Record-breaking AI funding
London-based AI startup Callosum secures $100 million in one of Europe’s largest seed rounds
Key Points
Callosum claims its heterogeneous compute approach delivers twice the accuracy, seven times the speed, and one-quarter the cost compared with a traditional GPU setup on complex agentic workloads.
Callosum has launched the first software layer for heterogeneous intelligence, providing APIs that let developers adopt frontier AI hardware without rewriting their applications.
Beyond Cerebras, Callosum has partnerships with Rebellions Inc and Axelera AI to build out its hardware ecosystem.
The seed round of $100 million is described as one of the largest seed rounds ever raised in Europe.
Callosum, a London-based AI startup, raised $100 million in a seed round led by European venture capital firm Atomico, marking one of Europe's largest seed rounds ever. The round included significant backing from early-stage investor Plural, US venture firm DCVC, and the UK Sovereign AI Fund—the government's first disclosed investment from its £500 million initiative to boost domestic AI capabilities tech.eu.
The company builds software that automatically routes AI tasks to the most cost-effective combination of models and chips, rather than relying on a single expensive setup. Callosum claims this "heterogeneous intelligence" approach delivers twice the accuracy, seven times faster speeds, and one-quarter the cost compared to traditional GPU-only systems on complex AI workloads sifted.eu.
How Callosum Cuts AI Compute Costs
Most AI companies rely entirely on expensive Nvidia GPUs. Callosum instead distributes work across different hardware makers—Nvidia, AMD, Google, and specialized chip makers—depending on which combination works best for each task thenextweb.com. The company's software automatically matches AI workloads to the right combination of chips and models.
This approach solves a real problem: as AI models grow larger, computing costs and processing time become major bottlenecks. By optimizing the entire compute stack rather than just buying bigger models, Callosum says companies can run AI faster and cheaper without rewriting their applications sifted.eu.
Building Partnerships With Hardware Makers
Callosum has already partnered with specialized AI chip companies including Cerebras Systems, Rebellions Inc, and Axelera AI sifted.eu. These partnerships help Callosum's software work seamlessly with newer, more specialized hardware that traditional software often ignores.
By forging deals across multiple chip makers, Callosum avoids lock-in to any single supplier. This open ecosystem approach appeals to companies tired of depending exclusively on Nvidia for their AI infrastructure.
UK Government Backs Domestic AI Innovation
The UK Sovereign AI Fund's first disclosed investment signals the government's serious push to develop homegrown AI capabilities. The £500 million initiative aims to spur innovation and job growth within Britain's AI sector, rather than leaving the field entirely to US tech giants uktech.news.
Callosum, founded in 2024 by Cambridge PhDs, represents exactly the type of infrastructure startup the government wants to back sifted.eu. The company's funding underscores Britain's bet that controlling the AI compute layer—not just models—is critical to long-term technological independence.
A Bet That Bigger Models Aren't the Whole Answer
The AI industry has obsessed over training ever-larger language models. Callosum argues the real bottleneck lies elsewhere: in how efficiently those models run on actual hardware. The startup's approach challenges the assumption that you need one expensive, centralized infrastructure to run AI well.
With $100 million in seed funding, Callosum now has the resources to build out its software layer and deepen partnerships with hardware makers. If the company's claims hold up in real-world deployments, it could reshape how enterprises think about AI infrastructure spending tech.eu.
Sentiment
Mostly PositivePublishers
20
Articles
3
Reach
23
