TL;DR: London-based Callosum has closed a seed round of roughly £78m ($100m), among the largest Europe has seen, with Atomico leading and Plural, DCVC and the UK Sovereign AI Fund joining. The company sells a routing layer that breaks an AI workload into constituent tasks and sends each to whichever model and chip suits it best.
The pitch rests on a shift in where AI money goes. Spending is migrating from training models to serving them, and firms built around AI can find inference consuming half their revenue or more, largely because every task gets pushed through the same general-purpose hardware regardless of what it needs. Meanwhile the silicon underneath is splintering — hyperscalers ship their own designs, and specialised processors from wafer-scale to optical are reaching production.
Callosum’s first product, a set of APIs letting developers reach new hardware without rewriting applications, went live alongside the raise. The company’s headline benchmark comes from its tie-up with US chipmaker Cerebras: on demanding agentic workloads in financial services, it reports four times the speed, 70% off compute costs and a 10-point improvement in task success against one frontier model running on conventional kit. Chip partners include Axelera, d-Matrix and Lumai, plus Rebellions and Tendrils, with HPE and Supermicro on the infrastructure side.
Danyal Akarca and Jascha Achterberg, the founders, met as Cambridge PhD students. Their research looked at why the brain gets its power from an assembly of narrow, purpose-built circuits instead of one enlarged general one. They have since published more than 70 papers between them and assembled a London team inside a year, drawing from DeepMind and Intel as well as Imperial, Oxford and MIT. Callosum is a founding member of ARIA’s Scaling Inference Lab.
The benchmark figures are the company’s own and have not been independently audited, which is worth holding in mind. The more consequential UK detail is the cap table. Resultsense has tracked the £500m Sovereign AI Fund’s investments since launch, and the central question has always been whether state equity attracts private capital or merely substitutes for it. A round of this size led by a tier-one European fund, with the state taking a minority position alongside, is the first substantial evidence for the crowding-in case.
Looking forward: The proceeds go towards scaling the technology and deepening silicon partnerships. Watch for a named enterprise customer outside the Cerebras relationship — until one appears, the performance claims rest on a single hardware pairing rather than the vendor-neutral layer the company describes.