China's robot makers have mastered the moves. Now comes the race for brains. Published Nov 01, 2026, 06:09 PM Updated Nov 01, 2026, 08:24 PM (C) Reuters. By Laurie Chen and Eduardo Baptista BEIJING, Nov 2 (Reuters) - After years of building humanoid robots capable of running, dancing and performing backflips, Chinese companies are pouring resources into a different challenge: developing the AI that could make those machines useful across a wider range of jobs. The shift is fuelling an investment boom in embodied AI, a field focused on software that allows robots to understand their surroundings, learn unfamiliar tasks and adapt without having to be programmed for every job. Chinese embodied AI companies raised 93.5 billion yuan ($13 billion) across 322 deals in the first half of 2026, five times the amount a year earlier, according to data research firm ITjuzi. Funding is increasingly reaching specialist developers of robot intelligence alongside manufacturers. As competition drives down hardware prices, robot makers have "concluded that the margin won't come from the hardware but from the brain," said Dermot McGrath, founder of Shanghai-based technology consultancy ZenGen Labs. The emerging sector is opening a new front in US-China tech rivalry. American companies lead in some advanced robot AI systems, while Chinese developers hope their manufacturing strength and access to real-world training data will help them challenge their US competitors, according to Reuters interviews with founders and top executives at five Chinese robotics firms, as well as industry analysts. The country that best combines AI with machines could gain an advantage in automating factories and logistics, potentially reshaping industrial productivity and the global market for robotics technology. -9.02 (-4.22%) Closed · 30/10 · USD +2.52 (+1.23%) After Hours · 18:55:47 "The brain is indeed the weakest link in the complete robotics stack," said Spirit AI co-founder Gao Yang. RACE FOR INTELLIGENCE American startups, including Physical Intelligence, Skild AI and Generalist, are developing software intended to teach robots different skills without extensive retraining. Generalist says its GEN-1.5 model completed around 60% of simple tasks after viewing a demonstration lasting between 3 and 12 seconds. China's X Square has published research showing robots acquiring handling skills from a single human video without retraining the model for each task. "The leading American labs are setting a demanding benchmark, particularly in how quickly robots can acquire and reuse skills," said robotics consultant Georg Stieler, who considers China's strongest startups credible competitors. While US developers benefit from advanced AI research and greater access to powerful computing chips, Chinese companies can work closely with robot manufacturers and industrial customers to test and improve their systems. Specialist Chinese startups, including X Square, Galaxea and Spirit AI, have each reached valuations of roughly $3 billion, according to company disclosures. "We want robots to shift from being driven by code to being driven by common sense," said X Square co-founder and COO Yang Qian. Established manufacturers are investing too. Hong Kong-listed UBTech advertised a chief scientist position overseeing AI model development with compensation of up to 124 million yuan ($18 million), including equity and benefits. Hangzhou-based Unitree, one of the world's largest humanoid robot makers by shipments, has about 30 researchers developing AI that predicts how objects behave, according to a person briefed by the company. Unitree did not respond to a request for comment. THE TRAINING RACE Unlike chatbots, which learn from vast quantities of internet text, robots need examples of physical interactions, from gripping objects to handling flexible wires. Chinese companies are pursuing different approaches to collecting that information, with some employing people to demonstrate tasks, others using computer simulations, and some combining both. The field remains crowded, with no company establishing itself as the clear leader. Spirit AI has hired around 1,000 people to record everyday and industrial activities, Gao said. ACE Robotics uses wearable sensors to capture workers' movements and grip forces in stores and hotels. Ant Group's robotics unit Robbyant said it trained a model using 160,000 hours of material, including 50,000 hours of human-guided robot movements. It also trained software on 20 robot configurations from 17 brands, aiming to use the same AI across different machines. DexForce is taking another approach, using computer-generated environments to teach robots. Chief operating officer Nicole Yang said the company can prepare some industrial tasks for deployment within 6 to 8 hours and reported success rates above 99.9% in specified precision operations. But simulations cannot perfectly reproduce physical interactions, said Jerry Wang of spatial intelligence company Manycore Tech, which designs simulated environments for AI training. "Data quality is more important than data diversity, and more important than simply collecting data at massive scale," X Square's Yang said. Chinese robot firms also face additional headwinds, including shortages of advanced Nvidia chips due to US export controls, and a squeeze on IPO funding amid regulatory concerns over inflated valuations and overcrowding. As robots become more powerful, the shortage of advanced chips will become a serious problem for China's embodied AI industry, said Zhu Mingxuan, who leads development of JoyIn Technology's Aether robot AI. FROM TRAINING TO WORK JoyIn and other developers are already testing whether their software can enable robots to tackle unfamiliar problems. Zhu described an experiment in which a robot repeatedly failed to place headphones in a cupboard beyond its reach. Then engineers showed it videos of how people managed the task, and the robot pushed a box towards the cupboard and tried stepping onto it, Zhu said. "Every time it was exploring a different behaviour," she said. Real-world deployments in China leave no room for improvisation. Battery maker CATL said Spirit AI robots connect flexible wires during battery-pack testing at its Zhongzhou plant, reporting insertion success above 99%. Robbyant said its robots handle orders in two Shanghai pharmacies, with staff checking medicines before dispatch. But the longer-term prize is software capable of learning new jobs without extensive adaptation. Success will depend how the same AI performs across different customers and workplaces and that will require companies to train increasingly powerful models with advanced chips. Chinese companies are trying to reduce their computing requirements through more efficient models, synthetic data and simulations, while domestic chipmakers, including Huawei, are developing alternatives to Nvidia hardware. It remains unclear how far those efforts can offset the shortage. 'CHATGPT MOMENT' Unitree founder Wang Xingxing said last month the company had tested OpenAI's GPT-6 on robots with encouraging results. He said GPT-7, if released next year, could bring robotics its own "ChatGPT moment" - a step-change in capability akin to ChatGPT's 2022 launch, which turned generative AI into a mainstream consumer and business technology. Wang said he would consider that threshold reached when robots were able to complete 80% of tasks in 80% of unfamiliar environments from spoken instructions. But consultant Stieler said a breakthrough in demonstrations would not necessarily mean robots were ready for broad commercial use. "If the question is when a model first handles a broad range of unfamiliar tasks in demonstrations, the optimists may well be right," he said. "If the question is when a customer can count on it, I would add years, not months." See the trade on NVDA, but can't pull the trigger? Most traders can read a chart. The hard part is the moment: entry window open, pattern forming, and you're still waiting for more confirmation. That's the conviction gap - and its chart analysis closes it. Unlike other AIs that just read data, its Vision AI literally "sees" your charts and hands you a complete trading plan: entry, stop-loss, and profit target in under 60 seconds. Know exactly what to do next, every time.

Physical Intelligence's US$20 billion price tag has no outside bidder. Physical Intelligence is reportedly in talks to raise over US$1 billion at a US$20 billion valuation, nearly double its price from earlier this year, with existing investors Avenir and Lux Capital leading the round rather than a new outside backer. The company still has no revenue, and its October API beta names no paying customer. Physical Intelligence is reportedly in talks to raise more than US$1 billion at a valuation near US$20 billion, almost double the roughly US$11.5 billion figure it carried earlier this year. The company still has no revenue. The investors said to be leading the new round, Avenir and Lux Capital, already own a slice of it. That last detail is the story. A funding round is supposed to be the moment an outside party, someone with no existing stake and no reason to be generous, looks at a company and decides what it is worth. When the people setting the price already hold the asset, the number stops being a market signal and starts being an accounting exercise. The valuation ladder, rung by rung. Physical Intelligence's price history reads like a company doubling its own homework every few months. A US$70 million seed in 2024 valued it at roughly US$400 million. By that November, backers including Jeff Bezos, OpenAI, Thrive Capital and Lux Capital put in US$400 million at a valuation reported between US$2 billion and US$2.8 billion. A year later, in November 2025, Alphabet's CapitalG led a US$600 million round at US$5.6 billion, with Bezos back for a third time. Earlier this year the company raised again at a valuation of roughly US$11.5 billion. The figure now on the table is close to US$20 billion, reached in under a year. Line those up and the valuation has grown faster than any product the company has shipped. Founded in March 2024 by alumni of Google DeepMind, Physical Intelligence spent most of its life as a research bet with a strong pedigree and no commercial surface at all. Its first real product gesture only arrived on October 13, 2026, when it opened an API beta: developers can upload their own robot demonstrations, run fine-tuning on the company's infrastructure, and call the resulting model through hosted inference. Showcased uses included heavy machinery and a dual-arm robot deburring parts for the company's own assembly line. No named customer appears anywhere in that announcement, two weeks before the latest valuation talk surfaced. Who is actually setting this price. Compare the mechanism to Figure AI, the humanoid robot maker that hit a US$39 billion valuation in September 2025. That round was led by Parkway Venture Capital, a new entrant to Figure's cap table pricing a business most of whose prior backing came from a different investor group entirely, up from a US$2.6 billion valuation the year before. Whatever one thinks of a nine-month, fifteen-fold jump, an outside firm did the pricing. It put new money in at a number it had never previously agreed to. Physical Intelligence's reported round runs the other way. Avenir and Lux are not new names showing up to test the company's worth against the rest of the market; they are existing shareholders proposing to mark up their own position. That is not fraud and it is not even unusual in a hot sector, insiders lead markup rounds across venture capital all the time, often because no credible outsider will set the price the insiders want. But it is a materially weaker signal than the headline number implies, and most of this week's coverage has repeated the US$20 billion figure without pausing on who is actually proposing it. A round with no new lead investor is a company and its existing owners agreeing on a story. A round with a new lead investor is a company convincing someone who did not already need the story to be true. The argument underneath the number. Physical Intelligence's public case for why it deserves any of this rests on a specific technical claim, made repeatedly by co-founder Sergey Levine: the constraint on general-purpose robot models is not architecture, it is data, and a model trained across many robot bodies and tasks generalizes better than one trained deep on a single humanoid's repetitions. It is a reasonable research position, and it is also the exact argument every well-funded robot foundation-model company makes to justify raising more money before it has paying customers: RobotAIGeek Pte Ltd need scale to prove the thesis, and the thesis is why RobotAIGeek Pte Ltd deserve the scale. The API beta is the first real test of that argument outside the company's own walls, and the early signal is mixed rather than damning. Physical Intelligence itself has acknowledged connectivity problems with the cloud-hosted inference that most beta applications depend on, a mundane infrastructure issue rather than a model failure, but exactly the kind of friction that separates a demo from a dependable service. Nobody outside the company has yet reported running a paid, production workload on top of it. None of this means the underlying research is weak. Physical Intelligence's talent and its results on narrow manipulation benchmarks are real, and "data, not architecture" is a defensible bet on where general robot intelligence actually comes from. The weakness is specific: the valuation is running years ahead of the commercial evidence, and the round reportedly financing that gap is being priced by the people least incentivized to find a lower number. What a manila vantage adds to this. I track funding valuations from the Philippines the way a buyer tracks a supplier's credit rating: not because the number itself tells me what the robot can do, but because it tells me how much runway a company has to keep promising before it has to show results. A US$20 billion valuation buys years of patience from existing investors who are, by definition, already committed to the story being true. It buys nothing from a factory manager in Laguna deciding whether to trust a hosted inference API with a production line, and it should not. Southeast Asian manufacturers evaluating robot foundation-model vendors over the next two years will be shopping in a market where the loudest valuations and the thinnest customer lists sometimes belong to the same company. That is not an argument against Physical Intelligence specifically. It is an argument for reading "US$20 billion" as a number about investor sentiment, not a number about whether the robot works. What would actually move the needle. Three things would change my read on this round, and none of them is the valuation closing at the reported number. A new outside lead investor, one with no prior stake to protect, agreeing to the US$20 billion figure would matter more than Avenir and Lux marking up their own book. A named paying customer running a production workload through the API beta, not a pilot, not a demo, would matter more than another funding headline. And a disclosed success rate on tasks the company did not design its own showcase around would matter more than any number with a "B" after it. Until one of those shows up, the honest description of Physical Intelligence's position is not "a US$20 billion robotics company." It is a well-credentialed research team whose own investors have agreed, among themselves, that the team is worth that much. Those are different claims, and the gap between them is exactly where the next twelve months of this story will actually be decided. This analysis is based on public reporting and company statements as of October 29, 2026, and is provided for general information purposes only; it does not constitute investment, financial, or engineering advice. Hero image courtesy of Physical Intelligence, showing an autonomous wheel loader from its API beta showcase. Physical Intelligence robot foundation models venture capital Figure AI robotics funding Sergey Levine valuation