Today we are officially sharing two pieces of good news: ATOMIX has completed its strategic merger and closed a new round of strategic funding.
With this round, ATOMIX's circle of partners has grown a little wider once again.
Zhipu AI, StepFun and SenseTime — joined by Alibaba, which has been walking alongside us for some time — are among the most substantial companies on China's large-model track, and all of them chose to come on board at this moment. Industry partners such as Huaqin and SAIC Hengxu, together with institutions including Jinpu Capital, Yuanhe Puhua, Xingzheng Capital, Suzhou Fund, CMBI, Houxue Fund, Zhengqi Nengke and Yunheng Industrial, also participated in this round, while existing shareholders continued to follow on.
In the second half of embodied intelligence, models and real-world scenarios must be deeply integrated. This merger is also the first time in China's embodied intelligence sector that models and scenarios have truly grown together.
“Models and scenarios have never existed in isolation. This merger brings the two together in a two-way convergence. Scaling embodied foundation models has to move from manual data collection into industrial scenarios, and ATOMIX's massive volume of real-world data is exactly the fuel that sets the flywheel turning. With this loop closed, ATOMIX is among the first to hold a ticket to the large-scale deployment of embodied intelligence. Next, our goal is to build the world's leading general-purpose embodied foundation model.”
— Tang Wenbin, Founder and CEO of ATOMIX
ATOMIX today
Following the merger, ATOMIX has built a complete full-stack footprint spanning models, infrastructure, hardware and application scenarios:
- DM0 · the world's first embodied-native foundation model. Ranked first globally in real-robot evaluation. Trained jointly on three data types — robot multi-sensor perception, autonomous driving and internet data — with manipulation precision beyond the sub-millimeter level.
- Dexbotic · the PyTorch of the embodied intelligence era. The only general-purpose open-source embodied framework in China and the second worldwide, serving dozens of leading universities including Tsinghua, Peking University and Princeton, as well as well-known companies such as Tencent and Beijing Humanoid Robot Innovation Center, and reaching more than a thousand developers.
- RoboChallenge · the world's largest real-robot evaluation platform for embodied intelligence. Co-initiated with Hugging Face, it has served more than 80,000 real-robot tests worldwide and has supported top-tier competitions such as CVPR and ICRA.
- ATOMIX · the largest real-world deployment network in the embodied intelligence sector. Its customers cover nearly 100 global brands in more than 20 countries, and it provides AI-native flexible warehousing solutions to leading companies such as UNIQLO, Coca-Cola and Mixue Bingcheng.
Recent “new products”:
- On June 15, ATOMIX officially launched the world's first three-stage sorting system for warehousing and logistics. With the DM0 embodied foundation model as its core brain, it will orchestrate mixed operations across multiple robot types, bringing the data flywheel into a high-frequency positive cycle.
- This July, the next-generation large model DM0.5, ATOMIX's first general-purpose robot and an entirely new application infrastructure will also be unveiled and officially launched to the world.
Embodied intelligence's picking moment
Looking back at the last wave of large models, the widely recognized dividing line was coding — whoever cracked that atomic task first held the key to unlocking general capability across the entire layer above it. We believe that in the era of embodied intelligence, that atomic task is picking.
ATOMIX ships more than 600,000 items a day and serves nearly 100 brands in more than 20 countries, making it the largest and most real-world source of logistics picking data in China's embodied intelligence sector. After the merger, this barrel of fuel feeds straight into the flywheel. Just as coding did for large models, the real-world data at scale, the verifiable success signals and the cross-task capability transfer that picking delivers will become the engine of the embodied intelligence breakout, setting off exponential growth for the industry.
Whoever gets the data flywheel running on picking first holds the ticket to the second half of embodied intelligence.
Starting from here
The merger is only the starting point. We believe general embodied intelligence is not a distant prophecy — it is walking toward us out of roaring factories and warehouses that never rest. We will accelerate the global rollout of general-purpose embodied foundation models, keep deepening the data flywheel in logistics scenarios, and inject the most advanced algorithms into the most real physical settings, taking robots into warehouses, factories, hospitals, commercial spaces, homes and places we cannot yet imagine, so that they become genuinely intelligent, useful and dependable.