London-based Callosum has raised $100m in seed funding to develop software that directs artificial intelligence workloads across different models and computer chips, as investors target the infrastructure required to make AI processing more efficient.
The round was led by Atomico, with participation from Plural, DCVC, and the UK’s Sovereign AI fund. Callosum has not disclosed the valuation attached to the investment.
The financing is unusually large for a seed-stage European technology company and reflects the volume of capital moving towards businesses positioned around AI infrastructure rather than end-user applications.
Callosum’s technology addresses a problem created by the diversification of AI computing. Instead of assuming that a workload will run on one type of processor or model, its software allocates tasks across different combinations of hardware and models according to requirements such as cost, performance, and latency.
The company describes the approach as heterogeneous intelligence. Its underlying proposition is that increasingly complex AI systems will depend on multiple types of silicon, making the layer that determines where each task runs more valuable.
Callosum was founded by Cambridge-trained neuroscientists Danyal Akarca and Jascha Achterberg and is headquartered in London. The company had already attracted government backing before the latest financing, becoming the first equity investment made by the UK’s £500m Sovereign AI fund.
That public fund is intended to help British artificial intelligence companies scale domestically rather than becoming entirely dependent on overseas capital as their funding requirements increase.
Compute has emerged as one of the principal constraints on AI growth. Training and running advanced systems requires high-performance processors, large amounts of electricity, data-centre capacity, networking, cooling, and software capable of managing those resources.
Much of the industry has historically centred on general-purpose GPU clusters. The arrival of alternative accelerators and specialised chips is producing a more fragmented infrastructure market in which different hardware can perform certain workloads more efficiently.
Fragmentation creates an orchestration opportunity. Software capable of routing a task to a lower-cost processor without reducing output quality can cut expenditure. Where speed is more important, the same layer can select infrastructure optimised for latency.
The commercial test is whether those decisions remain reliable as models and chip architectures change. AI infrastructure is developing rapidly, and orchestration software must keep adapting as new accelerators, inference techniques, and model designs emerge.
Competition is also substantial. Cloud providers, chip manufacturers, model developers, and independent infrastructure companies are all seeking ways to increase hardware utilisation and reduce the cost of AI workloads.
Callosum’s independence from any single hardware platform could become an advantage if customers increasingly adopt mixed infrastructure. The same independence raises the technical burden because the company must maintain compatibility across a continually changing ecosystem.
The funding arrives as enterprise AI spending shifts from experimentation towards economics. Companies that have demonstrated useful applications are now determining whether those workloads can operate at scale without producing unsustainable compute bills.
Inference — running a trained model to produce outputs — becomes particularly costly when systems serve large populations or perform repeated automated tasks. Small gains in routing and utilisation can therefore materially affect the commercial viability of high-volume products.
The Sovereign AI investment also gives the round an industrial-policy dimension. Britain retains significant AI research capabilities but competes with much larger US markets for funding, infrastructure, and experienced technical staff.
Government-backed capital cannot remove those structural differences, but it can provide domestic companies with another source of growth funding while encouraging private investors to participate alongside the state.
Callosum now has substantial resources to develop its platform during a period when AI computing architecture remains unsettled. Its longer-term position will depend on whether heterogeneous infrastructure becomes widespread enough for orchestration to emerge as a distinct and defensible part of the technology stack.
You must be logged in to post a comment.