Britain’s First Sovereign AI Bet Lands on Callosum’s $100M

Every other seed round is not a national policy statement. This is one, though. UK AI infrastructure startup Callosum has secured a $100 million seed round, one of the largest raises in Europe in early stage fundraising from Atomico and a significant check from the £500 million Sovereign AI Fund from the UK government. Plural and DCVC are also on the list of investors. However, the amount that should be taken note of here is not $100 million. It is the fact that it is the first investment disclosed by the British government in AI ever since its launch in April 2026. The UK government did not just invest in a startup. It placed a bet.(Electronics Weekly)

What Exactly Callosum Does

The core principle behind Callosum’s business is pretty straightforward: most companies perform all operations connected with an AI task using the same costly general-purpose hardware no matter what kind of operation they need to perform. Looking for information in a memory turns into a process as complicated as complex multi-step reasoning performed with the help of the same hardware. It resembles trying to deliver groceries using a sports car.

Callosum’s software splits a complex task into its sub-tasks retrieval, pattern recognition, reasoning, etc. and assigns each of them to the cheapest and most suited to perform this particular task chip. It could be a specialized processor made by Cerebras, Rebellions, Axelera and d-Matrix, or a regular GPU. The approach developed by Callosum is called “heterogeneous compute.” As the company claims, this solution contradicts the AI industry monoculture built on scaling the identical chips.

As Callosum’s website explains, the principle used by the company is inspired by neuroscience. Callosum was created by its co-founders, Danyal Akarca and Jascha Achterberg who worked together during their PhDs at Cambridge University. They noticed that the human brain is not intelligent because of scaling the number of neurons. Instead, different kinds of specialized circuits perform different tasks together. The scientific publications of Callosum’s co-founders appeared in Nature journals and they both have experience in working at Intel and cooperating with Google DeepMind.

The Numbers Behind the Hype

This is not the first milestone Callosum reached. In February 2026, the company secured $10.25 million from Plural and the Advanced Research and Invention Agency (ARIA) of the UK. At this point, the company claimed 2x accuracy gains, 7x speed improvements and 4x reductions in costs on tasks like autonomous computer use, compared to using identical hardware for everything.

For this $100 million round, Callosum has introduced a new and presumably more commercially relevant benchmark. According to it, the system created by the company in collaboration with US chipmaker Cerebras was able to perform complex agentic workloads in financial services four times faster, reduce compute costs by 70% and increase task accuracy by 10% compared to relying on a single frontier model on the regular infrastructure.

Note: Callosum changed benchmark categories from “autonomous computer use” in February to “agentic financial services tasks” in August and neither of its releases provided any technical details on how exactly these numbers were measured. Impressive figures, but the ones that should be checked by investors and enterprise buyers.

Why This Should Be Understood as a Story of UK AI Policy

This is the key element that was missed in most of the coverage. The round is important not because of the product developed by Callosum but because of who underwrites this bet. Choosing Callosum as the first publicly disclosed investment of the UK Sovereign AI Fund is a national strategy signal.

As the UK’s AI minister Kanishka Narayan explained, the success in AI would not be possible just because of having chips but because of using them effectively. Callosum, which aims at making the most of limited fragmented chip supply, seems to be a perfect fit to this vision.

In addition, this round comes at a crucial time for UK bets in AI chips. Chipmaker Fractile is said to be in talks for raising $600 million with the valuation of $6.5 billion and David Silver’s Ineffable Intelligence, another Sovereign AI Fund-backed company, raised $1.1 billion at $5 billion valuation shortly after launching. Both of them have larger valuations than Callosum, but the symbolic meaning of the first investment might turn out to be more important than the number of the dollars involved.

The Business Side: APIs Not Only Research

Besides securing money, Callosum is going commercial. According to the company, it offers a family of developer APIs that allows the use of heterogeneous chip infrastructure without rewriting the application. It is an important point to consider since the main problem of multi-chip usage has always been high engineering costs, not chip shortage. If this promise would prove true in practice, it would significantly reduce the switching cost for enterprises locked in the single-vendor GPU stack.

TheTweaks Verdict

Callosum’s technology bet is legitimate chip fragmentation exists, inference costs crush AI businesses and workload routing makes sense. The more interesting story here is the UK government using Callosum to make a statement in the sovereign AI race. A $100 million seed round would not change anything in Nvidia’s domination in the market. What it will do is to show you where the UK policymakers see the actual power in AI. It is not about owning the biggest model; it is about getting the most of the chips available. Watch whether Callosum’s benchmark claims prove true independently before considering this as the proof of concept rather than ambition.