Inside a record London seed round

Callosum Technologies, a London-based AI infrastructure startup, announced on 20 August 2026 that it had raised 100 million dollars, or roughly 85.4 million euros, in a seed round led by the venture capital firm Atomico. Plural, DCVC and the UK's own Sovereign AI Fund joined the round, which Bloomberg and Atomico both describe as one of the largest seed rounds in European history, arriving less than six months after Callosum's 10.25 million dollar pre-seed round in February 2026.

Callosum was founded in 2025 by Danyal Akarca and Jascha Achterberg, who met while completing PhDs at Cambridge University studying how the brain achieves intelligence through specialized circuits rather than by copying one type of neuron billions of times over. The company's name nods to the corpus callosum, the bundle of nerve fibers that lets the brain's two hemispheres work as one system instead of two separate ones, which is roughly the trick Callosum is trying to pull off with AI chips and models.

What Callosum is actually selling

Callosum's product, called Tailored Inference, is a family of APIs that dynamically routes an AI workload across different chip architectures and different models, choosing whichever combination is cheapest, fastest or most energy efficient for that specific task, rather than running everything through one frontier model on one vendor's GPUs. On complex agentic workloads in financial services, the company says the platform ran four times faster, cut compute costs by 70 percent and improved task success rates by 10 percent compared with the same job run on a single frontier model over conventional GPU infrastructure.

Alongside the funding, Callosum announced new technical partnerships with the chipmakers Cerebras Systems, whose flagship deal targets ultra-low-latency heterogeneous inference, and Rebellions, the South Korean AI-chip startup, plus smaller partners in the Netherlands and the UK and an infrastructure tie-up with Supermicro. Callosum is also a founding member of the Scaling Inference Lab run by ARIA, the UK's Advanced Research and Invention Agency, giving the company a formal seat inside a state-funded research effort on top of its private backers.

A sovereignty bet that skips the chip factory

Servola has tracked a run of moves this year aimed at reducing dependence on any single AI supplier: Japan pooling data that Nvidia itself cannot access, Nvidia, AMD and Qualcomm each separately backing EU AI gigafactories, Brazil deliberately splitting its AI infrastructure bet between Huawei and Nvidia rather than picking one, and five companies prepaying for data centers Mistral has not yet built to lock in non-US compute. Nearly all of that activity is about where compute physically runs or which company manufactures the chip underneath it.

Callosum is a different kind of bet: instead of building a rival to Nvidia's chips or a rival to a US cloud region, it sells a software layer that makes it commercially painless to move a workload off any single chipmaker's stack whenever a cheaper or faster option shows up. That does not require the UK or the EU to out-manufacture Nvidia at all; it only requires making the switch cheap enough that dependence on one vendor stops being permanent by default.

Why a state fund joined a seed round at all

The UK Sovereign AI Fund does not typically write seed-stage checks, and Callosum's round is its first investment of any kind, which is itself a signal about how the UK now defines the word sovereignty in AI policy. Rather than funding a homegrown chip fab, a project that would take a decade and tens of billions of pounds to even approach Nvidia's manufacturing scale, the state fund backed a company that shortens the distance between whatever chip a customer is using today and whatever chip is cheapest or fastest next quarter.

For a European or British business currently locked into one cloud provider's GPU pricing and roadmap, that is a more realistic near-term version of AI sovereignty than waiting for a homegrown chip industry to arrive: it does not remove Nvidia from the stack, but it removes the cost of leaving Nvidia's stack when a better option exists. Procurement teams sizing their next AI infrastructure contract now have a concrete reason to ask a vendor, or a startup like Callosum, exactly how expensive switching away would actually be.