Physical AI: Why Humanoid Robots Will Fill European Datacenters

Two signals in one week. In Berlin, Jensen Huang made Physical AI and robotics NVIDIA's next big European market. In Milan, Italy's Generative Bionics announced, with AMD, the construction of a physical data factory to train its humanoid robots — a first 5,000 m² hub in Italy, operational by late 2027, followed by two more sites. Behind the robots are datacenters.

The problem these projects solve is little known: language models learned from the internet, but robots have no equivalent. No library describes how to grasp a part, walk on uneven ground or handle a fragile object. This physical data must be produced — by robots performing real tasks, then multiplied through digital-twin simulation.

This is precisely the model of Generative Bionics' Physical AI Data Factory: replicated industrial environments where the GENE.01 robot collects vision, motion and touch data, and an AMD Helios computing infrastructure that extends these experiences through simulation. The company, which raised €70 million from CDP Venture Capital, AMD Ventures, Eni Next and Tether among others, explicitly pursues a sovereignty goal: data, models and know-how generated and validated in Europe.

A New Layer of Compute Demand

Industrial robotics adds a new load profile to the datacenter. Training and digital-twin simulation require dense GPU clusters; operating robot fleets then demands continuous inference, often close to production sites. It is the combination of both — heavy central compute and industrial edge computing — that makes this market structurally interesting.

And the competition is global: China is already multiplying robotic data collection centers, some exceeding 10,000 m². Europe starts with a different asset — its industrial fabric. Automotive, mechanical engineering, logistics: the environments in which robots will learn are precisely those where European industry is installed.

What This Changes for Investors

Physical AI turns a market of demonstrations into an infrastructure market: data factories, simulation capacity, proximity inference. For datacenter investors, this is additional demand that does not depend solely on hyperscalers and language models — a welcome diversification of growth drivers.

Digital sovereignty also finds a concrete outlet here: if robot training data must remain on the continent — for industrial property reasons as much as regulation — then the compute processing it must reside there too. This is an argument for sovereign capacity in Europe, even when the silicon remains American.

How We Apply This at FINXIA

In the TITAN DC AI strategy, we track Physical AI as emerging demand, not yet contracted demand: we do not invest on robot promises, but we select assets capable of serving this customer base — well-connected industrial sites, proximity inference capacity, closeness to manufacturing basins.

Our presence in the French, Spanish and Italian markets takes on particular meaning here: Italy is precisely hosting the first European hub of this kind. Metropolitan datacenters close to industrial zones will be the first candidates to host inference for these robot fleets.

Key Takeaways

Physical AI brings robotics into the infrastructure era: physical data produced in factories, large-scale simulation, distributed inference. For European datacenters, this is a new layer of demand — and a new field for digital sovereignty.

Worth watching: Generative Bionics' Italian site selection and its late-2027 opening, the first industrial humanoid contracts in Europe, and announcements of dedicated simulation capacity by datacenter operators.