LONDON, November 11, 2026 — Fractile, a London-based startup, has raised $720 million in Series C funding, valuing the company at $7 billion, to develop memory-first AI chips that promise to revolutionize AI inference.
The Memory-Bandwidth Bottleneck: Why It Matters
Memory-first architecture addresses a critical issue in AI processing: the growing gap between computing power and memory bandwidth. Over the past 25 years, chip compute has grown a millionfold, while memory bandwidth has only increased 40-fold, creating a 25,000-fold disparity, according to Fractile founder Walter Goodwin. This imbalance means processors often wait for data, slowing down AI tasks like inference—the process of generating responses from models.
How Fractile’s Design Challenges Traditional GPUs
Fractile’s solution stacks custom DRAM directly on top of the processor die, keeping data on-chip and eliminating the traditional processor-to-memory shuttle. This contrasts with GPUs, which rely on a memory hierarchy that separates processing and memory units, creating latency. By integrating memory and logic, Fractile claims its chips can run large language models up to 100 times faster than existing hardware while cutting operational costs by 90%.
The Numbers: Speed, Cost, and Market Potential
Fractile’s chips target the AI inference market, projected to reach hundreds of billions of dollars in the next five years, with 80% of AI spending focused on inference. The company’s full-stack approach includes a machine-learning compiler to maximize inference speed on its silicon. Early interest from Anthropic and backing from investors like Thrive Capital, Founders Fund, and the UK’s Sovereign AI Fund signal confidence in Fractile’s ability to challenge Nvidia’s GPU dominance.
What It Means Now: A Shift in AI Hardware
Fractile’s memory-first design represents a paradigm shift in AI hardware, prioritizing memory bandwidth over raw compute. As AI models grow more complex and demand for faster, cheaper inference rises, this approach could redefine the industry. With plans to ship chips to data centers next year, Fractile is poised to disrupt Nvidia’s stronghold, offering a glimpse of what’s changing in the AI hardware landscape. Watch this space—the trend underneath is clear: memory-first architectures are no longer just a concept but a competitive force in AI processing.
Fractile’s first chips are set to be ready for data center deployment next year, marking a critical test of its technology.
Sources
- London’s Fractile raises $720M for memory-first AI chips and hits $7B valuation
- Quantile
- Accelerating Intelligence: Our Investment in Fractile
- With $720M round, Britain's Fractile gets a leg up in the race for AI chips | TechCrunch
- Lightspeed and Alphabet's CapitalG back Fractile in £500M plus funding round
- Fractile Lands $720M to Challenge Nvidia in AI Chip Race
Photo: Anne Nygård / Unsplash (Unsplash License)
