Embedd raises £2M to be the software glue for physical AI

Embedd, a London startup, has raised £2M — about $2.7M — in a pre-seed round led by Seedcamp, to solve an unglamorous problem that sits underneath every robot, drone, and smart device: getting the software to talk to the chips. It is a small round for a bottleneck that gets bigger as “physical AI” scales.

The problem is real and largely invisible. Any intelligent machine runs software on top of dozens of different chips — processors, sensors, radios — and none of them speak a common language. Today, engineers write the glue by hand: board support packages, device trees, drivers, painstakingly re-done for each new chip and each operating system. It is slow, repetitive work, and it is exactly the kind of thing that stands between a hardware idea and a shipping product. Embedd’s platform models each chip as a digital twin and uses AI agents to generate that integration code automatically, which the company says lets manufacturers ship production-ready chip software up to six times faster. Since launching commercially in April, it has signed contracts with several semiconductor firms, including Microchip Technology.

The founders came to the problem the hard way. Michael Lazarenko, Maxim Gorinov, and Valentin Gololobov, all Ukrainian, were running a hardware company when the Covid-era chip shortage forced them to swap in whatever parts they could find and rewrite their software each time — and Russia’s invasion of Ukraine disrupted their supply chain again. That experience convinced them the integration problem would only worsen as more machines become intelligent. Seedcamp led the round, joined by Cocoa, Connect Ventures, and a long list of others.

It is worth reading alongside the UK’s growing cluster at the hardware layer of AI. Where Callosum works on matching AI workloads to the right chip, Embedd works one layer up, on making any chip usable by software in the first place. Both are bets on the same unfashionable truth: a great deal of the value in AI is not in the models at all, but in the plumbing that makes them run on real hardware. For a country with genuine ambitions in robotics and embedded systems, more of that plumbing being built — and funded — in London is a quietly encouraging sign.