Europe Builds the Training Ground for Physical AI

Welcome to Viewpoints, news and investment activity covering the next era of computing - brought to you by FOV Ventures.

This Month:

• European robotics scales: Sereact's $110M raise and TUM's RoboGym

• Apple at 50: John Ternus on spatial computing's inevitability

• Portfolio wins: Distance x Kia, Scenario MCP, nunu's William launch

Dave Haynes, Petri Rajahalme, Sointu Karjalainen & David Ripert (FOV Ventures)

The last 18 months have seen European robotics move from research-heavy to infrastructure-heavy. The capital is flowing into the enabling layer - AI models, training facilities, and sovereign data platforms that let robots learn faster and deploy more reliably.

April brought two significant announcements that illustrate this shift. Sereact raised $110M in Series B to fund Cortex 2, an AI model that trains robots on different physical behaviours and picks the one most likely to succeed. The focus is narrower, they don't build robots or sell services, just the AI model that runs across any hardware - single arm, dual arm, humanoid, fixed cell. Meanwhile, Neura Robotics and TUM launched the TUM RoboGym, Europe's largest Physical AI training center, with €17M committed and 2,300 m² of space where humanoid robots will train under real-world conditions from mid-2026. The data feeds Neuraverse, Neura's hardware-agnostic training platform, creating a sovereign data flywheel for European robotics.

We spoke with Dr. Máté Péntek from TUM Venture Labs, who's building this infrastructure, about Europe's advantages in physical AI and where the commercial opportunities are. Read the full conversation here

These are part of a broader pattern where the constraint in robotics is shifting from "can we build capable hardware" to "can we generate enough high-quality training data to make robots work reliably in unstructured environments," this month has also seen NVIDIA continuing to cement its position as the platform layer beneath robotics with new open models, and Japan committing $6.3B to physical AI as a matter of national survival.

We wrote last month about how the infrastructure layer is being built in real-time across geographies, with Europe's advantage is in manufacturing expertise, deep tech and operational environments that generate training data US labs can't easily replicate.

That infrastructure push extends beyond robotics into how we interact with intelligent systems in the physical world more broadly. Apple's 50th birthday this week brought that into sharper focus. John Ternus is stepping in as CEO with a hardware background/ And in his first interview as incoming CEO, he said Apple is still in the "early innings of spatial computing" and that combining the digital and physical world is an "inevitability."

He didn't give a timeline for smart glasses but reiterated Vision Pro as "a peek of the future." With Ternus' hardware background, the question now is which form factor wins and when.

FOV NEWS

Pathways: The Female LP Report 2026

There's a piece missing from the diversity conversation in venture: the LP layer. The people writing cheques to fund managers. They've never been measured.

We wanted to do our part. So we ran the first survey on female LP participation in European venture capital. Six months across 21 countries. Starting with data.

We expected a confidence gap. We found a design gap. 8 in 10 of the women we surveyed say they're ready to invest in venture capital today. Almost none have access to a structure that fits.

Launched live at the Pathways Summit in Helsinki with Tesi and 15 partners.

Europe's Physical AI Playbook: Q&A with Dr. Máté Péntek, TUM Venture Labs

We spoke with Dr. Máté Péntek from TUM Venture Labs about Europe's position in physical AI and robotics. Máté is at the center of the infrastructure buildout happening across European universities - including the TUM RoboGym we mentioned above.

The conversation covers why the built environment is the right anchor for physical AI, where Europe has credible advantages (precision engineering, industrial relationships, infrastructure expertise), how sovereignty and dual-use change the strategic picture, and why moving from research to revenue is the priority now.

FOV Portfolio News

• Distance Technologies announced their collaboration with Kia to integrate Distance's panoramic Lightfield HUD inside the Kia Vision Meta Turismo concept car. Distance's true 3D lightfield places images, graphics, and driving intelligence naturally within the real world at correct depth and distance. The windshield becomes the interface with the visual intelligence layer moving from smartphones and headsets into environments we already inhabit. more here

• Scenario opened their MCP server to public beta, connecting AI agents directly to Scenario's full generation stack - images, video, 3D, audio, hundreds of models, custom trained models, assets, and workflows. Works with Claude Desktop, Claude Code, Cursor, VS Code, and any MCP-compatible environment. more here

• nunu.ai launched William - the first AI agent with a phone. No SDKs, no integrations, no code. William sees the screen, taps, scrolls, types, and completes tasks end-to-end on real devices. Major gaming and enterprise companies are using it in production for QA, compliance, and market research workflows. The public beta is live now. more here

• M-XR launched their image-to-material foundational model. Input a clean 3D mesh and get six PBR maps back. The foundation is M-XR Marso measured data from years of physically capturing real-world materials. That measured PBR data is the moat, separating it from synthetic, scraped, or non-standardized alternatives. more here

• Ray Browser opened unlimited credits. Ray Unlimited lets you describe any app or workflow and builds it with no code. Gmail, Drive, Slack connected in a single prompt. The browser as AI interface layer is underappreciated, and this is the fastest way to understand why it matters. more here

• Flow Computing has been meeting with customers and investors across Silicon Valley. Strong signal: growing interest from both customers and investors, highly engaged technical conversations, and multiple discussions moving forward. Particularly encouraging traction with US-based investors focused on semiconductor and computing technologies. more here

• Spogen evolved from AI-powered machinery assistance into a broader operational knowledge platform for industrial equipment. Through customer deployments, they've seen how resource-intensive it is for industrial teams to build and maintain the structured operational knowledge that modern support and AI systems depend on. The platform now covers the full lifecycle from creation to delivery and continuous improvement. New website reflects this evolution. more here

FOV Job Board

• Roro - UI Programmer

• ai|coustics - Talent Pool: Audio Data Engineering & Machine Learning (Berlin, Germany, Mid-Senior Level)

• Waveline has multiple positions: Growth Intern (United States, Internship) and QA Product Support Specialist (Remote, Entry Level)

• GEEIQ - Client Manager (London, UK, Mid-Senior Level)

• Iconic AI has multiple positions in London, UK: AI Research Scientist/Engineer for Expressive Embodiment, AI Research Scientist for Open-Endedness & Reinforcement Learning, and AI Research Engineer for Model Optimization

• Levtek - Mechanical Engineer (Malmö, Sweden, Senior)

• M-XR - MTS Software Engineer (London, UK, Mid-Senior Level)

View all open roles here