ANTHROPIC SIGNALS PARTNERSHIP WITH SAMSUNG TO BUILD OWN AI CHIPS

Anthropic is in discussions with Samsung Electronics about manufacturing a custom AI chip. The project is still in its early stages with no detailed design yet, and the company is still working out what the chip would do and how powerful it needs to be. That much is clear from reporting that emerged this week.

What is less clear is how quickly a company best known for building large language models can become a chip designer, especially when it has not decided whether the chip will handle inference, training, or some as-yet-unspecified workload. Chips from AWS, Google, and Nvidia remain central to Anthropic's strategy, the company told TechCrunch, and a diversified hardware stack will continue to be pivotal. But the hiring of chip engineers, including Clive Chan, who is expected to build out a dedicated chip group, signals that Anthropic sees a future where it needs something off the shelf.

you may like

THE ERA OF PICKING JUST ONE AI LAB IS OVER, SAYS VERCEL CEO

SPACEX BUYS CURSOR AFTER HISTORIC IPO

SAM ALTMAN OFFERS 5% OF OPENAI TO US GOVT

WHY TALK TO SAMSUNG?

Samsung is an unusual dance partner for a company that relies on Nvidia hardware. Samsung is a major partner of Nvidia, producing chips that Nvidia needs for training and running models. Samsung also uses Nvidia's software to manufacture its chips, and the two are working together on an AI chip factory in South Korea. Samsung has also discussed partnering with Google on chip-making efforts, meaning the company's foundry capacity is already heavily committed.

For Anthropic, approaching Samsung means competing for fab time and engineering attention alongside some of the biggest players in the business. There is no indication that Samsung has committed to anything yet, and Anthropic declined to add anything further regarding a potential partnership. The talks may be exploratory, a way to understand costs and timelines before committing to a foundry partner.

THE BROADER CHIP LAND GRAB

Anthropic is not alone in wanting custom silicon. OpenAI recently unveiled Jalapeño, its first in-house inference chip developed with Broadcom, claiming it is more efficient and delivers better performance-per-watt than competitor chips. Amazon and Google offer custom-built TPUs as part of their cloud offerings. AWS, Google, and Meta all run custom silicon tuned for AI workloads. Nvidia is the undisputed leader of the chip industry, but the math that drives companies to build their own chips is simple: whoever can build and run AI infrastructure more cheaply keeps more of the revenue. Custom chips are how you get there.

For Anthropic, the timing is notable. The discussions with Samsung emerged shortly after Reuters and others reported that Anthropic was considering producing its own chips to address shortages. The Information reported on Thursday that the company had contacted Samsung. The Information also noted that Anthropic has not yet decided what the chip will be used for, how it will fit into the server, or how powerful it will be. That suggests this is a capability-building exercise as much as a product plan.

WHAT A CUSTOM CHIP WOULD ACTUALLY MEAN

If Anthropic follows through, the implications are significant. The company currently rents compute from its cloud partners. A custom chip would let it design hardware specifically for the mathematical operations its models need, potentially cutting costs and latency. It would also reduce reliance on a single supplier, which matters when Nvidia's supply is perpetually constrained, and its prices are rising.

But the barriers are real. Chip design is expensive, requires years of engineering, and founders like Samsung are already running at capacity. Anthropic is starting later, and it is starting without a clear specification. The chip group being built by Clive Chan will need to grow fast to make this viable.

WHAT ANTHROPIC HAS NOT SAID

Anthropic declined to comment on any chip roadmap of its own. The company declined to add anything further regarding a potential partnership with Samsung. The only official comment was that chips from AWS, Google, and Nvidia remain central to its strategy, a statement that leaves room for a custom chip as a complement rather than a replacement.

The most likely outcome is that Anthropic develops a chip for a specific workload, probably inference, and continues to buy Nvidia for training. That would mirror what OpenAI did with Jalapeño and what AWS does with its custom silicon. It would give Anthropic leverage in pricing negotiations with its cloud providers and a path to lower costs at scale.

what to read next

ALEX KARP SAYS AI COMPANIES DONT KNOW HOW UNLIKEABLE THEY ARE

IS AI THE NEW SPACE RACE? OPENAI IPO FILING KINDA CONFIRMS IT.

APPLE TAKES TO OPENAI TO COURT FOR POACHING AND STEALING TRADE SECRETS

LONG GAME

This is not a move that changes anything this year or next. Chip design cycles are measured in years, and Anthropic's chip is not even in the design phase yet. But the signal matters. In a market where the biggest AI companies are all racing to control their hardware supply chains, Anthropic cannot afford to be the only one still renting. The talks with Samsung are a bet on the future. For now, the company will keep running on other people's silicon. But the fact that it is even having these conversations says something about where the industry is heading.