Callosum Raises $100M Seed to Route AI Workloads Across Mixed Chips — With the British State on the Cap Table

Callosum, a Cambridge University spinout building chip-agnostic AI infrastructure software, has raised $100 million in a seed round led by Atomico — one of the largest seed rounds ever raised in Europe, and a round that puts the British state directly on the shareholder register.

The round, first reported by Bloomberg and confirmed across multiple outlets on August 20, 2026, includes participation from Plural, DCVC, and the UK’s Sovereign AI Unit, the £500 million government vehicle launched earlier this year. The valuation on the new round has not been disclosed, and neither the size of the state’s cheque nor its equity stake has been made public — a gap that has reportedly already drawn parliamentary attention.

From $10.25M to $100M in six months

The speed of the round is the headline within the headline. Callosum only came out of stealth in February 2026 with $10.25 million led by Plural, the fund founded by former Wise executive Taavet Hinrikus, alongside the government research agency ARIA and angels including Charlie Songhurst, Stan Boland, and John Lazar. Talks about a much larger round were first reported in May, at up to $100 million (~£75 million), and the company has now closed exactly that — a 10x step-up in capital just six months after launching publicly.

For Europe’s venture ecosystem, where seed rounds historically hovered in the single-digit millions, a $100M seed is a statement. It reflects how much capital is now chasing the infrastructure layer behind AI rather than the applications on top of it — and how quickly the economics of running models has displaced the economics of training them as the industry’s central concern.

What Callosum actually sells: orchestration, not silicon

Importantly, Callosum is not building chips. Its product is software that distributes an AI workload across mixed hardware — Nvidia, AMD, Google TPUs, and whatever else happens to be in the rack — routing each part of a task to the chip and model combination that handles it best, instead of assuming a homogeneous grid of GPUs.

The company claims that on complex agentic work, such as autonomous computer use, this approach delivers twice the accuracy, seven times the speed, and roughly a quarter of the cost of a conventional single-vendor GPU setup. Other coverage of the round cites claims of up to 70% cost savings and 4x speedups against general-purpose GPUs. The numbers are vendor-supplied and independent benchmarks remain thin, but the direction of the argument is clear: as inference bills scale with agent deployments that run around the clock, the “one model on one vendor’s silicon” assumption becomes an expensive luxury.

The founders are Danyal Akarca and Jascha Achterberg, who met while doing PhDs at Cambridge and have published in Nature journals, with stints at Intel and Google DeepMind between them. Their thesis is a direct bet against the prevailing architecture. “Big labs are currently betting that one model will rule them all,” one of the founders said at launch. “We think that’s wrong.”

Even the name is the argument. The corpus callosum is the bundle of nerve fibres connecting the brain’s two hemispheres; the founders’ claim is that intelligence emerges from separate specialised systems coordinating with each other, rather than from one very large system scaling further. Whether that holds as an engineering thesis at data-centre scale remains an open question in the field — but it is a considerably cheaper thesis to fund in Britain than building a frontier model from scratch.

Why the state’s involvement matters

The most strategically significant detail is the UK Sovereign AI Unit’s position as a direct equity investor. In April, Callosum became the first equity investment made by the £500 million unit, chaired by James Wise and launched with backing from technology secretary Liz Kendall and chancellor Rachel Reeves. Six other companies announced in the same wave — including Cosine, Prima Mente, Cursive, Doubleword, Twig Bio, and Odyssey — received up to a million GPU hours each on national supercomputing capacity rather than cash. Callosum got an actual stake sold to the state.

That structure turns the UK government from a grant-giver into a shareholder with upside, and it makes Callosum a test case for whether sovereign capital can meaningfully anchor a deep-tech company through the expensive years. Ministers have been unusually explicit about the framing: Kendall’s line at the fund’s launch was that the government is “betting on Britain.” Akarca, for his part, has pointed to the UK’s depth of talent across universities and labs like DeepMind as the reason the company stayed in Cambridge rather than relocating to San Francisco.

The inference-economics opening

Callosum is arriving at a moment when the cost of inference has become the industry’s central problem rather than a footnote. Training frontier models remains a concentrated, capital-intensive game played by a handful of labs — but the bill for running models sits with everyone: enterprises deploying agents, startups with thin margins, and governments building public-sector AI. That is precisely why Nvidia’s software moat has come under pressure from unusual directions this year, and why orchestration layers that promise to route around vendor lock-in are attracting nine-figure cheques.

Nor is Callosum alone on the British side of this trend. Velaura recently raised $110 million for low-power inference chips, and UK AI startups are collectively now valued at around $256 billion by recent industry counts.

Several things about the round remain undisclosed: the valuation, the full investor list, and how much of the $100 million came from the Sovereign AI Unit versus private funds. Companies House filings over the next quarter should settle the ownership question — a slower answer than a press release, but a more reliable one.

What is already clear is the signal: European seed capital can now write frontier-infrastructure-sized cheques, and at least one government believes routing software is a national asset worth owning a piece of.