London startup Callosum has landed a $100 million seed round, an unusually large early-stage raise for Europe, to build infrastructure that lets AI workloads move across different chips and models.
Atomico led the investment. Plural, DCVC, and the UK’s Sovereign AI also participated with substantial commitments, joined by several angel backers. The deal follows a $10.25 million raise in February when Callosum emerged from stealth.
The company’s founders, both Cambridge neuroscientists, reject the industry’s prevailing bet on a single monolithic AI model running on uniform hardware. Callosum argues that real-world problems demand varied capabilities, not one dominant system.
Callosum claims its method delivers faster and cheaper solutions for difficult, large-scale problems. A partnership with Cerebras, the US chip manufacturer, backs up that claim. Cerebras CEO Andrew Feldman has praised the startup, as has Kanishka Narayan, the UK’s Minister for AI.
Narayan framed the collaboration as a practical breakthrough. “By integrating Cerebras into Callosum’s platform, we’re building ultra-low-latency inference available exactly where it creates the greatest impact, enabling customers to build AI systems that simply weren’t practical before,” he said.
The team’s stance pushes back against concentration of power among Nvidia, OpenAI, and Anthropic. Callosum positions itself as an alternative to that consolidation, betting that heterogeneous systems will outperform a single scaled-up model.
With fresh capital secured, Callosum now faces the task of proving its architecture can deliver on that promise at scale.
