Callosum said on August 20, 2026 that it raised a $100 million seed round to build a platform that routes AI workloads across heterogeneous models and chips. Atomico led the financing, with significant participation from Plural, DCVC, and the UK Sovereign AI Fund.

The Cambridge company, founded by scientists Danyal Akarca and Jascha Achterberg, argues that the next bottleneck is not picking a single model. It is matching each task to the combination of models and silicon that meets cost, energy, and latency constraints. Callosum calls that approach programmable heterogeneity and is shipping tailored inference APIs for production use in areas such as cybersecurity and finance.

Partnerships and sovereignty angle

Alongside the round, Callosum announced a flagship partnership with Cerebras for ultra-low-latency multi-agent inference, plus work with next-generation silicon companies including Rebellions and with infrastructure providers and OEMs. The company said it is the first investment of the UK Sovereign AI Fund and is named in the UK’s £1.1 billion AI hardware plan.

UK AI minister Kanishka Narayan framed the thesis around chip efficiency as demand keeps rising. Cerebras chief strategy officer Andy Hock said integrating Cerebras into Callosum’s platform makes low-latency inference available where it has the biggest impact.

Decoded Take

Europe rarely posts seed rounds at this size, which makes Callosum a signal about how governments and VCs want to fund the middle of the stack rather than another chatbot. The bet is that sovereignty and cost control come from orchestration across many chips, not from owning one frontier lab. Watch whether the Cerebras and Rebellions partnerships turn into named customer deployments, and whether Callosum’s APIs become a procurement path for public-sector workloads that cannot standardize on a single US cloud vendor.