AtomBite.AI lands seed round for restaurant-kitchen embodied AI
What's the deal? AtomBite.AIDealroom has a profile for this one. Try Dealroom →, a Chinese embodied intelligence startup founded by former Meituan Waimai tech leaders, has closed a US$1.38M seed round led by Inno Angel FundDealroom has a profile for this one. Try Dealroom →, with participation from the Shuimu Tsinghua Alumni Seed FundDealroom has a profile for this one. Try Dealroom → and individual investors. The funds will go toward developing what the company calls a "world action model" for restaurant kitchens — starting with food delivery packaging and handoff to riders.
Founder and chief executive officer Wang Dong previously ran Meituan's food delivery tech division, overseeing a thousand-person team that supported tens of millions of daily orders. Co-founders Li Tao and Li Haozhe bring algorithm expertise and global commercialisation experience, respectively.
Why now? Despite years of digital transformation in food service — SaaS platforms, mini-programme ordering, dispatch systems — the physical steps between a kitchen finishing an order and a rider picking it up remain stubbornly manual. Packing, sealing, sorting, and handoff still cause high error rates, spills, and waste across the fulfilment chain.
Labour shortages compound the problem. Fast-food wages in North America keep climbing, while restaurants in China face chronic difficulty hiring and retaining staff. After months scouting markets in North America and Singapore, Wang concluded that restaurant back-of-house operations may be one of embodied AI's most bankable use cases — high frequency, clear ROI, and a global pain point.
What could go wrong? The gap between lab demos and a real kitchen running at lunch rush is enormous. Kitchens are hot, messy, and unpredictable — variables that stress even mature robotics. AtomBite's vision-and-touch world action model (VT-WAM) must handle subtle physics: liquid sloshing in cups, shifting centres of gravity, and varying friction from temperature changes.
The company also faces the classic chicken-and-egg problem of embodied AI: it needs real-world deployment data to improve its models, but it needs reliable models to earn deployments. And while it has secured partnership intentions from major domestic and international companies, none have reached scale yet.
The signal: Inno Angel Fund, the round's lead investor, is an established early-stage fund with a track record in Chinese deep-tech bets, and its involvement alongside the Shuimu Tsinghua Alumni Seed Fund underscores growing investor appetite for embodied AI startups that pair domain-specific data advantages with narrow, commercially viable use cases. AtomBite's bet — that the fastest route to a capable world model runs through millions of repetitive, real-world physical tasks rather than general-purpose lab training — mirrors a wider pivot in the sector toward vertical-first deployment, where the messiest environments may generate the most valuable training data.
Read more: 36Kr
