An agreement with Anthropic has pushed British chip company Fractile into discussions of a $6.5 billion valuation, underscoring investors’ willingness to reprice companies that could compete with Nvidia. More broadly, the development highlights how frontier AI companies are becoming influential buyers in the chip market, with customer commitments capable of lifting valuations before chips are even produced.

Fractile is currently in talks to raise about $600 million in funding at a pre-money valuation of $6.5 billion. Any additional capital could be priced differently, so simply adding $600 million would not produce an accurate post-money valuation. Anthropic and Fractile did not comment on the deal, which has not been finalized.

A sixfold jump built on one customer

Three months ago, Fractile was valued at about $1 billion after raising $220 million in May from investors including Accel, Founders Fund, and Factorial Funds. The latest figure is roughly six times higher, although the comparison is not exact because the May round used a post-money valuation, while the current figure is being discussed on a pre-money basis.

The key new factor is Anthropic. Fractile has signed a preliminary deal to sell about $250 million worth of chips to Claude’s creator, and both sides are said to be looking to expand the relationship. The chips are expected to launch only in 2027, making the agreement more prospective than operational.

That distinction matters. Fractile’s higher valuation appears to be driven by customer validation and expectations for the future inference market, rather than revenue from chips already in use. The scale of the repricing also reflects the concentration of the market it is entering: Nvidia still supplies the large majority of chips used for AI training and inference, and its CUDA software keeps most developers building on top of its hardware. Against that backdrop, even a preliminary order from a frontier AI lab functions as evidence that a challenger’s technology is being taken seriously.

Why Anthropic keeps spreading its chip bets

Anthropic’s strategy helps explain why its backing carries so much weight. On August 5, the company said it was assembling a team of engineers to develop its own chips while continuing to rely on products from Amazon, Google, Nvidia, and AMD.

It has also made major commitments elsewhere. 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 is following a pattern now common across the industry. OpenAI has partnered with Broadcom to develop custom accelerators, Microsoft is building its own Maia chips, and Meta designs its MTIA silicon, each as a way to secure supply and contain costs in a market where Nvidia remains the default supplier.

Anthropic’s growth shows how much computing power it needs. According to various sources, its annual revenue surpassed $30 billion by early April, reached $47 billion by mid-May, and climbed above $65 billion by the end of July. That marked a sharp increase from about $9 billion at the end of 2025.

The collaboration with Fractile gives Anthropic another route to secure the computing capacity it needs while reducing dependence on any single vendor.

An inference market that is getting crowded and expensive

Fractile focuses on inference, the computing used by pre-trained models to generate responses. The company argues that as reasoning tasks become longer, latency, memory bandwidth, and cost will matter more.

Fractile founder and CEO Walter Goodwin described the company’s thesis in May:

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

That argument tracks with how AI products have evolved. Reasoning-style models that work through problems step by step produce far more intermediate output per answer, multiplying the memory traffic and computing time required for every response a service generates.

Investors are making similar bets across a range of competing companies, even though those businesses are at very different stages of commercialization. Etched is ahead of the pack, having already deployed its technology to a customer and built working hardware, while Groq provides global inference infrastructure. Fractile and OLIX, by contrast, are still forecasting deliveries in 2027.

That gap between valuation and commercial execution is the central risk in the sector. Michael Ashley Schulman of Cerity Partners summed it up this way:

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

Fractile’s value is being lifted by Anthropic’s commitment, but the real test comes in 2027, when it will be clear whether the company can deliver on time and operate economically. Before then, the nearer-term markers are whether the funding round closes at the valuation under discussion and whether the preliminary Anthropic order hardens into a firm contract. If Fractile clears those hurdles, today’s valuation could still prove conservative. If it cannot, a further sixfold repricing would be much harder to justify.