London-based AI infrastructure startup Callosum has raised $100 million in seed financing to develop systems software that orchestrates machine learning workloads across heterogeneous processor environments.
The funding round was led by European venture capital firm Atomico, with participation from Plural and DCVC, alongside a major investment from the UK government's £500 million ($677 million) Sovereign AI Fund. Callosum did not disclose its post-money valuation.
Breaking Homogeneous Compute Dependencies
Most enterprise AI training and inference deployments remain architecturally constrained by homogeneous compute clusters, typically built from uniform banks of Nvidia GPUs. While monolithic hardware setups simplify distributed scheduling and standardized CUDA kernel compilation, they create substantial cost inefficiencies and supply-chain exposure as model architectures diverge.
Founded by University of Cambridge researchers Dr. Danyal Akarca and Jascha Achterberg, Callosum is engineering an orchestration platform that operates at the systems layer rather than the individual model layer. The platform coordinates workload execution across diverse silicon targets:
- Kernel-Level Workload Matching: Automatically decomposes inference and fine-tuning graphs into sub-tasks, mapping specific compute primitives to the most cost-effective and energy-efficient execution target available.
- Heterogeneous Hardware Dispatch: Dynamically schedules tasks across mixed hardware fleets, balancing execution across discrete GPUs, custom ASIC accelerators, and host CPUs without requiring manual kernel rewrites.
- Latency and Cost Profiling: Continuously evaluates hardware utilization, thermal boundaries, and per-token operating costs to route requests dynamically across available cluster nodes.
The startup previously raised $10.25 million in early 2026 before expanding the platform's multi-chip scheduling capabilities.
National Compute Strategy and Capital Allocation
The round marks one of the largest seed commitments in the European AI ecosystem and represents a direct equity deployment by the UK Sovereign AI Fund. As national AI strategies emphasize sovereign compute capacity and resilience against global hardware shortages, software architectures capable of utilizing non-standard accelerators are drawing increased state and institutional capital.
Callosum plans to deploy the capital toward engineering expansion in London, focusing on multi-accelerator compiler optimization and automated hardware dispatch runtimes for enterprise and sovereign data center deployments.
