Callosum Secures $100M Seed Funding to Slash AI Workload Costs with Smart Chip Orchestration

Callosum Closes $100M Seed Funding Round

Cambridge-based AI startup Callosum has secured $100 million in seed funding just six months after emerging from stealth with $10.25 million1

1

. Atomico led the round with participation from Plural, DCVC, and the UK Sovereign AI Fund2

2

. The deal marks a significant milestone for the company founded by Cambridge University neuroscientists Danyal Akarka and Jascha Achterberg, who met while pursuing PhDs and have published in Nature journals with stints at Intel and Google DeepMind1

1

.

First Investment from UK Sovereign AI Fund

Callosum became the first equity investment disclosed by the UK's £500m Sovereign AI Fund in April, though the exact cheque size and equity stake remain undisclosed1

1

. UK minister for AI Kanishka Narayan emphasized that success in AI development depends not just on accessing chips but using them as efficiently as possible3

3

. Technology secretary Liz Kendall described the government as "betting on Britain" when announcing the fund's launch1

1

. The fund also provided six other companies including Cosine, Prima Mente, and Cursive with up to a million GPU hours each on national supercomputing capacity1

1

.

Tailored Inference Platform Optimizes AI Infrastructure

Callosum's cloud service, Tailored Inference, addresses AI workload optimization by distributing tasks across mixed hardware from Nvidia, AMD, Google, and other manufacturers1

1

. The platform breaks down AI tasks into standalone software modules called blocks, then routes each block to the AI model best equipped to handle it2

2

. Simple tasks go to low-cost algorithms while more difficult work runs on frontier models, optimizing both model selection and underlying AI chip usage2

2

.

Performance Gains and Cost Reduction

On complex agentic tasks such as autonomous computer use, Callosum claims to deliver twice the accuracy, seven times the speed, and a quarter of the cost compared to conventional GPU setups1

1

. The company reports that its technology can complete some inference tasks 3.7 times faster than GPT-5.6 Luna with better output quality while reducing infrastructure costs2

2

. This approach challenges the prevailing architecture, with founders betting against the assumption that "one model will rule them all"1

1

.

Strategic Partnership with Cerebras Systems

Cerebras Systems announced its partnership with Callosum alongside the funding round, integrating its WSE series of wafer-size inference accelerators into Tailored Inference2

2

. The chipmaker debuted its WSE-3 Turbo on Wednesday, featuring the same core count as previous processors but delivering twice the performance through an increased clock rate2

2

. Cerebras ships the accelerator as part of the Wafer-Scale Backpack module, with three units forming the backbone of the CS-4 rack appliance capable of running models with 10 trillion parameters2

2

. Callosum has also secured deals with Rebellions Inc and Axelera AI3

3

.

Compute Orchestration as Industry Priority

CEO Danyal Akarca stated that "the next-generation of AI will be defined by how intelligently compute is orchestrated, not simply how much compute is available"2

2

. The company's arrival coincides with inference economics becoming the industry's central problem rather than a footnote, as model training costs concentrate in a handful of companies while the bill for running models sits with everyone1

1

. This shift has put pressure on Nvidia's software moat from an unusual direction, with UK AI startups collectively now valued at $256 billion according to recent industry counts1

1

. The company's neuroscience-inspired framing reflects the founders' argument that intelligence emerges from separate systems coordinating rather than from one very large system scaling further1

1

.