Last Updated: 20 August 2026
Callosum, a London startup building software that routes AI tasks to the best combination of models and chips, has raised $100 million in seed funding. The company was founded in 2025 by Danyal Akarca (CEO) and Jascha Achterberg (CTO), two neuroscientists who met during their PhD studies at the University of Cambridge. Their core idea is that AI should not depend on a single type of chip. Instead, different chips with different strengths should work together, an approach Callosum calls "heterogeneous intelligence."
The round arrives as the economics of AI are shifting fast. According to Gartner, global spending on AI inference (the process of actually running trained models to produce outputs) will surpass spending on model training for the first time in 2026, reaching $23.3 billion out of a $42 billion AI infrastructure market. Companies are moving from building models to running them at scale, and the cost of doing so has become a serious problem.
Atomico, the European venture capital firm, led the round. Plural, DCVC, and the UK Sovereign AI Fund also participated. The deal is the first equity investment made by the UK Sovereign AI Fund, a £500 million government venture initiative launched in April 2026 to back British AI companies.
For years, the standard way to scale AI was straightforward: run bigger models on more of the same chip, usually Nvidia GPUs. That worked well during the training era. But now that AI is being deployed in production, powering chatbots, coding tools, financial analysis, and autonomous systems, a growing number of specialised chips are entering the market. Companies like Cerebras, Rebellions, and Axelera build processors designed for specific types of AI work. The missing piece is a software layer that ties them all together. No widely adopted system exists to route different tasks to whichever hardware handles them best.
Callosum says it is delivering improvements in performance and cost efficiency across fields including cybersecurity and finance. The company also points to a broader strategic benefit: if AI can run well across many types of chips, countries are less dependent on any single hardware supplier for access to advanced intelligence.
"AI is nothing without the chips that underpin it - and the eye-watering demand for them is only going to grow. 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."
Kanishka Narayan, Minister for Artificial Intelligence, UK Government
The new capital will go toward scaling the platform and building out partnerships with chip makers and infrastructure providers worldwide. Alongside the funding, Callosum announced a partnership with Cerebras, the US chip company, focused on delivering ultra low latency inference for multi agent AI systems. It also announced partnerships with silicon companies Rebellions, Axelera, d-Matrix, Lumai, and Tendrils, along with infrastructure providers Supermicro and HPE.
On the product side, Callosum is developing what it calls "programmable heterogeneity." In simple terms, the software breaks an AI workload into its component tasks, then assigns each task to the model and chip best suited to run it, based on cost, energy use, and speed. The goal is to make room for new kinds of hardware, including chips that have not been built yet. These capabilities are delivered through a family of APIs called "Tailored Inference," which is now live for customers in cybersecurity and finance.
"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."
Andrew Feldman, CEO, Cerebras Systems
Akarca and Achterberg began working together during their PhDs at Cambridge around 2019. Their research sat at the intersection of the brain, computing, and AI, looking at how the brain produces intelligence not by repeating one type of circuit but by combining many specialised ones. They have published in multiple Nature journals and collaborated with organisations including Google DeepMind and Intel.
Here is what the product actually does. Callosum's software sits between AI applications and the chips that run them. When a company sends a task to the platform, the software splits it into smaller pieces and routes each piece to whichever model and chip combination handles it best. That might mean sending one part of the job to a Cerebras processor for fast inference and another to a conventional GPU. The customer sees better results and lower costs without managing the complexity underneath. The platform works across AWS, Google Cloud, and Microsoft Azure.
The team includes scientists and engineers from Cambridge, Oxford, Imperial College London, MIT, DeepMind, and Intel. Callosum is a founding member of the Scaling Inference Lab at ARIA, the UK's Advanced Research and Invention Agency, and has been named in the UK government's £1.1 billion AI hardware plan.
Atomico led the $100 million round. Plural, the early stage fund co-founded by Wise co-founder Taavet Hinrikus and Ian Hogarth, also participated. Plural had previously led Callosum's $10.25 million round in February 2026, which included angel investors Charlie Songhurst, Stan Boland, and John Lazar. DCVC, a deep tech venture firm, joined this latest round alongside the UK Sovereign AI Fund.
Callosum has also received grant funding from ARIA to support research on integrating new chip technologies into its platform. Across two rounds, the company has now raised over $110 million.
"Working with Callosum puts our architecture into systems alongside hardware chosen for different parts of the workload, instead of asking one chip to do every job. That's the difference between a partnership and a deployment that only works in one environment or geography."
Sunghyun Park, CEO, Rebellions
