---
title: Fractile raises 720 million dollars in Series C funding
url: https://www.dataloco.com/en/fractile-raises-720-million-dollars-in-series-c-funding
published: 2026-11-14T19:22:24+00:00
language: en
section: AI
source: https://www.techtalkthai.com/british-inference-chip-developer-fractile-raises-720m-in-funding/
organizations: Fractile, Thrive Capital, Founders Fund, Lightspeed, Redpoint, Alphabet
publisher: Dataloco
---

# Fractile raises 720 million dollars in Series C funding

Fractile, a British startup developing artificial intelligence inference chips, has completed a Series C funding round worth 720 million dollars. The company was founded in 2022 by Walter Goodwin, an artificial intelligence researcher who serves as chief executive officer. The new investment brings the company's valuation to 7 billion dollars. Thrive Capital, Founders Fund, Lightspeed, and Redpoint led the round. More than 12 other investors participated, including the U.K. Sovereign AI Fund, British Business Bank, and CapitalG, the investment arm of Alphabet. Fractile projects that its chips will deliver performance improvements of up to 100 times compared to current silicon. The company also states that its technology can reduce energy consumption by up to 90 percent. This efficiency is achieved by eliminating the need for artificial intelligence models to move data to external high bandwidth memory. Large language models generate answers through complex processing steps. After each step, the model sends results to external high bandwidth memory chips and retrieves them for the next step.

This repeated data movement consumes significant time and causes inference delays. Fractile is not the only artificial intelligence chip manufacturer addressing this bottleneck. Cerebras Systems, which went public in May, offers wafer-sized accelerators that reduce data movement by storing data in large on-chip caches. These caches are faster than external high bandwidth memory modules. The company's latest processor, the WSE-3T, can store up to 44 gigabytes of data. Cerebras uses static random access memory technology for its on-chip cache. Static random access memory cells consist of six transistors each. These transistors can read and write data faster than the single transistors found in high bandwidth memory cells. However, static random access memory has limitations, including high cost and limited capacity. Fractile uses a different approach to reduce data movement. Instead of integrating static random access memory into the inference chip, the company places standard dynamic random access memory directly on top of the chip. Positioning memory modules above the processor's logic circuits shortens the distance data must travel.

This configuration speeds up data transfer and reduces the electrical current required for data movement. Samsung Electronics is also developing similar technology. In August, the company launched a special high bandwidth memory version called zHBM. This technology can be stacked on artificial intelligence accelerators. It uses microscopic copper wires to send data to the main chip and receive data back. Samsung states that this technology will be eight times faster than high bandwidth memory version 5. High bandwidth memory version 5 is a future version with higher performance than currently deployed versions. Fractile estimates that its chips will allow large language models to process thousands of tokens per second. The company plans to deliver the processors as rack-scale systems with specialized software. Many artificial intelligence chip manufacturers offer custom compilers. These software tools convert customer large language models into formats compatible with the company's chips. Fractile expects to begin delivering its chips next year.

## This story in other languages

- [Français](https://www.dataloco.com/fr/le-developpeur-de-puces-fractile-leve-720-millions-de-dollars)