Galbot raises RMB 1.1 billion in a round led by CATL and Puquan Capital, lifting two-year funding above RMB 2.4 billion

Galbot, a Beijing developer of wheeled dual arm robots and the models that drive them, completed a new financing round of RMB 1.1 billion in June 2025, the Zhongguancun Science City administration said in a release published on 24 June. Contemporary Amperex Technology (CATL), the world's largest maker of electric vehicle batteries, and Puquan Capital led the round.

The administration listed the other participants as CDB Science and Technology Innovation, the investment arm of China Development Bank, the Beijing Robotics Industry Fund and GGV Capital. CATL invested through the strategic investment vehicle of its listed company. The release said Galbot had raised more than RMB 2.4 billion in total over two years. It did not give a round letter, a valuation, a dollar amount or the split between new and existing investors, and Galbot did not post separate terms on its corporate website.

The official account framed the deal in political as well as commercial terms. It said embodied AI had been raised to the level of a national strategy, and that bringing in a national investment platform, a Beijing municipal platform and an international venture firm underlined Galbot's position in one of the capital's core industries. It singled out CATL as the round's strategic investor and said its arrival opened room for joint development in applying Galbot's embodied models.

Galbot's pitch rests on training robots largely in simulation. According to the Zhongguancun release, the company launched GraspVLA in January 2025, describing it as the first end to end embodied model pretrained on billion scale synthetic action data and able to generalise zero shot from pretraining alone. The accompanying research paper, written by Galbot, Peking University, the University of Hong Kong and the Beijing Academy of Artificial Intelligence, describes a dataset called SynGrasp-1B with one billion frames covering 10,680 object instances in 240 categories, and reports roughly 90% grasp success in real world tests and results on the LIBERO benchmark above fine tuned versions of the open models pi0 and OpenVLA.

Two further models followed. The release said GroceryVLA was built for commercial retail use, and that TrackVLA, a navigation model, combines purely visual perception with natural language instructions to follow moving targets. The TrackVLA paper, posted on 29 May 2025, reports training on 1.7 million samples, inference at 10 frames per second on a real robot and an 85.1% success rate on single target tracking in the EVT-Bench evaluation.

The commercial case so far is retail. The administration said Galbot's wheeled dual arm robot runs 50 square metre unstaffed shops around the clock, handling 5,000 product types across 6,000 shelf lanes and more than 10,000 boxes of goods, covering stocktaking, restocking, picking and packing. A new shop takes one day to set up. Nearly ten such shops were operating in Beijing at the time of the release, and the company aimed to have them in use at 100 stores nationwide within 2025.

Industry was the second front. On 17 June 2025, a week before the funding release, Galbot announced a joint venture with Boyuan Capital, the investment arm of the Bosch Group, and signed a three party strategic memorandum with Bosch China and Boyuan Capital. Galbot's founder and chief technology officer is Wang He, an assistant professor at Peking University and a scholar of the Beijing Academy of Artificial Intelligence, and the company was founded in 2023.

The hardware is a wheeled machine rather than a walking humanoid. Galbot's current product page for the G1 lists a maximum height of 1,730 mm, a weight of 92.5 kg, 21 degrees of freedom including seven per arm, a payload of 5 kg per arm and 10 kg combined, a top speed of 1.5 metres per second, a 48 V 30 Ah battery rated for eight hours of work, and an NVIDIA AGX Orin 64GB module rated at 275 TOPS. That design trades stair climbing for stability and battery life in shops and factory aisles.

The round placed Galbot among the best funded robot model developers in the first half of 2025. In the United States, Apptronik raised USD 350 million in a Series A in February, and Persona AI raised USD 27 million in May. In France, Wandercraft closed a USD 75 million Series D on 11 June and Genesis AI announced USD 105 million on 30 June. At home Galbot competes with AgiBot, Unitree, the Hong Kong listed UBTech and younger rivals such as Galaxea AI and AI² Robotics, most of which did not disclose round sizes in the same period.

Questions remain. Galbot did not disclose revenue, unit shipments or the economics of its shops, and the 100 store target was an aspiration rather than a contract. The Zhongguancun release is a government account of a private deal, not a filing, and the precise stake CATL took is unknown. Simulation trained grasping has to prove it holds up as tasks move from boxed groceries to deformable parts and assembly.