Callosum's $100m seed round is the first disclosed investment by Britain's £500m Sovereign AI Fund

Callosum, a London startup founded last year, has raised $100m in a seed round led by Atomico, with participation from Plural, DCVC and the British government's Sovereign AI Fund. The round was announced on Thursday and is among the largest seed rounds ever raised by a UK company.

For the government the significance is narrower and sharper than the headline figure. This is the first disclosed investment from the £500m Sovereign AI Fund, the vehicle set up in April to take equity positions in British AI companies rather than simply hand out grants.

Callosum was founded in 2025 by Danyal Akarca and Jascha Achterberg and emerged from stealth in February this year with $10.25m. It is building what it calls heterogeneous intelligence: infrastructure that matches a given AI task to the model and the chip best suited to it on cost, speed and energy, rather than sending everything to the same frontier model on the same hardware.

At a glance

• $100m seed round led by Atomico, with Plural, DCVC and the UK Sovereign AI Fund participating
• Announced 20 August 2026; roughly €85.4m at prevailing rates
• First disclosed investment from the government's £500m Sovereign AI Fund, launched in April 2026
• Callosum was founded in 2025 by Danyal Akarca and Jascha Achterberg
• The company raised $10.25m in February 2026 when it left stealth
• Among the largest seed rounds ever raised by a UK startup

What the company is actually building

Most AI deployments today route work to a single large model running on a single class of accelerator, usually because that is the path of least resistance rather than because it is efficient. A summarisation job, a classification job and a complex reasoning job all end up on the same expensive hardware.

Callosum's pitch is that this is wasteful in three dimensions at once — cost, latency and energy — and that the routing decision can be automated. Its platform sits between the application and the compute, deciding which model and which chip should handle each request. If that works at scale it is a margin story for anyone running AI in production, which is a large and growing group.

Why the state is on the cap table

The Sovereign AI Fund was set up in April 2026 with £500m to invest directly in British AI firms. Taking equity rather than awarding grants is a deliberate choice: it gives the state upside if a company succeeds, and it signals to private investors that the government is prepared to sit alongside them on ordinary commercial terms.

It also reflects a specific anxiety. Britain has produced world-class AI research and repeatedly watched the resulting companies acquired or relocated. An equity stake does not prevent that, but it does mean the taxpayer participates in the outcome, and it gives officials visibility into a sector they are trying to regulate and buy from at the same time.

The wider chip push

The investment sits inside a broader £1.1bn plan to build a domestic AI hardware sector, announced earlier this year, which also backs chip designers directly. The government has said it will commit up to £100m, subject to due diligence, to buy high-performance AI chips for public supercomputers once British suppliers meet agreed benchmarks — using procurement as a guaranteed first customer rather than a subsidy.

Fractile, the London inference-chip company that raised $220m in a Series B in May, is the most visible beneficiary of that strategy. It has reportedly been seeking roughly $600m more at a $6.5bn pre-money valuation, a figure that would have looked implausible for a British chip startup two years ago.

What could go wrong

Seed rounds of this size compress the usual timeline. A company founded last year now has to build an enterprise-grade platform, hire against Silicon Valley salaries and prove its routing actually saves money in production, all while investors who paid a seed-stage price expect Series A progress on a short clock.

There is a competitive risk too. The large cloud providers have every incentive to do model and hardware routing themselves, inside their own stacks, and they own the customer relationship. Callosum's opportunity depends on enough buyers wanting a neutral layer that spans providers — a bet on the market staying plural rather than consolidating.

What happens next

The fund's next disclosures will show whether Callosum is the template or the exception. An infrastructure company with an unusually large seed round is a conspicuous first pick for a state investor, and the pattern of subsequent investments will say more about the strategy than this one deal does.

For Callosum the immediate task is unglamorous: turn a research idea about matching workloads to silicon into software that enterprises will put in front of production traffic. The money buys time to do that. It does not, by itself, prove the thesis.