It was a controversy laced with pride for He Xiaopeng. In November, He, the founder and CEO of Chinese physical AI firm XPeng, had just debuted his new humanoid robot, IRON, whose balance, posture shifts, and coquettish swagger mirrored human motion with such eerie precision that a slew of netizens accused him of faking the demonstration by putting a human in a bodysuit.
[time-brightcove not-tgx=”true”]To silence the naysayers, He boldly cut open the robot’s leg live on stage to reveal the intricate mechanical systems that allow it to adapt to uneven surfaces and maintain stability just like the human body. “At first, it made me sad,” He tells TIME in his Guangzhou headquarters. “The robot is like our classmate, our child. But later, I was proud.”
There’s plenty to be proud about across China’s burgeoning AI industry these days. Despite significant headwinds, such as restrictions on Nvidia’s most advanced GPUs, China’s core AI industry was estimated at $160–$170 billion last year. China has more than 5,300 AI enterprises and leads the world in GenAI patenting, with six times as many registered as the U.S.
It’s been quite a turnaround. When ChatGPT exploded onto the scene in late 2022, policymakers in Beijing were aghast. But then last January, an obscure AI startup named DeepSeek released a generative AI platform, R1, which was comparable to ChatGPT but purporting to use just a fraction of Nvidia’s bleeding-edge chips. Although those claims are hotly contested, R1’s debut sent markets reeling and frissons through Silicon Valley.
DeepSeek also sparked a rush of AI investment in China, spotlighted by the rise of six AI unicorns—StepFun, Zhipu AI, Moonshot, Minimax, 01.AI, and Baichuan—which became known as China’s “AI Tigers” as the nation’s cities and regions clamored to seed a local champion. “I don’t know where the ‘tiger’ thing came from,” MiniMax CEO Yan Junjie tells TIME with a chuckle. “We just continue to advance our model to increase our revenue share and serve our global audience better.”
MiniMax grew to prominence with low-cost AI video generation, though today boasts multimodal models that rank favorably across text, video, or audio benchmarks. But Yan doesn’t see his competition as with the other tigers—DeepSeek’s Liang is a friend, he says—but undercutting the industry behemoths. “The battle really lies with the big tech giants—not just the Chinese ones, but globally.”
While the common perception is that America leads the AI race, with its advanced AI models, chips, and private investor cash, China has its own advantages. These include a massive cohort of engineers and STEM graduates, lower costs, and a state-led development model. “The idea that China does not have a technology industry is lunacy,” says Nvidia CEO Jensen Huang. “Why would we underestimate such an extraordinary country?”
Moreover, while policymakers in Washington have repeatedly alluded to a “Manhattan Project” pursuit of Artificial General Intelligence (AGI)—human-level cognitive abilities to understand, learn, and apply knowledge across virtually any intellectual task—China is more focused on diffusing AI across society.
“I don’t think about AGI a lot,” says Robin Li, founder and CEO of Baidu, one of China’s foremost full-stack AI companies. “We are training our models, but the reason we train our model is to solve our application problems. I don’t think we should come up with a similar super smart AI that can be everything for everyone.”
Not that China is shy about its AI ambitions. In August, Beijing unveiled its AI+ Initiative, which grandiosely aims no less than to “reshape the paradigm of human production and life,” with AI to be integrated into 90% of China’s economy by 2030. Hot on the heels of AI+ came the latest draft of China’s 15th five-year plan, which sets ambitious targets to achieve “independent and controllable” innovation to upgrade its economy and slip the U.S. technological noose.
These goals are underpinned by a wholesale revamping of China’s scientific and innovation ecosystem, strengthening project-based funding for researchers, reorienting the education system toward STEM, while incentivizing private companies to spend more on basic research with whopping tax breaks. Already, semiconductor R&D has tax relief of 120%—for every $1 million a firm spends on chip development, they deduct $1.2 million from their taxable income—and the draft suggests this could increase to 200%.
China’s presumptive position at AI’s vanguard marks a seismic shift. For decades, China relied heavily on resourceful harvesting of American IP for all its tech ambitions. Many of China’s hitherto top names in AI—including ByteDance founder Zhang Yiming, Xiaomi cofounder Lin Bin, and Alibaba CTO Wang Jian—cut their teeth at the Microsoft Research Asia (MRSA) lab in Beijing.
However, four companies founded by former MSRA staff—SenseTime, Megvii, Yitu Tech, and CloudWalk Technology—have been sanctioned by the U.S. for their roles in the mass surveillance and repression of China’s Muslim ethnic minorities in what Amnesty International has dubbed a “dystopian hellscape.” In addition, Xiao Rong, president of the U.S.-sanctioned Shenzhen Intellifusion, which provides facial recognition technology to Chinese police, worked at MSRA for over 10 years.
Under pressure from Washington, Microsoft has drastically reduced its China operations. Instead, a new generation of self-taught Chinese AI visionaries are moving fast and breaking things. Upon launching DeepSeek’s V2 model in April, Liang pointedly crowed to Chinese media that its development “doesn’t include anyone returning to China from overseas—they are all local. The top 50 experts might not be in China, but perhaps we can train such talents ourselves.”
Not least when the mighty Chinese state provides lavish support. With dominance of AI now a core government policy, every city and region is offering incentives to AI start-ups. In Shanghai’s eastern Pudong district, humanoid robots firm Agibot enjoys tax- and rent-free premises as well as many other perks. Agibot shipped 1,000 units in 2024 and aims for 10,000 this year, when co-founder and CTO Peng Zhihui expects to move into profit with 30% of revenue coming from overseas.
“Because Shanghai is a major manufacturing city—Tesla has its factory here—we have very solid strengths in supply chains,” Peng tells TIME. “And there are many top universities in Shanghai, so we can get high-quality talent in AI and robotics.”
Like DeepSeek’s Liang, Peng has never studied overseas, first joining OPPO’s AI Lab before a stint with telecoms giant Huawei and then co-founding Agibot in early 2023. “It’s been my dream since childhood to build a humanoid robot,” says Peng. “I know many top [Chinese] entrepreneurs dropped out from Stanford University, but I think the best thing in business is to discover issues from the real world.”
In echoes of how the U.S. overtook the U.K. and Germany following the last industrial revolution, real-world adoption may determine the race for AI dominance. Until China can close its hardware deficit regarding GPUs, its ambitions will hinge on whether it can undercut Western competitors with embodied AI platforms to supercharge diffusion.
That goes for the models, too. MiniMax is focusing on offering comparable services to OpenAI at around one tenth of the cost, with Yan noting how many users in South or Southeast Asia would balk at the $200 monthly subscription cost for ChatGPT Pro. Yan says minimizing outgoings and attracting a broad customer base is the surest route to success at a time when many U.S. companies have bottomless pockets for R&D—raising fears of an AI bubble.
“Being a sustainable business is definitely very important,” says Yan. “Our Capex and R&D costs are significantly lower compared to our U.S. counterparts, so for us it’s easier to achieve sustainable status.”
It’s not just the developing world that sees the benefits of China’s low-cost approach. So does Silicon Valley. In October, Airbnb CEO Brian Chesky revealed he was ditching ChatGPT for Alibaba’s Qwen, praising the Chinese model as “fast and cheap.” That same month, Social Capital CEO Chamath Palihapitiya disclosed that his investment firm preferred Moonshot’s Kimi K2 as “a ton cheaper” and “way more performant” when pitted against American models.
Up until now, the U.S. had far outpaced China when it comes to diffusing general purpose technologies throughout an economy, which typically takes multiple decades. The question is whether the same dynamics will apply to AI. “There’s a difference between demonstrated capabilities on isolated benchmarks and actual integration with improving business productivity,” says Jeffrey Ding, author of Technology and the Rise of Great Powers: How Diffusion Shapes Economic Competition. “The U.S. is very well positioned, but it’s going to be a diffusion marathon.”
China’s determination to break free of the U.S. stranglehold on chips is no secret. While the Trump Administration recently brokered a carve-out for Nvidia to export its second-most powerful H200 chips to Chinese firms, Reuters now reports that Beijing is blocking their import—presumably to boost demand for homegrown equivalents like Huawei’s Ascend 910C, which offers around 76% of the H200’s processing power and two-thirds of its memory bandwidth. Huawei has also launched CloudMatrix 384, a China-based AI system to rival Nvidia’s GB200 NVL72 but built using homegrown chipsets.
In November, Baidu also unveiled its latest M100 AI chips and plans to build a supernode capable of supporting “millions” of chips by 2030. Still, Li is candid about China’s progress in semiconductors—saying that China is still “a few years” behind on chips though fast catching up with regard to models. “So I’m not so worried about the restrictions on chips, although I’d very much like to get access to the most advanced Nvidia chips.”
Those Nvidia chips are, of course, fabricated by TSMC predominantly in Taiwan—the self-ruling island that China claims as a renegade province and whose “reunification” President Xi Jinping has called a “historical inevitability.” In a May speech to the Shangri-La Dialogue security forum in Singapore, U.S. Defense Secretary Pete Hegseth said a Chinese invasion of the democracy of 23 million “could be imminent.” Beijing is certainly honing its capabilities with AI at the fore.
In September, Beijing’s Tiananmen Square hosted China’s biggest ever military parade to mark 80 years since Japan’s surrender in World War II. After striding to the rostrum flanked by North Korea’s Kim Jong Un and Russian President Vladimir Putin, Xi watched as an AI-enabled inventory of advanced weaponry filed past, including smart tanks, robot dogs and wolves, and lasers designed to thwart the kind of autonomous drone swarms currently terrorizing Ukraine. “AI features very prominently in Chinese military strategy,” one former PLA officer tells TIME.
The peerless role TSMC plays in global supply chains has been dubbed a “silicon shield” that raises the cost of a Taiwan conflict to prohibitive levels, estimated at $10 trillion by Bloomberg Economics, or some 10% of global GDP. Whether the current push to homeshore semiconductor manufacturing to the U.S. weakens that shield is a huge question—not least since Nvidia began fabricating its latest Blackwell chips in TSMC’s Arizona plant in October. Still, “the vast majority of manufacturing will still be done in Taiwan,” says Nvidia’s Huang. “We have redundancy, we have diversity, but there’s no question we’ll be dependent on Taiwan for a long time.”
Until China manages to compete with Taiwan on chips, or forcibly brings the island to heel, Beijing hopes to steal a march on Washington through frontier AI applications. Last March, China’s Ministry of Industry and Information Technology and civil-aviation and transport regulators released a six-year plan for the low-altitude economy, exploring regulations for aerial tolls, pilot licenses, and establishing trial areas where early-stage eVTOL (electric vertical takeoff and landing vehicles) can fly around actual city environments.
Other than humanoid robotics, Xpeng is due to roll out its first commercial eVTOL in the third quarter of 2026. For less than 2 million RMB (approximately $280,000) customers receive a six-wheeled hybrid AeroHT Land Aircraft Carrier that contains a self-deploying six-propeller drone, which seats two adults and can be either piloted or flown autonomously for up to 45 minutes. New regulations had to be introduced for every step, from the number of wheels on the landing module to charging certification and even a bespoke pilot qualification. “We cannot deliver this product without laws and regulations,” says He. “That’s why we had cross ministerial meetings. Overall, it has been very effective.”
Still, China isn’t content to merely play catch up. The office of Zhu Song-chun feels more fitting for an 18th-century philosopher than globally renowned AI professor. It occupies an island within a sun-dappled ornamental lake in Peking University, whose campus was once an imperial garden, blanketed with crabapple blossoms in spring and ginkgo leaves in the fall. Guests seeking an audience with Zhu must hop across polished stepping stones—but once they do, they will incongruously find one of the world’s foremost futurists, into whose research the U.S. once poured $30 million in federal grants.
Zhu was born in China during the bedlam of “Great Helmsman” Mao Zedong’s Cultural Revolution. In 1992, he moved to the U.S. for his doctorate at Harvard and later ran one of the world’s most prolific AI research centers at UCLA. For a decade, he undertook AI projects for the Pentagon and the U.S. National Science Foundation. But in August 2020, Zhu took up professorships at two top universities in Beijing, where he also leads the state-sponsored BigAI institute, which today boasts 300 researchers fixated on realizing AGI.
Zhu points to many reasons for his abrupt homecoming, including the first Trump Administration’s 2018 China Initiative, which investigated hundreds of prominent Chinese-American academics for alleged espionage, with some 250 losing their jobs. Trump 2.0’s assault on learning institutions writ large has done nothing to convince Zhu he erred by returning to China. “Actually, they’re not only going against Chinese people [but] the whole of academia,” says Zhu. “If I stayed in UCLA right now, it’s pretty miserable, because I don’t have top students, I don’t have a lot of funding. And I’ve got so many big ideas.”
Chief among them is a revolutionary new cognitive architecture for AI. Zhu posits that the “big data, small task” approach of LLMs like ChatGPT will never achieve true intelligence, instead championing a “small data, big task” variant that better mimics human learning. Big data models learn by copying existing actions—how to fold clothes, write a dissertation, or create a video of a pig hoverboarding through space—by parsing previous examples and extrapolating similar behavior.
“But you can only solve tasks on that data set, so therefore it’s not intuitive, and does not generalize to other tasks,” Zhu tells TIME. “We have an infinite level of tasks to solve, so it does not contribute to true intelligence.”
Indeed, this pabulum only scratches the surface, argues Zhu, who in keeping with his scholastic digs has burrowed deep down into the very core of human identity: our genetic disposition to like certain shapes, tastes, and odors; our ability to learn and be influenced by the community; and our motivation to build relations with others, especially with shared interests.
By mimicking these parameters, Zhu created an AI model, dubbed TongTong, which exists in a virtual city with her own drives, motivations, and limited autonomy. Rather than simply learning via existing data or repetition—a token or statistical chain—TongTong learns via a causal one: If I choose one chair over another, what factors lie behind that decision? Is it the chair’s comfort, position far from a draft, or close to a friend. And at an even higher level, what value chain lies behind that choice: disliking hard surfaces, chilly breezes, or valuing company? “Large language models stay at a token level,” says Zhu. “But they don’t necessarily know the causality.”
Regulation is another area where China is trying to get an upper hand. While the U.S. operates a negative regulatory environment, whereby everything is legal unless explicitly banned, China has a positive regulatory environment, with only explicitly described commercial activities permitted. A hands-off regulatory approach may seem like a boon for American companies, but in practice it means a lot of lawsuits and bickering when unleashing AI applications in the real world.
Take autonomous driving: Uber sold off its self-driving business in 2020 after a fatal collision. Ford abandoned its stake in its robotaxi developer Argo AI two years later. In 2023, GM paused all its Cruise driverless operations, despite already plowing in $10 billion, following collisions that led to the suspension of California licenses.
By contrast, Baidu’s Apollo Go robotaxis have completed over 17 million rides globally—the most in the world—and 250,000 rides per week, neck and neck with Alphabet’s Waymo. But while the U.S. firm’s 1,000 vehicles operating in San Francisco occupies around a quarter of the city’s total ride-hailing market, Apollo Go has the same number ferrying passengers around China’s central city of Wuhan—but comprising just 1%, spotlighting vaster potential to ramp up and hone operations while exploiting economies of scale. China’s ride-hailing companies are already competing overseas; all three of China’s leading robotaxi firms—Apollo Go, WeRide, and Pony.ai—operate in the UAE and vie for market share in Europe and beyond.
“Whether other companies are also in a foreign market isn’t really a factor for us,” says James Peng, CEO of Pony.ai, which runs robotaxis across China, Europe, East and Southeast Asia, and the Middle East. “Rather we look at the market potential, the size, and how fast can the regulatory framework be ready.”
Zhu can’t help shaking his head when he sees the amount of cash the big Silicon Valley firms are plowing into AI but without, he says, a true understanding of the fundamental technology. Exhibit A: a company like DeepSeek can create its V3 model for $6 million, whereas Meta has plowed tens of billions of dollars into AI with few tangible results.
“A lot of these big guys who bet on AI, like Mark Zuckerberg or Elon Musk, I don’t think they have a deep understanding,” says Zhu, reeling off the various distinct fields such as computer vision, robotics, cognitive reasoning, natural language understanding, multiagent systems, and machine learning that all require their distinct expertise. “AI is such a big area you need to really immerse yourself into it for many, many years.”
But despite his breathless pursuit of AGI, Zhu still says that it will be realized by a gradual diffusion throughout industries rather than the sudden “boom” breakthrough akin to the Manhattan Project.
“Our big advantage is we can work with many local governments and industries,” says Zhu. “I can go to see a governor, the minister, and big CEOs to try to apply our technology quietly. To let TongTong improve the agents used in hospitals or the factory monitoring systems.”
TongTong—which Zhu claims already has the reasoning abilities equivalent to a six-year-old—has already proved more intuitive than big data models, using tools to solve problems and collaborating with peers without explicit instruction. The next step will be to move TongTong from the virtual to real world. “TongTong today is a system software,” says Zhu. “Of course, they can put it into a robot to become hardware. And then finally to new AGI chips that can replace Nvidia GPUs. That’s the key thing in the future.”
—With reporting by Billy Perrigo/Silicon Valley
Hence then, the article about how china caught up on ai and may now win the future was published today ( ) and is available on Time ( Middle East ) The editorial team at PressBee has edited and verified it, and it may have been modified, fully republished, or quoted. You can read and follow the updates of this news or article from its original source.
Read More Details
Finally We wish PressBee provided you with enough information of ( How China Caught Up on AI—and May Now Win the Future )
Also on site :