BEIJING, Aug. 21, 2026 — Humanoid robots could experience a breakthrough in artificial intelligence comparable to the arrival of ChatGPT by the end of 2027, according to the chief executive of Chinese embodied-AI startup ACE Robotics, as technology companies race to make robots capable of handling unfamiliar real-world tasks.
ACE Robotics CEO Wang Xiaogang said advances in so-called embodied intelligence could soon allow robots to move beyond highly scripted demonstrations and perform more complex jobs in changing environments.
The prediction comes as competition intensifies over what the robotics industry increasingly calls the “brain” of the robot — the AI software that allows a machine to perceive its surroundings, decide what to do and translate those decisions into physical actions.
Wang said the breakthrough could arrive by the end of 2027, although widespread commercial adoption would likely take another four to five years.
ACE is betting heavily on world models, AI systems designed to build an internal representation of the physical environment and predict what may happen after a robot takes an action.
The technology is becoming a major battleground in robotics because hardware has improved more quickly than robots’ ability to operate independently in unpredictable environments.
ACE’s Kairos model is designed to learn from general video, human behavior and interactions involving real robots, allowing the system to predict both future visual states and physical actions. The company’s Kairos-4B model has posted leading results on several embodied-AI benchmarks despite having about 4 billion parameters.
Real-world data becomes the next AI race
One of the biggest obstacles is data.
Large language models were trained using enormous amounts of information already available online. Robotics companies do not have an equivalent supply of high-quality data showing how people manipulate objects and navigate real environments.
ACE plans to tackle that shortage by collecting tens of millions of hours of real-world data over the next two years, including by equipping workers with lightweight sensors that record physical activities that can later be used for robot training.
The strategy reflects a wider shift in the AI industry toward physical AI, where companies are attempting to apply advances in machine learning to machines operating outside computers and data centers.
The race is attracting significant investment.
ACE Robotics, founded in July 2025 and backed by investors including Ant Group and SenseTime, has raised more than $100 million in 2026 and is preparing for a possible initial public offering, according to Reuters.
From demonstrations to businesses
ACE is also attempting to prove that its technology can generate revenue outside research laboratories.
Its systems are being used in areas including retail, hospitality and logistics, with robots carrying out tasks such as store fulfillment and hotel-laundry operations.
The company has said it plans to expand its retail systems to around 1,000 locations within a year and approximately 10,000 locations within two years.
That commercial push comes as humanoid-robot companies face growing pressure to show that increasingly sophisticated machines can perform useful work consistently rather than simply dance, run or demonstrate carefully rehearsed tasks at technology exhibitions.
China races for the robotics lead
ACE is competing with a growing group of Chinese robotics companies, including Unitree and X Square, that are developing their own approaches to embodied intelligence.
The timing of the prediction is particularly notable because Unitree CEO Wang Xingxing made a similar argument at the World Robot Conference in Beijing this week.
Wang said the industry is moving toward a “ChatGPT moment” in which a robot could enter an unfamiliar environment and successfully perform most requested tasks using simple voice or text instructions.
His timeline, however, is considerably less certain. Unitree has said the software breakthrough necessary to reach that level could still be two to 10 years away.
The disagreement illustrates one of the biggest uncertainties facing the robotics sector.
Companies broadly agree that AI software must improve before humanoid robots can become truly general-purpose machines, but there is little agreement over when that threshold will be reached.
China already holds a commanding position in humanoid manufacturing. More than 40,000 humanoid robots were shipped by Chinese companies during the first half of 2026, accounting for about 97% of global shipments, according to data cited by Reuters.
But producing robots at scale and making them intelligent enough to work independently are different challenges.
ACE believes better world models and dramatically larger real-world datasets can help close that gap.
The company is also using both Nvidia processors and chips from Chinese suppliers, a strategy aimed at reducing costs while avoiding dependence on a single hardware source.
Whether the robotics industry reaches its “ChatGPT moment” in 2027 or several years later remains uncertain.
What is becoming clearer is where the next phase of competition is moving.
After years spent building increasingly capable robot bodies, some of the world’s most ambitious robotics companies are now racing to build the intelligence that can make those bodies genuinely useful.
