Robot brains could have their “ChatGPT moment” by the end of 2027, according to ACE Robotics Chairman Wang Xiaogang, who says advances in AI models and real-world training data are bringing humanoid robots closer to commercial use. The comparison is deliberate: ChatGPT, launched in November 2022, reached an estimated 100 million users within about two months — the kind of mainstream breakout that embodied intelligence has yet to achieve.

Humanoid robots can already walk, dance, and box, but getting them to perform useful work reliably in the unpredictable physical world remains one of artificial intelligence’s biggest challenges. Wang believes new AI models combined with environmental data capture could soon give robots the intelligence needed to move beyond demonstrations and into commercial deployment, according to a report by Reuters.

“We expect to reach the ‘ChatGPT moment’ for embodied intelligence by the end of next year, driven by world models and environmental data capture,” Wang told Reuters.

Founded in July 2025, ACE Robotics is a Chinese startup developing AI models for humanoid robots. Backed by Ant Group and SenseTime, the company raised more than $100 million in the first half of 2026 and, according to Reuters, plans to pursue an IPO “as early as permitted.” The startup also plans to deploy its technology in 1,000 stores over the next year, while aiming to collect tens of millions of hours of real-world training data within two years — a resource that remains in critically short supply. It is one of several well-funded Chinese players in the field, alongside Unitree, UBTech and Agibot, competing in a sector that Beijing has designated a strategic priority, with the industry ministry setting mass-production targets for humanoid robots.

While large language models such as ChatGPT and DeepSeek have spread rapidly, robots still struggle to perform a wide range of tasks in unfamiliar environments, and a lack of training data remains a major obstacle for the field.

Embodied AI enables robots and other physical agents to perceive their environment, reason about it, and convert decisions into actions through sensors and actuators. World models, meanwhile, help AI understand how the physical world works by learning how objects and environments behave. For robots, that means anticipating what might happen when they move, pick something up, or otherwise interact with their surroundings before taking an action.

“Over the past few years, the entire industry has accumulated data of roughly 100,000 hours, which is far from enough to train embodied foundation models,” Wang said.

Researchers elsewhere are experimenting with similar approaches. In October, researchers unveiled HumanoidExo, a wearable exoskeleton that captures human movements to train humanoid robots.

Other companies are developing their own AI models for robots. In January, Boston Dynamics unveiled the production version of its Atlas humanoid, saying advances in AI helped bring the robot closer to commercial deployment. In June, Alibaba introduced its Qwen-Robot Suite, a set of AI models designed to help robots navigate, perform physical tasks, and simulate real-world environments. Tesla is developing its Optimus humanoid for factory work, and Figure AI has tested its robots on a BMW production line. Wang’s timeline now gives the industry a concrete marker to watch: whether deployments at scale, such as ACE’s planned 1,000 stores, can produce the real-world training data that demonstrations alone cannot.