Fractile’s $6.5B talks show how one Anthropic deal reprices AI chips

An agreement with Anthropic has led British chip company Fractile to discussions of a $6.5 billion valuation, indicating investors’ eagerness to re-evaluate companies capable of competing with Nvidia. The larger message is equally significant: frontier AI companies are now influencers in the chip space, as customer commitments may lift company valuations even before chips are produced.

Fractile is currently in discussions to raise about $600 million in funding at a pre-money valuation of $6.5 billion. Additional investment could come in at a different valuation, meaning that simply adding the full amount of $600 million would result in an inaccurate post-money valuation. Anthropic and Fractile did not provide comments about the deal, which is not finalized yet.

A sixfold jump built on one customer

Three months ago, Fractile was valued at approximately $1 billion after raising $220 million in May from investors, such as Accel, Founders Fund, and Factorial Funds. The latest figure is about six times bigger, even if this is not a perfect comparison since the valuation from May uses a post-money valuation while this one was discussed using a pre-money valuation.

The major new factor is Anthropic. Fractile has signed a preliminary deal to sell approximately $250 million worth of chips to Claude’s creator, with both parties looking forward to enlarging their cooperation. The chips are expected to be launched only in 2027, which means that the deal is rather prospective than operational in its nature.

This detail is very important for the following reason. Fractile’s revaluation is driven by customers’ validation and expectations about the future inference market, not by profits generated by chips in use.

Why Anthropic keeps spreading its chip bets

Anthropic’s approach sheds light on why the company’s endorsement is so powerful. The firm announced on August 5 that it is putting together a team of engineers to develop its own chips while still relying on products from Amazon, Google, Nvidia and AMD.

It has already made huge commitments outside. In April, Anthropic committed more than $100 billion to AWS technologies over 10 years for up to 5 gigawatts of computing power. It has also expanded its relationship with Google and Broadcom for multiple gigawatts of TPUs starting in 2027.

Anthropic’s success illustrates just how much computing power it is going to require. Its annual revenue exceeded $30 billion by early April, reached $47 billion by mid-May, and then broke the $65 billion mark at the end of July, according to various sources. This was a big jump from the approximately $9 billion it made at the end of 2025.

The collaboration with Fractile provides Anthropic with another way to acquire the computing power it needs while relying less on any one vendor.

An inference field is getting crowded and pricey

Fractile concentrates on inference, the type of computing utilized by pre-trained models to produce a response. The company holds that longer reasoning tasks will make latency, memory bandwidth, and costs increasingly significant.

Fractile founder and CEO Walter Goodwin put the company’s thesis this way in May:

“Inference is both the revenue engine of the AI industry and the rate-limiting factor on expanding it.”

Investors are placing similar investments in a number of competing companies, even though those companies vary greatly in the phases of commercialization they are in.

Etched is already well ahead of the pack, having deployed its technology to a customer and developed working hardware, while Groq provides global inference infrastructure. Meanwhile, Fractile and OLIX are still predicting deliveries in 2027.

That difference between valuation and actual commercial success is the risk behind the growing industry. Michael Ashley Schulman of Cerity Partners explains it well:

“Semiconductor history is littered with brilliant chips that never became great businesses.” – Michael Schulman

Fractile indeed derives value from Anthropic’s commitment, but the actual challenge is in 2027, when it must be seen if its chips can be delivered on schedule and perform economically. If so, today’s valuation may still be quite low. If not, any move to reprice six times from here would be much tougher to justify.

Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.