Fractile Raises $220 Million Series B to Revolutionize AI Hardware
UK-based startup Fractile has successfully raised $220 million in a Series B funding round, positioning itself as a pioneer in the development of next-generation inference hardware for artificial intelligence (AI). This impressive round of funding was led by venture capital giants Accel, Factorial Funds, and Founders Fund, with significant participation from investors such as Conviction, Gigascale, O1A, Felicis, Buckley Ventures, and 8VC.
Addressing the Bottleneck in AI Inference
Founded in 2022, Fractile operates on the belief that the future of AI progress is constrained by the time and cost necessary to produce meaningful results at scale. As AI systems evolve to tackle increasingly complex tasks requiring the generation of millions of tokens, Fractile is committed to developing chips and systems that make faster inference economically feasible. This involves innovations in AI research, chip microarchitecture, and foundry processes.
According to Walter Goodwin, CEO and founder of Fractile, the company was built on the premise that the most successful AI systems will eventually be limited not by their capabilities, but by the time they take to deliver useful results. Goodwin emphasizes the need to radically reinvent the hardware that powers cutting-edge AI models to unlock latent value and make speed viable at scale.
Economic and Technical Constraints in Inference
Goodwin highlights a critical issue: the raw capabilities of AI have advanced to a point where the time from query to output is the main limitation of their peak capabilities. As models improve, so does their ability to orchestrate longer output sequences. However, the economic constraints of inference have become a significant barrier.
He notes that current Large Language Models (LLMs) are already generating up to 100 million tokens to solve challenging problems. At the rate of approximately 40 tokens per second on existing chips, a single release of that length could take a month. The primary hurdle is memory bandwidth, which has not kept pace with current architectures, and this is the fundamental problem that Fractile is addressing.
Future Prospects and Expansion
Looking forward, Goodwin envisions not just accelerating current workloads but exploring entirely new workloads that Fractile’s hardware will enable. The company is actively recruiting talent in London, Bristol, San Francisco, and Taipei as it continues to grow and innovate.
For more detailed information, the original source can be accessed here.
“`
