Tuesday, November 17, 2026Tue, Nov 17 Updated

Mergers, acquisitions and startup deals, and what they mean.

Fractile Raises $720 Million at $7 Billion Value for AI Chips

British AI chip startup Fractile raised $720 million in a Series C led by Thrive Capital and others at a $7 billion value. Shipments start next year.

Detailed view of RAM sticks and microprocessors on a motherboard.
Photo: Sergei Starostin / Pexels

Fractile Ltd., a British developer of artificial intelligence inference chips, has raised $720 million in a Series C round that values the company at $7 billion, SiliconANGLE reported, citing Reuters. The money comes from a group led by Thrive Capital, Founders Fund, Lightspeed and Redpoint.

The company expects to start shipping its chips next year.

Who put in the money

About a dozen other investors joined the four lead firms. They include the U.K. Sovereign AI Fund, the British Business Bank and CapitalG, the investment arm of Alphabet Inc. Fractile was founded in 2022 by AI researcher Walter Goodwin, who is its chief executive.

What the chips are meant to do

Inference is the stage where a trained AI model produces answers to prompts. A large language model completes a series of steps for each response, and after each step it writes its results to an off-chip memory chip called HBM, then reads them back to start the next step. SiliconANGLE says those repeated round-trips take time and slow inference down.

Fractile’s design removes the need to move data to off-chip HBM. The company reportedly expects up to 100 times better performance than current silicon and promises a 90% cut in energy use. It estimates its chips will let language models process several thousand tokens per second, and it plans to sell them in rack-scale systems bundled with custom software.

How rivals tackle the same bottleneck

Fractile is not alone in going after data movement. Cerebras Systems Inc., which went public in May, builds wafer-size accelerators that keep data in a large onboard cache. Its newest processor, the WSE-3T, can hold 44 gigabytes. The cache uses SRAM, which reads and writes faster than the memory cells in HBM because each SRAM cell has six transistors against one. The trade-off is that SRAM is expensive and limited in capacity.

Fractile takes a different route. Instead of adding SRAM, it stacks standard DRAM directly on top of its processors. Putting memory right above the logic circuits shortens the distance data travels, which speeds up transfers and lowers the electricity needed to move it.

Samsung Electronics Co. is working on a similar idea. In August it introduced an HBM variant called zHBM that can sit on top of AI accelerators and uses tiny copper wires to link to the host chip. Samsung says it will be eight times faster than HBM5, a future version of HBM.

What this means for the AI chip market

The round adds a well-funded British name to the field of companies building alternatives to current inference hardware. Fractile’s performance and energy figures are the company’s own expectations and promises, and the chips are not yet shipping, so customers have no production results to compare against rivals such as Cerebras.

The first shipments are due next year, which is when those claims can be tested.

Source: SiliconANGLE