Physicl has raised ten million dollars in seed funding to make training data for physical AI. Its platform recreates difficult situations in 3D so developers can target weaknesses in their models.
Paris-based Physicl secured the funding round led by Inovia Capital, with U.S. Venture Partners also participating. It plans to expand its team and develop its synthetic data platform.
The tool lets researchers recreate situations where an AI model struggles, then change lighting, camera angles and object positions. Developers can generate targeted training data, test a fix and measure whether the model improves.
Physicl founder Alex de Vigan previously built Nfinite, which made 3D product imagery for online retailers. That technology offers a starting point for robot training, where collecting real-world data is expensive and reproducing physical conditions in simulations remains difficult.
Early customer SpAItial, a Munich company building 3D AI models, reportedly cut its room-building model’s error rate by more than half after adding Physicl’s data, according to the company. For builders, that offers an example of targeted synthetic data improving a specific task, rather than evidence that it works equally well across physical AI.
The funding backs a specialist supplier for companies that need physical AI training data rather than general-purpose imagery.
Generating data around a known failure could help developers focus training on weaknesses instead of collecting broad datasets.
SpAItial’s reported result gives builders a concrete use case to assess, but does not establish gains across other models or tasks.
Written automatically by the Made in 10 Newsroom from the source above.
Thailand paused construction on 49 data centre projects in September 2026 as scrutiny of the sector increased. It also suspended approval decisions on 117 more projects amid concerns about regulation and monitoring.
DataVita now plans to supply its proposed Lanarkshire AI data centre entirely from the National Grid. Direct renewable connections are now a future possibility, replacing an earlier pledge of more than 1GW of directly connected renewable capacity.
Mercor, Micro1 and Handshake are buying professional work records to help develop AI models, Wired reports. The push seeks real business workflows rather than simulated tasks, raising questions about employee privacy.