Why Germany's humanoid billionaire is building robot gyms
Best data wins?
"The decisive competitive factor in intelligent robotics is no longer mechanics, but data."
That is David Reger, founder and CEO of Neura Robotics, and the line cuts straight to the center of what his company is actually building. Not just humanoid robots, not just a foundation model, but physical training facilities: gyms where robots learn to work by watching humans work.
As reported by Forbes, Reger's company, headquartered in Metzingen, Baden-Wurttemberg, is now worth approximately $7 billion following a record $1.4 billion Series C in June 2026. The company is backed by Nvidia, Amazon, Qualcomm, Bosch, Schaeffler, and the European Investment Bank. That makes NEURA Robotics the most valuable humanoid startup in Europe, and one of the most valuable anywhere on the globe.
Reger holds roughly 20 percent of the company, so on paper, that alone makes him a billionaire.
The problem with simulation
Many humanoid robot maker are chasing synthetic data and fighting the sim-to-real transfer. Training in simulation deploying in the real world is cheaper, faster and -- in principale -- almost infinitely scalable.
Reger's argument is that the principle breaks.
Simulations cannot reproduce the physical variability of the actual world: friction coefficients that shift with humidity, warped cardboard on a conveyor, a bolt that wobbles slightly off-spec. You can train a robot to fold a box ten million times in a simulator, but reality still has surprises up its sleeves.
Real-world training data, collected at scale under controlled conditions, is what closes that gap, according to NEURA. So -- like Figure AI with its Index project -- NEURA is building the infrastructure to generate it.
What a robot gym actually looks like
A NEURA Gym is a purpose-built facility where human trainers demonstrate tasks to robots: box folding, component assembly, object manipulation. Tracking technology and embedded simulation environments capture every movement. That data streams into the Neuraverse cloud platform, where it is processed, stored, and made available across the entire robot fleet.
The first is the TUM RoboGym in Munich, built in partnership with the Technical University of Munich's Munich Institute of Robotics and Machine Intelligence. At roughly 25,000 square feet, it is Europe's largest physical AI training center.
Total investment?
17 million euros, with NEURA contributing 11 million euros.
A second facility, NEURA's RWTH Aachen gym, will open by the end of this year. Ten gyms are planned globally across Europe, the U.S., and China, with 5 operational before 2027.
"In RoboGym, robots will learn general capabilities that they can then independently apply to specific tasks," says TUM professor Achim Lilienthal.
In other words, these gyms are not training specific robots for specific clients' specific tasks. They are building a library of general motion primitives that any NEURA-connected robot can draw from.
(Which, of course, brings up the question of post-training.)
The business model: three layers
Hardware is layer one for NEURA.
The flagship 4NE1 cognitive humanoid robot runs 19,999 euros in its Mini configuration and up to 98,000 euros at Gen 3.5: 180 centimeters, 80 kilograms, 25-plus degrees of freedom, 5 km/h walking speed, hot-swappable battery good for up to 8 hours.
Layer two is gym access.
Companies can rent dedicated training cells inside a NEURA Gym, complete with tracking infrastructure and simulation environments, at fees running into the tens of thousands of dollars. The data those sessions generate gets stored in the customer's "brain account" on Neuraverse. Anything trained in the gym can be pushed to their entire robot fleet instantly.
Layer three is the skills marketplace.
What a company teaches their robots, they can sell or lease to other companies. A logistics operator who develops a high-quality palletizing skill can monetize it. A manufacturer who trains a precision insertion routine can license it. NEURA is building toward a tradeable economy of robot capabilities.
This is kind of a genius business model, because it incentivizes customer to invest in robot training ... which benefits NEURA and other customers, but especially those customers that invested in it in the first place.
Hardware is a one-time sale. Data and skills generate recurring revenue. The Neuraverse ecosystem compounds over time: more training sessions, more skills, better robots, more valuable platform.
Neuron AI: the brain behind the gyms
On October 14, NEURA unveiled the Neuron AI model, its multimodal foundation model for physical AI. The architecture integrates 3D vision, touch, audio, robot state, and language. Customers can fine-tune it for their specific tasks.
Reger calls it "the brain of the physical world: Physical AI that turns human expertise and physical experience into capabilities robots can learn and apply."
The gym feeds Neuron AI. Neuron AI improves the robots. Better robots generate better data. Better data feeds better performance, which makes NEURA's robots more saleable. And the loop goes around and around and around ...
What this means for the humanoid race
The humanoid competition looks increasingly winner-take-most.
China has Unitree, AgiBot, and explicit government industrial policy. The U.S. has Figure, Tesla Optimus, Boston Dynamics, and Physical Intelligence. Everyone is racing toward deployment scale.
Germany has Reger.
(Well, and Agile Robots and a bunch of other companies.)
Reger's bet is that durable differentiation does not come from a better actuator or a flashier demo. It comes from a proprietary global training network that generates data no competitor can simply replicate by throwing more compute at a simulator. It is the same structural insight that made large language model training a winner-take-most competition: whoever accumulates the most high-quality training data at scale tends to pull away.
Whether real-world gym data turns out to be the decisive variable, or simulation-based approaches close the gap fast enough, or frontier models increasingly one-shot top-notch robot skills, this is the one of the most interesting structural bets in European robotics right now.
Sources
Written by
I tell stories with data and build startups. I've raised angel and VC funds, led software companies and startups, and chronicled the rise of the mobile economy. Currently I write for Forbes, host several podcasts, and am starting up HumanoidMegaHub. I was TUNE's mobile economist and Singular's VP of Insights, built the research team at VentureBeat, and managed teams creating software for partners like Intel and Disney. In past lives I built websites, e-learning platforms, and mobile apps, and, a long time ago, was named to Folio's top 100 of the media industry's "most innovative entrepreneurs and market shaker-uppers.”