Generative Bionics to build AI data factory for humanoid robots in Italy

Robotics firm Generative Bionics will build a 5,000 sqm (53,820 sq ft) AI data center in Italy to train its humanoid robots for industrial work. The company says the site, called the Physical AI Data Factory, could be partially operational by Q4 2027.

The facility will sit at an undisclosed location in Liguria, the region where the company is based. IT capacity has not been disclosed. The infrastructure is confirmed: AMD will deploy its double-width Helios rack-scale system, a platform designed specifically for AI hardware.

Generative Bionics frames the project as more than a training site. The factory is designed to "move humanoid robotics from laboratory demonstrations to reliable industrial deployment and to advance European humanoid sovereignty, ensuring that the data, models and know-how behind Europe's industrial humanoids are generated, trained and validated on the continent," the company said.

How will the factory actually train robots?

The mechanism has two halves, physical and digital.

First, fleets of the company's first robot, Gene.01, will perform tasks continuously in real-world conditions inside the facility. "These operations will generate high-fidelity physical data for training world action models and validating the complete robot system, including perception, control, safety behaviour and task execution, before deployment at customer sites," the company said.

Second, every physical environment and robot configuration will be recreated as a digital twin. That allows tasks to be repeated across thousands of variations in layout, objects, materials, operating conditions and failure modes. Reproducing those variations physically would be impractical.

The two halves form a loop. Models trained and tested in simulation go back to the robots for validation, and data from physical operations feeds the next training cycle.

CEO Daniele Pucci described the approach plainly: "We are replicating the world, physically and digitally, to train humanoid robots at scale before they ever enter it."

He added: "The physical ai data factory is how we do it: fleets of Gene.01 learn from real industrial work, we stress-test those skills across thousands of digital-twin variations on AMD compute, and we return validated world action models to the field."

Pucci also stressed the European dimension of the build. "With AMD at our side since day one, on board Gene.01 and now at the heart of the data factory, we are building this network here in Europe, so that the data, the models and the infrastructure behind our industrial humanoids are developed and operated on the continent."

What does this change for AMD and for Europe?

For AMD, the deal extends an existing relationship. The chipmaker is already an investor in Generative Bionics and its technology runs on board Gene.01. Now Helios compute sits at the core of the training facility.

Keith Strier, SVP of global AI markets at AMD, said: "Physical AI represents an exciting opportunity to bring the power of AI into the real world, creating new possibilities for automation and industrial innovation."

He called the collaboration "an important step in advancing European AI and industrial sovereignty, enabling the development and deployment of next-generation intelligent robotics applications while strengthening Europe's capacity to innovate."

Who is behind the company?

Generative Bionics was founded to commercialize physical AI research from the Istituto Italiano di Tecnologia. It has raised €70 million ($78m) to date from backers including AMD.

The company is not stopping at one site. It hopes to build a second factory elsewhere in Europe, and a third outside the continent to serve global markets.

The AMD infrastructure is expected to deliver the scalable compute needed for AI model training, processing physical and synthetic data, and running digital-twin simulations.

No customer names or deployment dates for Gene.01 fleets have been announced. The Q4 2027 partial-operations target is the firmest date on the record.