Atomico is leading a $100 million seed round for Callosum, one of the largest such investments in Europe, as the company addresses a critical cost problem in the expanding AI economy. Industry data reveals AI companies routinely spend at least half of their revenue on inference, yet face a fragmented hardware landscape with emerging specialized processors.
Callosum acts as an intelligent intermediary, routing AI tasks to the optimal model and chip; the company states the goal is not simply cheaper tokens but more useful intelligence from every unit of compute. The company also launched a family of APIs allowing developers to adopt new AI hardware without application rewrites.
Callosum’s $100M Seed Funds Heterogeneous AI Compute Layer
The APIs abstract away the complexities of increasingly diverse AI processing units, enabling broader adoption of specialized silicon. Callosum’s platform aims to alleviate this burden by intelligently routing AI workloads to the most appropriate model and chip, optimizing for both cost and performance.
In early trials, conducted in partnership with Cerebras on complex financial services workloads, the platform achieved a four-fold increase in speed, a 70 percent reduction in compute costs, and a 10 percent improvement in task success rates, these gains were realized simultaneously, demonstrating that performance improvements do not necessitate trade-offs in cost or accuracy. Beyond Cerebras, Callosum is actively building a network of silicon partners, including Rebellions, Axelera, d-Matrix, Lumai, and Tendrils, as well as infrastructure collaborations with Supermicro and HPE. This strategy positions the company to benefit from the ongoing fragmentation of the chip market, rather than being dependent on a single vendor.
The company’s founders, Danyal Akarca and Jascha Achterberg, previously conducted research at Cambridge exploring how the brain utilizes specialized circuits to achieve intelligence, a principle they are now applying to AI hardware, Callosum says. Their work, encompassing over 70 published papers, informs the core design of Callosum’s system, which seeks to co-evolve intelligence and hardware for optimal performance.
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
