Robotics startups raised $9.9B globally in Q3, Crunchbase data shows

Robotics startups took in $9.9 billion of venture money in the third quarter of this year. That is more than double the $4.3 billion raised in the same quarter of 2025, according to Crunchbase News. Across the first three quarters, funding is up around 185 percent against the same months last year.

Where the money is going

Crunchbase News says China has led the year with $13.2 billion raised, followed by the United States at $12.1 billion. Germany is a distant third at $1.7 billion.

The largest Chinese rounds went to Xpeng Robotics in Guangzhou, which raised $900 million. Xpeng builds AI models and physical robots. Shenzhen's AI² Robotics raised a $736 million Series C to build robots for manufacturing, healthcare and services.

In the United States, Pittsburgh-based Skild AI raised $1.4 billion for its Skild Brain foundation model for robots. Atoms, run by Travis Kalanick in Los Angeles, raised at the billion-dollar level too. Atoms covers ghost kitchens plus robotics divisions in mining and transport. Germany's Neura Robotics raised $1.4 billion for cognitive robots and the data behind them.

The unicorn count has doubled this year

Crunchbase News writes that around 75 robotics companies on its unicorn board have raised money in the past three years. More than half of them, 45, joined the board just this year. Many of the most valuable ones are building the data and the models that other robotics firms will plug into.

The analysis leaves defence-related companies out. That is a real exclusion. Robotics for defence has its own money and its own buyers, and mixing it in would change the picture.

Why the money is moving now

The piece quotes Ulrik Stig Hansen, co-CEO of San Francisco-based Encord. Encord sells the data layer for physical AI. "We're just seeing physical AI take off," he told Crunchbase News at its Long Horizon Physical AI Summit this month.

Hansen's point on the engineering is useful. Physical AI is harder than a language model because there is no internet of physical AI data. The data has to be collected. Sensor readings have to be turned into action instructions a robot can run on. The data sets combine vision, tactile, force and torque. All of it has to be stored, cleaned, enriched and validated. He said physical AI is still in its pre-training phase and that the post-training phase is coming.

The hard part is the data, not the model

Physical AI is now the fastest-growing part of AI spend. For a team building AI products, this is where the next set of engineering jobs is opening up. The hardest problem is not the model. It is the data pipeline that turns raw sensor readings into training examples.

A software team that already runs AI products has useful skills in orchestration, evals and deployment. The missing piece is the sensor side. The gap between a model team and a robotics team is narrower than it looks. The data layer still has to be built either way.

Source

Crunchbase Data Shows Q3 Robotics Startup Funding Jumps 185% As US And China Lead Physical AI Push, Crunchbase News.