London's Callosum lands $100M to make AI models and chips work together

Jan de Vries · 2026-08-20

Here's a funding story that could quietly change how AI gets built and deployed. Callosum, a London-based startup, just raised $100 million in a Seed round. That's a massive number for a seed-stage company, and it signals something big is happening in the AI infrastructure world. The round was led by Atomico, with serious backing from Plural, DCVC, and the UK Sovereign AI Fund. The UK government launched its $652 million Sovereign AI initiative in April 2026 to support British AI startups that matter for both economic growth and national security. Callosum says this money will help it "redefine how humanity computes for AI's next chapter." ### What Callosum actually does Founded in 2025 by Jascha Achterberg and Danyal Akarca, Callosum isn't building another chatbot or model. Instead, it's building the glue that connects different AI models to different chips. The idea is simple: instead of relying on one giant model running on one type of processor, why not orchestrate many models across many types of silicon, routing each task to the hardware that handles it best? That concept, which the company calls "heterogeneous intelligence," could be a game-changer. As AI shifts from training to inference, the question isn't just who builds the best model anymore. It's how you deliver intelligence efficiently across a growing mix of models and chips. Niklas Zennström, CEO of Atomico, put it well: "They're building the software layer that intelligently routes AI workloads to the right compute, making AI faster, more efficient and more accessible without adding complexity." ### Three big moves behind the funding Callosum isn't just talking about its vision. It's shipping. Here are the three major developments it announced alongside the funding: - **Global compute partnerships**: A flagship partnership with Cerebras Systems to bring low-latency, multi-agent AI to customers at scale, plus deals with chipmaker Rebellions and other infrastructure providers worldwide. - **Programmable heterogeneity**: The ability to break AI workloads into smaller compute tasks and route each one to whichever model and chip run it most efficiently, tailored for constraints like cost, energy, and speed. - **Tailored Inference in production**: A family of APIs already delivering performance and cost improvements in cybersecurity and finance. ### Why this matters for the AI race Here's the thing about AI right now: the demand for chips is eye-watering, and it's only going to grow. But having access to those chips isn't enough. You have to use them as efficiently as possible. That's exactly what Callosum is tackling. Andrew Feldman, CEO of Cerebras Systems, said it best: "By integrating Cerebras into Callosum's platform, we're making ultra-low-latency inference available exactly where it creates the greatest impact, enabling customers to build AI systems that simply weren't practical before." Kanishka Narayan, the UK's Minister for Artificial Intelligence, echoed that sentiment. "AI is nothing without the chips that underpin it," he said. "In the race to develop and use AI, success will depend not just on having access to those chips, but on using them as efficiently as possible." ### A quick look at the numbers This Seed round follows a $10.25 million pre-Seed raise from February 2026, led by Plural with support from the UK's Advanced Research + Invention Agency (ARIA). Callosum is also the first-ever investment from the UK Sovereign AI Fund and has been named in the UK government's $1.4 billion AI hardware plan. With a founding team that spans neuroscience and hardware-software co-design, and members from Cambridge, Oxford, MIT, Imperial College London, Microsoft, ETH Zurich, and Intel, Callosum is clearly playing in the big leagues. The question now is whether their orchestration layer can become the standard for how AI gets computed in the years ahead.