Global venture funding for robotics startups reached $9.9 billion in Q3 2026, more than double the $4.3 billion invested in Q3 2025, according to Crunchbase data. Funding for Q1 through Q3 is up around 185% year over year, marking a run of quarters near or above $10 billion. The concentration of capital in the US and China signals where commercial physical AI capacity is being built.
Where the $9.9 billion went in Q3
China attracted $13.2 billion in robotics funding so far this year, slightly ahead of the $12.1 billion directed to US-based companies. Germany ranks third with $1.7 billion, far behind the two leaders. The largest Chinese rounds went to Guangzhou-based Xpeng Robotics, which raised $900 million for AI models and physical robots, and Shenzhen-based AI² Robotics, which raised a $736 million Series C for intelligent robots for manufacturing, healthcare and services.
In the US, billion-dollar rounds included Los Angeles-based Atoms, founded by Travis Kalanick, which combines a ghost kitchen business with robotics divisions for mining and transport. Pittsburgh-based Skild AI raised $1.4 billion for its robotics foundation model Skild Brain. In Germany, Neura Robotics raised $1.4 billion for cognitive robots with AI and data infrastructure to build intelligence. The analysis excludes defense-related companies.
The funding pattern reflects a shift from software-only AI toward systems that act in the physical world. Ulrik Stig Hansen, co-CEO of San Francisco-based Encord, a provider of the data layer for physical AI, said at the Long Horizon Physical AI Summit in San Francisco that physical AI is taking off. Compute and storage capabilities are being developed to support these systems, yet building for the physical world takes longer than training large language models on internet text. The result is parallel investment in robots, models and infrastructure.
What physical AI means for business
For companies that deploy automation, the funding concentration points to a wider choice of vendors for manufacturing, logistics, heavy industry and services. Chinese and US suppliers are scaling simultaneously, while European options such as Neura Robotics add a third source for cognitive robots. Larger enterprises can pilot competing platforms for different tasks, from factory handling to service operations. Smaller firms may gain access to standardized robots and models without building data infrastructure in-house.
The constraint is data, not just capital. Hansen noted there is no internet of physical AI data, so data must first be collected and converted from sensor tokens to action tokens. Physical AI datasets combine vision, tactile, force and torque inputs and are orders of magnitude larger than text datasets. Storing, curating, enriching and validating that data is described as a very expensive operation. Buyers should therefore check how a vendor collects real-world data, validates it and moves from pre-training to post-training.
A marker to watch is whether the 45 companies that became unicorns this year, out of 75 robotics unicorns funded in the past three years, convert capital into deployed systems over the next year. New large rounds and customer deployments across manufacturing, logistics and household use would confirm that funding translates into operating capacity. Absence of such deployments would suggest capital is waiting for the post-training phase.
