Callosum Secures $100M Seed Financing for Chip-Routing AI Platform

London-based AI infrastructure startup Callosum has secured $100 million in seed financing to develop its platform for routing AI tasks across various chip vendors. This is one of the largest seed rounds in Europe, with participation from Atomico, Plural, and DCVC.

The company's founders argue that the current monoculture in enterprise AI, where all workloads are dispatched to Nvidia GPU clusters, is flawed. They claim that complex tasks require heterogeneous compute profiles, which can be achieved by routing subtasks to different chip architectures.

Callosum's platform operates through a two-path software architecture: planning and scheduling for task decomposition and profiling, and fast-path routing for dispatching subtasks to the optimal hardware. The platform currently supports Nvidia GPUs, AMD processors, Amazon Trainium and Inferentia silicon, and specialist chips from Cerebras, Rebellions, Axelera, d-Matrix, and Lumai.

The company's co-founder and CTO, Jascha Achterberg, stated that complex tasks are structurally heterogeneous by nature and cannot be solved with a single model. He explained that different subtasks within a workflow may require entirely different compute profiles, which can be achieved by routing each subtask to its ideal hardware.

The platform has published benchmark results showing compounding gains in performance, cost savings, and task success rates compared to running workloads on a single frontier model on conventional infrastructure. Callosum's founders argue that their approach is more efficient and effective than existing LLM routing tools, which only select the cheapest model capable of meeting a given quality threshold.