Callosum, a London-based startup that orchestrates AI workloads across different chips and models, said Wednesday it has raised $100 million in a seed round led by Atomico. The funding, which included participation from DCVC, Plural, and the UK Sovereign AI Fund, is one of the largest seed rounds Europe has seen, though the company declined to disclose its valuation.
The raise follows an earlier $10.25 million pre-seed round led by Plural in February, plus a $2.9 million grant from the UK's Advanced Research and Invention Agency. It also arrives during a particularly frothy moment for European AI infrastructure deals. In April, another London company, Ineffable Intelligence, closed Europe's largest seed financing on record: a $1.1 billion round that made even Silicon Valley observers pause.
What Callosum is building sits at the intersection of two trends reshaping enterprise AI. First, the proliferation of specialized chips beyond Nvidia's dominant GPUs. Second, a growing recognition among engineers that no single model handles every task optimally. The company's bet is that intelligently routing different pieces of an AI workload to different processors and models can outperform simply throwing more compute at a monolithic system.
Co-founder and CEO Danyal Akarca, who holds a PhD from the University of Cambridge focused on computational principles of brain network development, framed the approach as a challenge to prevailing assumptions. "What the industry has believed is that the fundamental units of intelligence we should be thinking about are models, parameters, and weights," he said in a DCVC blog post. "What we believe is that that's not inherently true."
The company calls its philosophy "Heterogeneous Intelligence." In practice, that means breaking AI tasks into smaller blocks and routing each piece to whatever chip-and-model combination can handle it most efficiently. According to SiliconANGLE, Callosum announced its Tailored Inference product alongside the funding round, describing it as a cloud service that spans multiple accelerators and can complete certain tasks substantially faster than some competing models while improving output quality. The company has said its workflows run across AWS Trainium chips, Cerebras wafer-scale engines, SambaNova systems, and other accelerators.
Akarca's co-founder, Jascha Achterberg, also earned his doctorate at Cambridge before becoming a Junior Research Fellow at St. John's College, Oxford, where he conducted research with Google DeepMind and Intel. The two have published joint papers on brain-inspired AI architectures, including work exploring how heterogeneous expert models form processing pathways similar to those found in biological brains.
"Everyone assumed chip diversity was a disadvantage to be managed," Achterberg told UKTN in an earlier interview. "We saw the opposite, that it's an advantage to be exploited."
On Wednesday, Callosum also announced a partnership with Cerebras to integrate the chipmaker's wafer-scale inference capabilities into its platform. The collaboration, aimed at what the companies described as "ultra-low-latency, heterogeneous agentic inference," comes as Cerebras pushes deeper into European markets, HPCwire reported.
The UK Sovereign AI Fund made the startup its first equity investment this spring, a move the Department for Science, Innovation and Technology positioned as anchoring next-generation AI infrastructure in Britain. "As the Sovereign AI Fund's first investment, Callosum sets that direction: backing the infrastructure that will shape how AI is computed globally, and anchoring it in the UK," the department said in a press release at the time.
The company incorporated last year and, according to LinkedIn, has been hiring across accelerator systems software, cluster orchestration, and inference engine roles, all based in London. The team size remains modest for now, though that may shift quickly given the fresh capital and the complexity of the technical challenges ahead. Orchestrating workloads across radically different chip architectures is not a problem with obvious solutions, which may be precisely why investors wagered nine figures on two neuroscience PhDs willing to question industry orthodoxy.
