$27.6B in robots. Now the hard part.

Last month a German robotics company raised €1.2 billion in a single round. NEURA Robotics is now worth about $7 billion, backed by Nvidia, Amazon, Qualcomm and the European Investment Bank. Capital stopped being the bottleneck in physical AI. The right kind of capital became it.

Deep Dive: Why physical AI needs a different kind of capital

Why physical AI can’t be funded like software

Robotics and physical AI startups raised $27.6 billion in 2025, more than double the year before. In Silicon Valley, AI took 93% of every venture dollar. The money has arrived. Whether it can actually build robots is a separate question.

Software AI rewards speed. Ship, measure, ship again. Robotics runs on different physics. A humanoid that climbs a flight of stairs in a demo is a long way from a fleet that runs a warehouse for a year with nobody stepping in.

The autonomous vehicle sector already ran this experiment. More than $100 billion went in. A decade later, a handful of US cities have driverless commercial service. Capital alone did not collapse the timeline, and it rarely does when atoms are involved.

When generalist money meets hardware, three things tend to go wrong. Valuations drift away from technical milestones. Founders skip the unglamorous engineering to chase a moonshot demo. And early cap tables fill with story investors who go quiet the moment an actuator redesign eats six months. You can see the incentive in the pitch language: too many early-stage teams now introduce themselves as “the OpenAI of robotics” instead of explaining what they have actually built.

Specialist capital behaves differently because it has to. It understands the physics, the supply chain and the competitive map. It can sit still through a twelve-month detour into a new actuator design, because it has lived through that detour before. As Lancaster and Paseman put it, niche investors have an edge precisely because they are not trying to be everything to everyone.

That is the thesis we have been running since 2022, across 33 companies at the convergence of AI, robotics and spatial computing. None of them is a magazine cover. All of them have to work on day one. Four are below.

Read David Ripert’s full essay, Why Physical AI Needs Specialist Capital 

In the Field

Four companies from the essay above, each building physical AI that has to work in the real world, not just in a demo.

Makiina — The general factory for physical intelligence
Raw material in, different form factors of physical agents out. Makiina designs a shared manufacturing and software back end from day one, with common actuation, electronics, perception and policy layers spanning multiple robotic form factors. This month it showed its arms operating in real time and fully autonomously.

Why it matters: One back end across many form factors compresses the cost of building physical agents faster than any single-robot roadmap in the press.

🎥 Watch the arms demo below where robots are assembling their own actuators.

Levtek — Robot working companions for physical labour
Levtek’s Levkart, a ride-on autonomous robot, is in live logistics trials with PostNord, driven in part by vision-language-action models. An operator can redirect it by voice in a space it has never seen. No facility retrofit required.

Why it matters: Levtek fits the world as it is. No QR floors, no rebuilt aisles, which means a sales cycle that closes in months, not years.

Distance Technologies — Deviceless mixed reality for vehicles and cockpits
Distance signed Galvion, integrating its AR computational optics and Acuity OS into Galvion’s head systems, with market-ready products targeted for 2027. It also joined Project Orbion alongside Niantic Spatial, ICEYE and BlackSky, deployed for the first time at the US Coast Guard Aviation Training Center.

Why it matters: Defence and aerospace buyers move on multi-year evaluation cycles. That is the patient, expert-capital story in practice. Spatial computing reaches scale by upgrading the surfaces people already use, not by selling them new headsets.

🎥 Watch the Eurosatory demo below.

Spogen — Voice and vision AI assistants for heavy machinery
Spogen’s assistants are now running in tractors and seed drills through the EIT Food Test Farms programme, with Valtra, Väderstad, Lännen Tractors and NHK Group. Construction, forestry, mining and ports are next.

Why it matters: Heavy machinery is fragmented, slow to sell into, and gets no hype. It is exactly the kind of physical-AI problem generalist capital walks past and specialist capital underwrites.

🎥 Watch Spogen's Talking Tractor (AGCO Corporation) demo below.
📖 Read → Natalia Leinonen on machine knowledge beyond the manual.

Radar

Beyond our portfolio, here’s what caught our attention this month.

• NEURA Robotics raised €1.2 billion. The number to watch now is the deployment count, not the valuation. Europe’s largest robotics round of the year, backed by Nvidia, Amazon, Qualcomm and the EIB at roughly $7B. The Deep Dive above in one headline: the money is here, the milestones have to catch up. (CNBC)

• Europe’s industrial robots are pulling growth money now, not just seed. THEKER raised a €73M Series A out of Barcelona, led by CRV with Samsung and LVMH, to put AI robots on tier-one factory floors. The deployable-robot thesis, funded at scale. (EU-Startups)

• Jeff Bezos is now a physical-AI founder. His startup Prometheus raised $12B at a $41B valuation to build an “artificial general engineer” — AI that designs and manufactures physical systems, from jet engines to drugs. When the second-richest person alive bets his next act on atoms, this stops being a contrarian thesis.

• The world-model money is here. The benchmark below says the tech isn't yet. Odyssey, built by autonomy veterans Oliver Cameron (ex-Cruise) and Jeff Hawke (ex-Wayve), raised a $310M Series B at a $1.45B valuation, led by Natural Capital with Amazon, AMD and GV, to build world models that learn physics from the real world — for game worlds and robotics. Capital is racing into the exact frontier the benchmark below just called brittle. The edge goes to whoever lands these models on real hardware first. (TechCrunch)

• When DeepMind opens its first robotics accelerator, it opens it in London. Google DeepMind’s Accelerator: Robotics kicked off with a 15-startup cohort, several of them European. The platforms and the talent are converging on this continent. (PPC Land)

• The week the US cut Europe off from the best AI models was the strongest case for tech sovereignty yet. In June, Washington forced Anthropic to block every non-US user from its most capable models, overnight. Europe runs on AI it does not control, and a dependency that can be switched off in a day is not a foundation. The continent needs its own stack. (Al Jazeera)

• The billion-dollar world-models bet just met its first hard benchmark, and the models came back brittle. A new evaluation of Yann LeCun’s AMI research direction found today’s world models still fragile. We think that’s healthy. The category needs benchmarks more than it needs bigger seed rounds. (TechTimes)

One Thing to Watch

The teams turning world models into something a robot can actually use are the ones to watch.

Yann LeCun’s $1B raise put world models back on the front page, but the benchmark that followed (see Radar) showed today’s models are still brittle in the real world. The work we find most interesting in Europe is narrower: not a model that reasons about everything, but one compressed enough to give a specific robot a reliable guess at what happens next, on real hardware, in one environment.

We’ve taken a handful of meetings, and the signal is getting louder. If you’re shrinking a world model down to something that runs on a machine that has to work on Monday, we want to hear from you.

Hiring across the portfolio

Our companies are building, which means they are hiring. A sample of open roles this month:

• M-XR — Applied Research & Machine Learning (London)

• ai|coustics — Systems Software Engineer, Rust/ML (Berlin)

• Flow Computing — Compiler & CPU-modeling engineers (Oulu)

• Levtek — Mechanical Engineer (Malmö)

• Iconic AI — AI Research Engineer (London)

• Scenario — Full-stack ML Engineer (Paris)

• Starstuff — AI/ML Engineer (Helsinki)

Each company name links straight to its careers page — more roles listed there.

Save the date: Minus One, the day before Slush Investor and Founder Days

Google returns as a partner for the fourth year running.

On Monday 16 November, the day before Slush opens its doors, we host Minus One in Helsinki. Three editions in, it has become one of the sharpest rooms of the week: 300+ investors and operators, north of €60B in AUM, gathered before the noise starts.

It is invite-only, and the first invites go out this week. If you are on the list, watch your inbox. And book your flights accordingly — for those travelling into the city, you’ll need the earlier flight on Monday.

Your turn

We are mapping the European founders building physical AI in markets nobody calls exciting: heavy machinery, inspection, agriculture, logistics. Building one? Know someone who is? Hit reply, we read every response.

— The FOV Team

Viewpoints is brought to you by FOV Ventures, the leading European fund investing in the next era of computing.

LinkedIn | Website | Pitch us