A Software Layer, Not a Chip, Just Got Britain's Backing
Callosum, a London startup that decides which AI model and which chip should handle each piece of a workload, confirmed on 20 August 2026 that it had closed a 100 million dollar seed round, one of the largest seed rounds ever raised by a European AI company. Venture firm Atomico led the financing, with Plural and DCVC joining alongside a significant stake from the UK government's own Sovereign AI Fund.
The company was founded in London by two Cambridge PhDs, Danyal Akarca and Jascha Achterberg, and builds what is best described as a traffic-control layer for AI compute: instead of running every task on one large model and one type of chip, Callosum's software splits a workload into its component tasks and sends each one to whichever model and processor handles it most cheaply and effectively.
The Number Sat Hidden For Four Months
Callosum's relationship with the state fund is not new. Back on 16 April 2026, the UK government named Callosum as one of the first companies to receive backing from its 500 million pound Sovereign AI Fund, alongside six other startups working on drug discovery, cheaper supercomputing, and related fields, but that announcement did not disclose how much money changed hands. Founder Danyal Akarca said at the time that the future of compute is heterogeneous and that making that complexity usable is the next frontier, framing the deal as a bet on infrastructure rather than a single flashy model.
Bloomberg's reporting four months later is what finally puts a number on the round: 100 million dollars, disclosed alongside the private investors who actually led it. The fund's own contribution within that total is still not public, which matches a broader pattern in early state-linked AI funding: governments are often quicker to announce a relationship with a startup than to publish what they paid for it.
What The Bet Signals For Buyers Of AI Compute
For a European business watching where sovereign AI money actually goes, the Callosum round is a data point against the assumption that state AI budgets mostly chase chip fabs or headline-grabbing models. The Sovereign AI Fund's support package for its portfolio, which includes equity plus up to a million GPU hours of supercomputer access per company and fast-track visas for specialist hires, is explicitly aimed at the unglamorous layer that determines how efficiently AI compute gets used once it exists.
That has a direct consequence for any company already paying for multiple AI models or cloud GPU contracts: workload-routing software like Callosum's is a bet that the next round of AI cost savings comes from smarter allocation of existing compute, not from buying more of it. Whether Callosum's own product delivers on that promise at scale is unproven outside its early pilots, but the direction of the state's money is now on the record.
Read next: Cisco Quadrupled AI Orders. Margins Paid For It. | OpenAI's Fastest Mode Skips Nvidia's Chips
