Fractile Hits $1B Unicorn Valuation with Massive $220M Funding to Break Nvidia’s AI Grip

The race to build the hardware powering the future of artificial intelligence just got a massive structural shakeup. London-based AI chip startup Fractile has officially closed a staggering $220 million Series B funding round, propelling the company into the coveted unicorn bracket with a post-money valuation hitting $1 billion.

The mega-round was co-led by heavy hitters Accel, Factorial Funds, and Peter Thiel’s Founders Fund, alongside notable angel backing from former Intel CEO Pat Gelsinger. This massive capital injection signals a major shift in venture capital focus: the industry is moving past basic software apps and pouring billions into fixing AI’s ultimate scaling bottleneck—inference hardware.

The Massive Problem: The High Cost of AI Responses

While the tech industry spent the last few years obsessed with training massive foundation models, the financial strain has rapidly shifted to inference—the actual process of running a deployed AI model to generate text, code, images, or real-time reasoning steps.

As models take on complex, multi-step agentic workflows that require generating tens of millions of tokens, standard hardware architectures are starting to buckle under the pressure. Conventional AI accelerators, including Nvidia’s dominant H- and B-series GPUs, separate the processing unit from the high-bandwidth memory. Moving massive amounts of data back and forth between these two components creates an expensive "latency and energy tax" that limits how fast an AI can think.

Fractile’s Secret Weapon: In-Memory Compute

Founded in 2022 by Dr. Walter Goodwin out of the University of Oxford’s Robotics Institute, Fractile is attacking this bottleneck by entirely rewriting chip micro-architecture.

Instead of shuttling data back and forth, Fractile utilizes an in-memory-compute approach. By running matrix multiplications directly inside specialized computer memory cells located right alongside the compute logic, the architecture eliminates the traditional memory bandwidth constraints.

The Bold Performance Claim: Fractile’s internal metrics claim their specialized architecture can execute frontier model workloads up to 25 times faster while slashing data center power consumption and operational costs down to one-tenth (10%) of current GPU infrastructures.

Big Tech is Already Watching

Though Fractile's commercial chips aren't expected to hit the market until 2027, top-tier AI labs are already positioning themselves to reduce their reliance on Nvidia. Early reports indicate that Anthropic (the creators of Claude) has already held preliminary discussions with Fractile to explore integrating the startup's hardware as a multi-supplier hedge alongside Google’s TPUs and Amazon’s Inferentia chips.

Fractile plans to utilize the $220 million to aggressively scale its engineering hubs across London, Bristol, the US, and Taiwan. The funding is specifically sized to carry the design through its critical tape-out phase, software-stack development, and early enterprise customer integrations.

As sovereign AI pushes accelerate across Europe and the US, Fractile’s massive funding round proves that the battle for the physical backbone of artificial intelligence is just getting started.