Callosum's $100m seed is a 10x on February, and a UK state fund's first cheque

Callosum, a London startup whose software routes AI workloads across different models and chip architectures, has raised a $100m seed round led by Atomico, with Plural, DCVC and the UK Sovereign AI Fund taking part ^[1][2][3]. In February the same company raised $10.25m in a pre-seed led by Plural, so the cheque size has grown roughly 9.8 times inside a year, which is the part of this story that matters to anyone raising in Europe ^[23][19].

Two cautions before the trend-spotting. The $100m is round size, not valuation: Callosum has not disclosed what it is worth ^[17]. And the February figure is reported inconsistently by TechFundingNews, which describes $10.25m as both the amount raised and the valuation ^[23][24], so the 9.8x is a comparison of rounds, not of price. Total raised since founding is about $110m ^[8][20].

The state involvement is the genuinely new element. Callosum was the first company to receive money from the UK's £500m Sovereign AI Fund, and according to the fund's own website that investment was made in April ^[2][9][10]. The company has also been named in the UK's £1.1bn AI hardware plan ^[18]. Kanishka Narayan, the UK's AI minister, framed it as an efficiency play: "AI is nothing without the chips that underpin it, and the eye-watering demand for them is only going to grow" ^[11].

What the money is buying is a bet on the founders and a thesis. Danyal Akarca and Jascha Achterberg met doing PhDs at Cambridge in neuroscience, computing and AI, have published in Nature journals, and have held positions at Intel and Google DeepMind ^[4][5]. Their platform, which they call "programmable heterogeneity", splits a workload into smaller tasks and assigns each to the model and processor best suited to it ^[6][7]. Callosum claims this is twice as accurate, seven times faster and four times cheaper than uniform hardware on complex tasks, but has not published the benchmark figures ^[25]. Sifted dates the founding to 2024; TechFundingNews says 2025 ^[21][22]. Either way, an investor writing US-scale cheques here is paying for people and partnerships ahead of verifiable performance data.

The partnerships are real and named. Cerebras, which TechFundingNews reports went public in May 2026 at a valuation of nearly $56bn, is now a partner focused on low-latency inference at scale ^[12][13], and South Korean chipmaker Rebellions is also signed up ^[14]. Rebellions chief executive Sunghyun Park put the commercial logic plainly: working with Callosum puts its architecture "into systems alongside hardware chosen for different parts of the workload, instead of asking one chip to do every job" ^[15]. The obstacle is scale: Nvidia's CUDA ecosystem, on roughly 85% of the GPU market, remains the main competitor ^[26], while Etched and SambaNova are attacking inference efficiency with their own silicon ^[27].

The demand case is the strongest thing Callosum has. AI companies frequently spend half or more of revenue on inference ^[16], and Nvidia has said the AI infrastructure market could reach at least $1 trillion by 2027, driven mainly by inference ^[28].

Watch for three things: published, third-party benchmarks against the 2x/7x/4x claims; whether the Sovereign AI Fund's second and third investments follow this template or something more conservative; and whether the Cerebras and Rebellions agreements convert into paying deployments rather than press releases.