AWS Physical AI Blog
Built in Europe: What 500 Founders Tell Us About Scaling Physical AI
Research conducted in collaboration with Strand Partners
Physical AI is moving from the lab onto the factory floor, and Europe is well placed to lead that shift. The region combines some of the world’s strongest robotics research institutions with a dense industrial base that urgently needs to automate. European robotics startups raised a record €1.45 billion in 2025, more than double the previous year.
To understand where Europe really stands, AWS worked with advisory firm Strand Partners on Built in Europe: The Rise of Physical AI. The report draws on insights from 500 founders and senior leaders of Physical AI startups across seven European markets, alongside interviews with founders, investors, enterprise leaders, and AWS and Amazon experts. It examines what these companies have built, who is buying from them, and what stands between a working pilot and deployment at scale.
The result is a sector that is further along than its reputation suggests. In this post, we share four findings from the report, spotlight three European startups building on AWS, and look at where infrastructure helps founders move from pilot to production, and where the real blockers remain.
Four findings that stand out
- Europe’s Physical AI startups are already selling. More than half (51%) have their technology commercially deployed, including 14% at scale. Only 12% have no paying customers yet. Demand is close to home: 85% say their first production deployments are, or will be, with European customers, and 61% have more pilot demand than they can currently serve.
- Europe’s strengths line up with what deployment needs. Most founders rate Europe’s access to industrial customers (80%), technical talent (73%) and cost of building and operating (72%) as strong, at or above other leading markets. 78% say being close to European industrial customers helps them reach production faster.
- Capital is the main constraint. 71% say access to capital is harder in Europe, particularly at the manufacturing and scale-up stage, when hardware companies need to invest ahead of revenue. Even so, 77% would choose to headquarter in Europe if they were founding their business today, though the risk of companies relocating to scale remains.
- The road from pilot to production is where startups stall. 68% agree that most European Physical AI startups get stuck at the pilot stage. Long enterprise procurement cycles are the obstacle founders name most often (64%), ahead of integration with customers’ existing systems (48%). Among startups already in production, the most important factor in getting there was a customer willing to co-develop with them.
Startup spotlights
NEURA Robotics (Germany) trains robots and the companies that use them, side by side. Businesses want humanoids faster than they can deploy them, so NEURA created NEURA Gyms: facilities where partners train robots on their own tasks while generating the real-world data Physical AI depends on. Its Neuraverse platform runs on AWS, with Gym training pipelines integrated with Amazon SageMaker AI.
Flexion (Switzerland) builds the AI that lets humanoid robots learn new tasks quickly. Until now, every new robot or task meant a custom engineering project. Founded in Zurich in 2024 by researchers from ETH Zurich and NVIDIA, Flexion trains robots in simulation before they move to the physical world. An AI planner breaks each job into steps, so a robot can complete a long task from a single plain-language instruction. Its model, Reflect, already runs on more than 15 robot designs and can integrate a new one within hours.
PhysicsX (UK) is replacing traditional engineering simulation with AI. Its platform predicts physical behavior in seconds instead of hours or days, so a designer can check aerodynamics or crash safety without handing work to separate specialists, and factory operators can see how changing machine settings will affect output. PhysicsX serves large manufacturers and asset operators across aerospace and defense, semiconductors, industrial machinery, and energy.
How AWS fits in
Infrastructure rarely blocks production, but it does shape how fast founders can build. 73% train their Physical AI wholly or partly in the cloud, 56% say they could not have built their business without cloud computing, and 71% call simulation and synthetic data critical to developing their technology. Only 7% name access to compute or infrastructure as an obstacle to full production, the lowest of any blocker tested.
Founders still point to real constraints: managing multimodal data pipelines (64%), the cost of simulation and training compute at scale (58%), and latency at the edge (57%). That is the gap the Physical AI Toolchain on AWS is built to close. It provides reference architectures and deployment automation for the full lifecycle, from synthetic data generation and training through simulation, validation, and edge deployment, integrated with the NVIDIA Physical AI stack.
But technology alone does not get a pilot into production. As David Randle, Worldwide Head of Go-to-Market for Physical AI at AWS, says in the report:
“Production happens when someone who runs the plant, the warehouse, or the fleet owns the outcome. It also means designing for the second site on day one. A pilot proves something works once. Production proves it’s repeatable.”
Read the full report
Built in Europe: The Rise of Physical AI covers seven chapters of findings, perspectives from 10 European startups, HV Capital, Speedinvest, Audi, and NVIDIA and AWS leaders, and four recommendations for turning Europe’s pilots into production.
Download Built in Europe: The Rise of Physical AI to see the full data and the path from pilot to production, and explore the Physical AI Toolchain on AWS to start building.