiyear deal with Synopsys to build a model specialised for Synopsys tools.
  • Benchmark: Claude Fable 5.1 scored best on the ChipAgents/Nvidia chip-design benchmark, but only 16.6%, so full automation is far off.
  • Data: Anthropic has approached some EDA companies about buying or using chip design data (one startup executive's account).
  • Supply chain: past talks with Samsung Electronics as a manufacturing partner; backers Amazon, Google, Nvidia and AMD have committed about $90B to Anthropic.
  • Caveats: no disclosed commercialisation plan; customers may prefer self-hosted open-source models for privacy.

What The Information may be missing

The piece doesn't quantify the claimed time savings or name which chips or design stages Claude is used for, and the main sources are investor-meeting accounts rather than documents. It doesn't size the AI-for-EDA startup field (ChipAgents and others) or their funding, nor how Cadence and Synopsys are responding beyond the OpenAI-Synopsys deal. It also leaves open the conflict-of-interest question for the S-1: whether custom-chip plans appear as a risk to existing compute contracts with Amazon, Google and Nvidia.

Read the full article: The Information

More top stories