The London Startup Solving AI's Biggest Physical Bottleneck

Jan de Vries · 2026-08-24

Let's talk about a problem you might not see until you're knee-deep in it. The next wave of artificial intelligence isn't just about chatbots and image generators. It's about powering the physical world—factories, vehicles, robots, our critical infrastructure. But there's a massive, hidden bottleneck slowing it all down. A London-based startup called Embedd just raised $2.7 million in pre-Seed funding to tackle it. And their story, born from crisis and hardware headaches, is one every tech professional should hear. The round was led by Seedcamp, with a strong group of backers joining in, including Cocoa, Connect Ventures, 2100 Ventures, and several others. It's a vote of confidence in a solution for a problem that's about to get a whole lot bigger. ### The Invisible Friction in Physical AI Here’s the core issue, explained simply. Every smart machine—from an autonomous warehouse robot to a next-gen medical device—relies on dozens of different chips to function. The catch? None of these chips speak the same language. The software that needs to run the machine has to be painstakingly hand-coded to communicate with each one. Michael Lazarenko, Embedd's co-founder and CEO, puts it bluntly. "Every change in hardware creates huge complexity for software teams," he says. "That friction is already massively slowing innovation." His team knows this pain intimately. They're Ukrainian tech entrepreneurs who previously ran a hardware company. They were first hit by global chip shortages during COVID, and then their operations were disrupted by the war. A constant struggle? Every time they had to source a new component, engineers would spend weeks buried in thousands of pages of documentation, rewriting code from scratch. They realized this wasn't just their problem. It was *the* bottleneck for the entire move toward physical AI. ### How Embedd's Digital Twins Cut the Cord So, what's their fix? Embedd uses digital twins and AI agents to automate that grueling integration layer. Think of it like this: - Instead of an engineer manually translating between 50 different chip 'languages,' Embedd creates a perfect digital replica—a twin—of the hardware. - Their AI agents use this twin as a guide and a testing ground. - The system then automatically generates the necessary code, making the chip instantly usable by the software stack above it. It turns a process that can take months into something far more efficient. The company reports enabling customers to deliver production-ready software for new chips up to **six times faster**. That's not a marginal gain; it's a fundamental acceleration. ### The Road Ahead and Why It Matters Since its commercial launch, Embedd hasn't been quiet. They've already signed contracts with multiple semiconductor companies. One notable partner is Microchip Technology, where Embedd is specifically enabling support for the popular Zephyr real-time operating system. This $2.7 million in funding is fuel for expansion. Lazarenko sees the promise of physical AI as enormous, but the path is currently blocked by hardware fragmentation. "This funding enables us to expand our platform," he notes, "and help more semiconductor companies bring their devices into emerging software ecosystems." For professionals watching the intersection of European startups, deep tech, and AI infrastructure, Embedd's story is a compelling case study. It’s about solving a foundational problem that others are just beginning to feel. It's about building the plumbing so the next generation of intelligent machines can actually get built. And sometimes, the most impactful innovations aren't the flashy AI models themselves, but the tools that let them step off the screen and into our world.