Callosum Raises $100M to Rethink How AI Uses Compute

Callosum is challenging one of AI infrastructure’s most entrenched assumptions: that better intelligence requires increasingly powerful versions of the same models running across increasingly large fleets of similar chips. The London company is building a different architecture, where complex workloads are broken into smaller tasks and each one is assigned to the model and hardware best suited to its particular requirements.

That thesis has attracted a $100 million seed round led by Atomico, with significant participation from Plural, DCVC and the UK Sovereign AI Fund, alongside other investors and angels. The financing follows Callosum’s $10.25 million pre-seed announced in February and gives the company substantial capital to expand its platform, compute partnerships and production infrastructure.

Callosum was founded by Danyal Akarca and Jascha Achterberg, whose backgrounds span neuroscience, AI and computing systems. The wider team brings experience across Cambridge, Oxford, MIT, Imperial College London, Microsoft, ETH Zurich and Intel. That mix reflects the company’s core belief that models, algorithms and silicon should be designed as parts of one computing system rather than optimized independently.

Its Tailored Inference platform applies that idea in production. Instead of sending an entire workload to one large model, Callosum can decompose it into specialized tasks and select different models, algorithms and processors based on cost, latency, energy and capability. The company has announced partnerships with Cerebras and Rebellions, while working with additional infrastructure and hardware providers globally.

Callosum says early deployments show why that flexibility matters. In cybersecurity work with HelmGuard, one configuration cut costs by 77 times while another delivered decisions ten times faster. In finance automation with Round Treasury, average document-processing time fell from 15.9 seconds to 2.1 seconds. These are company-reported results.

The larger bet is that AI’s next efficiency breakthrough may come from orchestration rather than scale alone. As more specialized chips and models enter the market, Callosum wants to become the layer that decides which combination should do each piece of work.