The founder of a UK chip developer looking to take on Nvidia and US tech firms in the AI chip race says he hopes to bring its chips to market in 2027- and claims they can run LLMs (Large Language Models) faster, cheaper and more energy efficient than rivals.
Speaking on the Tech.eu podcast, Walter Goodwin, Fractile CEO, said: “We are sprinting towards sort of implementation of really what is our first sort of product chip. We plan to send that to the fab in the second half of next year.”
Fractile is hoping to get its first chips into data centres in 2027.
Fractile is focused on LLM inference, the process where a pre-trained LLM generates text output, like answers to user prompts.
Goodwin says that Fractile's chips, unlike the usual approach to inference in which the AI chip refers to a training model database, fuse memory and inferences in a single chip.
However, Fractile, which has raised around £16m, and is backed by Kindred Capital, NATO Innovation Fund, Oxford Science Enterprises, Cocoa VC and Inovia Capital, is likely to require significant further capital as it looks to take on Nvidia, the hyperscalers and a raft of other well-capitalised tech firms in this highly competitive field.
Goodwin thinks more UK government support should be accorded to chips.
He said: "Today, to compete directly in the foundation model space, that’s something which it takes 10s of billions of dollars and ultimately a lot of that is predicated on hardware.
“A lot of that capital accrues to hardware. The main kind of victor so far, in terms of bottom line, from the current time in AI, has been Nvidia.
“I think semi-conductors are a great place for Britain to focus."
Would you like to write the first comment?
Login to post comments