Humanoid Robots Could Have Their “ChatGPT Moment” by 2027, ACE Robotics Says

Humanoid robots could reach a major intelligence breakthrough by the end of 2027, according to ACE Robotics chairman Wang Xiaogang, who believes advances in world models and real-world training data could give embodied AI its own “ChatGPT moment.” The prediction does not mean millions of fully autonomous robots will suddenly enter homes next year. Wang’s argument is narrower and more important: robot software may reach an inflection point where machines can understand unfamiliar environments, reason about physical tasks and transfer what they learn across different bodies and situations far more effectively than today.

ACE Robotics is betting that the next leap in humanoid robots will come from the “brain” rather than another dramatic improvement in walking, dancing or backflips. The Chinese startup says its Kairos world model combines perception, multimodal understanding, physical simulation and action planning, while its data strategy aims to expand robot training from an industry measured in roughly 100,000 hours to datasets measured in tens of millions of hours.

That makes 2027 a potentially important year, but not a guaranteed deadline. Wang himself says broad commercial implementation could still take another four to five years after the breakthrough. Rival executives give different timelines, and real-world deployments in factories, warehouses and homes continue to expose problems with reliability, autonomy, safety and cost.

The better question, therefore, is not whether humanoid robots will become universally useful in 2027. It is whether their intelligence could improve enough that the industry begins to scale the way generative AI did after ChatGPT made advanced language models useful to ordinary people.

Humanoid Robots in 2027: What ACE Robotics Is Actually Predicting

Wang told Reuters at the 2026 World Robot Conference in Beijing that ACE expects embodied intelligence to reach its inflection point by the end of next year. The company’s thesis is that better world models, combined with far more environmental data, will allow robots to predict what happens next in the physical world and choose actions accordingly.

The phrase “ChatGPT moment” is easy to misunderstand. ChatGPT did not invent large language models. What changed was usability and public access. A comparable breakthrough for humanoid robots would mean moving from carefully prepared demonstrations toward machines that can handle a much wider range of useful tasks with less programming and supervision.

Humanoid robots already exist in factories and warehouses. The missing piece is not walking on two legs. It is dependable intelligence.

What Would a Real “ChatGPT Moment” for Humanoid Robots Look Like?

A convincing breakthrough would require several capabilities to arrive together. Humanoid robots would need to understand natural instructions, generalize when objects or environments change, reason about physical consequences, plan multi-step tasks and recover from mistakes without calling a human operator every few minutes.

This is why the intelligence problem is so much harder than a viral demonstration suggests. Software can generate a plausible paragraph in milliseconds and try again if it makes a mistake. A physical robot interacts with gravity, friction, fragile objects and humans. A bad action can damage equipment or injure someone.

The News Ink’s earlier analysis of how humanoid robots are moving beyond science fiction reached the same practical dividing line: the industry has already moved beyond laboratory prototypes, but mass adoption depends on reliability, autonomy, safety and return on investment.

ACE Robotics Is Trying to Solve the Data Problem

The most interesting part of ACE’s 2027 prediction may be its data strategy.

Large language models improved partly because developers could train them on enormous quantities of text, code and other digital information. Robotics does not have an equivalent supply of high-quality physical-interaction data.

Wang estimated that the entire embodied-AI industry has accumulated only around 100,000 hours of useful training data. ACE wants to move several orders of magnitude beyond that by collecting tens of millions of hours within two years.

Instead of relying only on teleoperation, where humans repeatedly control robots through exoskeletons or other interfaces, ACE says it is fitting people on real production lines with lightweight sensors. Human workers can continue doing ordinary jobs while their movements and environmental interactions generate training data.

If that works at scale, it could address one of robotics’ biggest bottlenecks.

Humanoid robots need examples covering different objects, body positions, mistakes, recoveries and interruptions. More data alone will not guarantee intelligence, but the lack of diverse real-world experience is one reason robot models have not followed the same scaling curve as language models.

Kairos Is ACE Robotics’ Bet on the Robot “Brain”

ACE Robotics was founded in July 2025 and is backed by Ant Group and SenseTime. Reuters reports that it raised more than $100 million in the first half of 2026 through several financing rounds. Wang also said the startup intends to pursue an IPO as early as Chinese listing rules permit.

Its flagship technology is Kairos, an open-source world model designed for embodied AI. ACE describes the 4-billion-parameter model as a system that can combine understanding, future-state generation and action prediction across different robotic embodiments.

Public model materials describe Kairos as using general videos, human behavior and real-robot interaction to build representations of how the world changes over time. The aim is not simply to identify an object in an image, but to predict how the scene will evolve if the robot takes a particular action.

Reuters reports that Kairos-4B currently ranks first on public benchmarks used to evaluate world models, ahead of larger systems including Nvidia’s Cosmos 3 and Ant Group’s Lingbot. ACE separately says Kairos achieved leading results across RoboTwin 2.0, LIBERO-Plus, WorldModelBench Robot and DreamGen.

Benchmark leadership is useful evidence, but it should not be treated as proof that a robot can work autonomously in any home or factory. Benchmarks test selected capabilities under defined conditions. The commercial test is whether those abilities survive noise, wear, human unpredictability and long hours of operation.

That is where 2027 will either validate or weaken ACE’s prediction for humanoid robots.

ACE Is Already Targeting Retail, Hotels and Delivery Warehouses

ACE Robotics is not building only a research model.

The company says its embodied AI is being used with humanoid hardware from Chinese manufacturers including Unitree, AgiBot and Fourier in unmanned retail, hotel services and instant-delivery warehouses.

Wang said ACE plans to deploy in at least 1,000 stores over the coming year and scale to 10,000 stores in two years.

That strategy is important because the fastest route to general-purpose robots may not begin with a machine that can do everything.

Retail and logistics provide repeated tasks, structured environments and large amounts of operational data. A robot can learn shelf handling, item transfer, sorting, simple delivery and customer-service routines without confronting the full complexity of a private home.

Factories and warehouses are following the same logic. They give humanoid robots controlled environments in which companies can measure uptime, cycle speed and cost against human labor or conventional automation.

The News Ink has already covered a practical example in aviation through humanoid robot ground-handling trials, illustrating how the technology is entering narrowly defined real-world workflows before becoming a general household product.

The Rest of the Industry Is Moving From Demos to Work

ACE’s forecast arrives during an unusually important week for robotics.

The 2026 World Robot Conference in Beijing brought together more than 300 companies, over 2,000 exhibits and more than 150 new product launches. But the conversation has changed. Investors and customers are asking less about dancing robots and more about useful work, reliability and return on investment.

Unitree had delivered roughly 18,000 bipedal humanoid robots across its product range as of July. Yet founder Wang Xingxing has been more cautious than ACE about the timeline for general intelligence. He has said a true “ChatGPT moment” would involve a robot entering an unfamiliar environment and successfully completing most tasks from natural-language instructions.

Depending on the formulation, Wang has placed that breakthrough anywhere from a few years to much longer away. Another Chinese startup, Galbot, has publicly pointed to the end of 2028.

Those differences show that 2027 is a forecast, not an industry consensus.

There is, however, growing agreement on what the missing capability is. Hardware has improved rapidly. Robots can walk, balance, lift, carry and manipulate objects. The difficult problem is building software that can generalize those skills reliably enough to make the machines economically useful.

Real Deployments Show Both the Progress and the Gap

Current deployments show both progress and the gap.

Agility Robotics says Digit has moved more than 100,000 totes in a live GXO logistics operation. That is meaningful because it measures repeated work in an operating facility rather than a single demonstration.

Boston Dynamics began production of its new electric Atlas in 2026, with deployments committed to Hyundai and Google DeepMind. The company describes Atlas as a production-ready industrial humanoid rather than a research demonstration.

Figure says it has produced more than 350 Figure 03 robots and demonstrated a manufacturing rate of one robot per hour at its BotQ facility. Figure 03 has also been deployed at BMW’s Spartanburg plant for more complex logistics work after its previous generation participated in production involving 30,000 vehicles.

At the consumer end, 1X is taking orders for NEO, with U.S. deliveries beginning in 2026. Yet complex tasks can still be handled through scheduled remote expert supervision. The pattern is clear: humanoid robots are becoming commercial products, but general autonomy remains unfinished.

Why 2027 Could Be Plausible

Several trends make ACE’s forecast plausible even if the deadline remains uncertain.

• Production is scaling: Figure, Unitree and Boston Dynamics are moving beyond handcrafted prototypes, creating larger fleets and more operational data.
• World models are improving: developers are training systems to predict physical outcomes instead of following only fixed trajectories.
• Training data is broadening: internet video, human demonstrations, simulation and real-robot experience can now be combined.
• Edge AI is getting stronger: better chips allow more reasoning to happen on the robot instead of relying constantly on the cloud.
• Commercial deployments create feedback loops: every warehouse or retail deployment produces examples of success, failure and environmental variation.
• Capital is abundant: ACE alone raised more than $100 million in the first half of 2026, while the wider sector is attracting public and private investment.

These forces do not guarantee a humanoid robots breakthrough in 2027, but they make the forecast more credible than it would have looked only a few years ago.

Why the Prediction Could Still Be Too Optimistic

Physical AI has several disadvantages compared with software-only AI.

Each humanoid robot needs motors, batteries, sensors, actuators, cameras, processors and maintenance. Reliability standards are also higher: a chatbot can be corrected after a bad answer, while a warehouse robot that drops a heavy object or repeatedly stops working can erase the economic case for deployment.

Safety, cybersecurity, privacy and liability add further constraints.

Cost is another major barrier. Reuters reported this week that many Chinese humanoid robots still cost roughly 300,000 to 500,000 yuan, making customers sensitive to whether the machines deliver enough productive hours to justify their price.

Homes are even harder than factories.

A factory can standardize floors, lighting, tools and workflows. A home contains pets, children, clutter, stairs, liquids, deformable objects and endless variation. That is one reason a “ChatGPT moment” in robot intelligence could arrive years before a genuinely mass-market home robot.

A 2027 Breakthrough Would Not Mean 2027 Mass Adoption

This is the most important distinction in the entire prediction.

Wang says that even if ACE is right and embodied intelligence reaches an inflection point by late 2027, broad commercial implementation may take another four to five years.

That pushes the larger adoption window toward roughly 2031 or 2032.

The timeline makes sense. ChatGPT itself became an immediate cultural phenomenon because software could be distributed to millions of people through browsers and smartphones. Robots cannot scale that way.

A useful humanoid model would still need to be integrated with hardware, tested for safety, manufactured in large volumes and serviced in the field. In 2027, the real change could be faster task learning, better transfer between robot bodies, less teleoperation and more reliable natural-language control.

Those shifts would be transformative even if most people still do not own a humanoid robot.

The Jobs Question Will Get Louder if ACE Is Right

A breakthrough in humanoid robots would inevitably intensify the debate over employment.

Early deployments are concentrating on repetitive logistics, material handling, manufacturing and selected service tasks. If humanoid robots remain expensive and need frequent supervision, they will mostly complement workers. If one embodied model lets a robot learn dozens of tasks quickly, the economics change and employment concerns become much more immediate.

That is why The News Ink’s broader analysis of whether AI will replace jobs becomes much more concrete when AI gains a physical body. Software automation primarily changes information work. Humanoid robots could eventually bring similar pressures into physical workplaces.

China’s Robotics Ecosystem Adds Momentum

ACE’s forecast also has a geopolitical dimension.

China has become a major manufacturing base for humanoid robots, supported by dense supplier networks for motors, sensors, batteries, chips and precision components.

ACE says it trains its models using Nvidia hardware as well as Chinese AI chips from Rhino Tech and Digua Robotics, a diversification strategy aimed partly at lowering cost and supply risk.

The News Ink’s coverage of the Google-Marvell deal shows how specialized AI hardware is becoming strategically important across the industry. Physical AI extends that race onto the robot itself.

What Would Prove ACE Right by the End of 2027?

The industry should judge the prediction by measurable capability rather than a dramatic robot demonstration.

A convincing humanoid robots milestone would include several signs:

• robots completing unfamiliar multi-step tasks from natural-language instructions;
• significantly less teleoperation and human recovery;
• one model controlling multiple robot designs without extensive retraining;
• reliable work across changing objects and environments;
• measurable reductions in cost per useful task;
• sustained deployment measured in months rather than staged demonstrations;
• safety performance strong enough for continuous operation around people;
• customers expanding fleets because the economics work.

If those changes appear together, the humanoid robots “ChatGPT moment” may be justified.

If robots are still mainly performing carefully rehearsed routines, relying heavily on remote operators or struggling with ordinary environmental variation, the 2027 prediction will have been too aggressive.

Frequently Asked Questions

Who says humanoid robots could have a ChatGPT moment by 2027?

ACE Robotics chairman Wang Xiaogang told Reuters on August 21, 2026 that he expects embodied intelligence to reach a ChatGPT-like inflection point by the end of 2027, driven by world models and larger-scale environmental data.

Does ACE expect humanoid robots in every home in 2027?

No. Wang said broad commercial implementation could take another four to five years even if the intelligence breakthrough occurs by late 2027.

What is ACE Robotics?

ACE Robotics is a Chinese embodied-AI startup founded in July 2025 and backed by Ant Group and SenseTime. It develops the Kairos world model and is deploying its software with robots from multiple manufacturers.

What is the biggest problem holding humanoid robots back?

General-purpose intelligence and high-quality real-world training data remain major bottlenecks. Robots can already perform useful physical tasks, but reliable adaptation to unfamiliar environments is still difficult.

Are humanoid robots already being used commercially?

Yes, but mostly in structured settings and relatively narrow workflows. Digit is operating in logistics, Figure robots are working with BMW, Boston Dynamics is deploying Atlas in industrial settings, and Chinese companies are testing humanoids in retail, factories and services.

The Real 2027 Test Is Intelligence, Not Spectacle

Humanoid robots have spent years winning attention through dancing, running, boxing and carefully choreographed demonstrations. The next phase will be much less glamorous and far more important.

Can a robot understand what a person wants, work out how to do it, adapt when the environment changes and recover when something goes wrong?

ACE Robotics believes the answer for humanoid robots could change dramatically by the end of 2027.

Its case rests on world models, rapidly expanding physical training data and commercial deployments that generate experience outside the laboratory. Kairos is already producing strong benchmark results, while the company plans to push its software into thousands of commercial locations.

There are good reasons for caution. Competitors disagree on the timeline. Home environments remain exceptionally difficult. Manufacturing costs are high. Reliability, safety and maintenance still determine whether a robot creates economic value.

And even ACE does not predict mass deployment immediately after the breakthrough. Wang’s own estimate suggests another four to five years may be required before embodied world models are broadly commercialized.

That is why 2027 should not be treated as the year humanoid robots suddenly take over homes and workplaces.

It may instead be the year the robot brain becomes good enough to change what companies believe is possible.

If that happens, the comparison with ChatGPT will make sense not because robots become universal overnight, but because a technology that has long looked impressive in demonstrations finally becomes useful enough to scale.

Follow The News Ink

Stay connected with The News Ink for business, markets, technology, world news and major economic developments.

Follow us on X, Instagram and Threads, or read our longer-form analysis on Medium.