Galaxea AI raises $291M

Galaxea AI has raised another huge round, and the size matters almost as much as the company. The Beijing startup said it closed a Series B+ worth nearly 2 billion yuan, or about $291 million, at a valuation above 20 billion yuan, roughly $2.9 billion. That came only two months after a separate Series B round of nearly 1 billion yuan. In other words, investors just priced this company up at extraordinary speed. It was founded in 2023. Now it is one of the most heavily funded names in China’s embodied AI boom. That phrase, embodied AI, can sound like marketing fog. Here it means software that does not stop at generating text or images. It has to perceive a room, track objects, plan motions, and move a machine through the world without breaking things. Galaxea says it is building that stack end to end, from robot bodies and core modules to data systems and foundation models. Its stated goal is blunt: give robots a general-purpose “brain” for work in factories, warehouses, and service settings. That ambition would be cheaper to pitch than to build if Galaxea were still a lab project. It is not presenting itself that way. The company says the new money will go to mass production of its robots and to more research on its vision-language-action models. It has also tied itself to manufacturing partners. One of the investors in the new round is Lens Technology, a major electronics supplier, and Galaxea says the two will work together on hardware supply chains and large-scale production. That is the tell. This round is not just paying for smarter demos. It is paying for the hard part after the demo. The hardware gives that away too. Galaxea’s best-known machine, the R1 Pro, is a wheeled humanoid platform with dual arms, 26 degrees of freedom, a vertical operating range up to 2 meters, and onboard sensors including cameras and lidar. The design is less about making a robot that looks uncannily human and more about making one that can actually reach shelves, pick items, and move through industrial spaces. Galaxea has also released lighter variants such as the R1 Lite. The company says it plans to begin mass production at the 10,000-unit scale in 2026. That number would be easy to dismiss if the software story were thin. Instead, Galaxea has spent the past few weeks pushing a specific technical claim. In March, researchers tied to the company released Fast-WAM, a world action model aimed at cutting robot inference latency. The central idea is simple enough to matter: a robot may not need to generate imagined future video frames at test time to act effectively. In the paper, Fast-WAM runs at about 190 milliseconds of latency, more than four times faster than slower imagine-then-execute approaches, while staying competitive on benchmark and real-world tasks. For embodied AI, speed is not a benchmark trick. A robot that thinks too slowly is just a hazard with a battery. That helps explain why money is moving into companies like this even as the broader AI market keeps arguing about models. Investors are not funding Galaxea to win a chatbot contest. They are funding a company that claims it can turn foundation models into machines that sort, carry, and operate in places where delays and failure rates are obvious. China’s private and state-linked capital has been pouring into that category, and the rounds are getting larger, not smaller. Galaxea says nearly 20 investors joined this one. A company that did not exist three years ago is now talking about 10,000 robots.