Anthropic Touts Claude’s Chip Design Skills to IPO Investors

Anthropic’s AI models have helped uncover serious cyber vulnerabilities and transform the software coding profession. Now the company is seeing just how much of an advantage its models can give it in unfamiliar terrain: designing its own AI chips.

The company, which is barreling toward an initial public offering later this month that could value it around $1.5 trillion, has told investors in recent meetings that its new 15-person chip design team is already using its own Claude models for engineering and design work on its custom chips, according to four people familiar with the discussions. Its engineers have found Claude cuts down the time it takes to design chips compared to their estimates of traditional workflows at large chip designers such as Nvidia and Advanced Micro Devices, said some of the people.

The efforts come as rival OpenAI has been highlighting chip design as a potential source of revenue. OpenAI’s chief financial officer, Sarah Friar, said at a Goldman Sachs conference in September that chip design is among the business sectors it is targeting, alongside financial services and life sciences. She said the ChatGPT maker is relying on its own frontier AI models to design its Jalapeño custom chip, which it plans to start using in significant quantities next year.

Both companies have largely focused on how AI could improve the way chips run and train their own models. It’s not clear how they will achieve that, or how the companies plan to turn such advances into products for their customers. Anthropic hasn’t disclosed any plans to commercialize its AI for chip design.

Still, helping customers design and test chips using AI represents a potentially huge market for both companies. Publicly traded Synopsys and Cadence Design Systems both sell software and other services to help chip engineers design, simulate and test different semiconductor chips, circuit boards and electronics systems before they are manufactured. The two companies are expected to generate a combined $16 billion for their current fiscal years.

But the custom chip efforts of Anthropic and OpenAI could potentially stoke tensions with some of their biggest backers, which also supply them with chips via cloud computing contracts or directly with data centers. Anthropic uses chips made by Amazon, Google, Nvidia and AMD, which have committed to investing about $90 billion altogether into the startup. If Anthropic or OpenAI end up making AI models that are far better at chip design than others’ models, they could also keep the most cutting-edge versions of those models to themselves, at least at first, giving them a leg up over other chip design firms.

Anthropic’s efforts to use Claude for chip design follow the company’s August announcement that it is building an in-house team to design ​custom chips to train and run its Claude AI models. It previously held talks with Samsung Electronics as a potential manufacturing partner, The Information reported.

It can take years to develop new chips, largely due to the tedious effort of testing and verifying different potential designs that could involve trillions of tiny circuits. But OpenAI said it took just nine months to develop Jalapeño with Broadcom using OpenAI’s most advanced models. And executives at startups that aim to use AI to speed up chip design said AI might be able to cut the time needed to design a fully functioning chip to a few months—or eventually even less time.

“Hardware design is difficult and unforgiving—exactly the sort of domain we want Claude to excel at,” wrote Anthropic in a posting for a research engineer to advance its models’ silicon-design abilities.

Anthropic’s models have proven the most adept at design by one measure. According to a benchmark released by startup ChipAgents and Nvidia last month that aims to measure how good various AI models are at chip design, Claude Fable 5.1 scored the best. That said, there still seems to be a long way to go for AI to fully take over chip design workflows; the model scored 16.6% on the test.

If Anthropic eventually decides to sell its AI to help other companies design chips, such Claude-based tools would compete with Cadence and Synopsys, which develop electronic design automation software. (OpenAI in September announced a multiyear agreement to develop a model specialized for using Synopsys’ tools for semiconductor design.)

One executive at a startup selling AI tools for chip design said Anthropic has approached some EDA companies asking to buy or otherwise use some of their chip design data—presumably to help make its models better at chip design.

Still, it may be an uphill battle for Anthropic to convince some of the larger EDA or chip manufacturing companies to use its models across all stages of the chip design process. Due to potential privacy concerns, the highly secretive chip companies may prefer to run instances of open-source models on their own hardware.