Fractile, a British AI chip startup, has raised $220 million to support the development of its specialised inference chips.
The series B was led by venture capital firms Accel, Factorial Funds and Peter Thiel’s Founders Fund, with participation from Conviction, Gigascale, 01A, Felicis, Buckley Ventures and 8VC.
The company was established in 2022 and targets a growing sector of the AI market: inference chips. These are built to generate tokens more quickly than systems that run on GPUs like those supplied by industry leader Nvidia, with the trade‑off that they are not as useful for training artificial intelligence.
It said in a release announcing the financing that “raw AI capacity has already reached the point where time from query to output is the key limit to frontier capabilities”, and that current token rates are insufficient for solving complex problems.
Current systems can produce around 40 tokens per second, meaning problems can take months to solve, Fractile said. The company’s systems are built “from the ground up” to shorten this process, targeting production of over 1,000 tokens per second.
The company does not provide technical specifications of the chips it is designing, but it told Forbes its chips do not rely on traditional high‑bandwidth memory chips, nor on‑chip static RAM, two of the most commonly used forms of memory in AI. In April, it was reported that the company’s chips were static‑RAM‑based.
The company has not specified what it will use the funds for, but it was reported in April that they intended to use them for the creation of a hardware engineering facility in Bristol.
Another AI chip manufacturer, Cerebras, launched its initial public offering yesterday at $185 a share, seeking to raise $5.5 billion.
