China experiences physical AI surge – and how the U.S. should respond

Humanoid robots box at the World Robotics Conference in Beijing. Credit: Georg Stieler

The latest data from the International Federation of Robotics indicates that, out of roughly 520,000 industrial robots installed worldwide in 2024, 54% went to China, 17% to the European Union, 8% to Japan, and 7% to the U.S. While China continued to grow from an already high base, shipments declined in the other three regions.

Robot density, a key indicator of a nation’s automation level, tells a similar story. Twelve years ago, the U.S. rate was almost five times China’s. Since then, the country has overtaken the U.S. and is now almost than twice its level.

Motivated in part by the prospect that smart, connected manufacturing could trigger reshoring to early‑industrialized economies, no country has automated its factories as decisively as China.

China scales automation industry

Backed by state policy and a willingness to accept short‑term pain to gain technological leadership, China has built a highly competitive industrial ecosystem. Domestic vendors’ share of the industrial robot market has risen from under 30% a decade ago to more than 50% today. In collaborative and mobile robots, local suppliers hold about 90%.

The country now sets the pace in electric vehicles (EVs), batteries, photovoltaics, and drones; application know‑how is increasingly flowing from China to Europe. In autonomous driving, China is running neck and neck with the U.S. More than half of the publicly listed companies in the humanoid robot supply chain are Chinese, estimates J.P. Morgan.

In mechatronics, no one brings new products to market faster, at high quality and competitive prices. Chinese industrial robots are roughly one‑third cheaper than those from Europe or Japan. Robot exports have grown about 65% annually since 2022.

The gains reach beyond local brands: a substantial portion of the hardware for Tesla’s Optimus humanoid is reportedly sourced from China. To stay competitive, German automation firms are expanding research and development in China. One recent example is STIHL’s decision to move development of its robotic lawnmowers to China.

China has invested in STEM education. Credit: Georg Stieler

Talent pool and AI provide tailwinds

China’s emphasis on science, technology, engineering, and mathematics (STEM) education is paying off. The country graduates about 3.5 times as many science majors as the EU and roughly 4.5 times as many as the U.S.

At neural information processing systems conference NeurIPS last year, U.S. institutions still produced the most papers, but scholars of Chinese origin formed the single largest group, underscoring the depth of the talent pool.

Vendors such as DeepSeek are setting new markers with capable, resource‑efficient open‑source large language models (LLMs), enabling applications that previously lacked a business case.

As The Economist reported last month, 80% of startups pitching at Andreessen Horowitz were using a Chinese open model instead of a U.S. one. In robotics, access to low‑cost hardware encourages experimentation and opens new use cases.

Semiconductors and state support

A key weakness remains the shortage of high‑performance domestic chips, though progress toward self‑reliance is arriving faster than expected. U.S. sanctions have created new opportunities along supply chains.

At the same time, demand for compute is surging. Models that capture the physical world are far more complex than LLMs or straightforward vision‑language models (VLMs). Huawei is working to position itself as a nexus in this stack.

China applies EV playbook to physical AI

After several financially lean years for the Chinese tech sector — with a drastic drop in venture capital and a record low in IPOs — Beijing is trying to reinvigorate the field. In February, Xi Jinping, president of the People’s Republic of China, met with leading tech entrepreneurs. Alibaba’s Jack Ma was publicly rehabilitated. Companies such as DeepSeek and Unitree stand for a new dynamism.

Humanoid robots and “physical AI” are the hottest sector. By the end of May 2025, investment in robotics and embodied intelligence exceeded the full‑year 2024 total.

That pace accelerated in Q2 2025. By July, we calculate that China invested $3.4 billion in new robotics ventures — 42% more than the U.S. and five times that of Europe. Venture funds are back, large tech companies such as Meituan and Alibaba are active again, and state‑backed funds are co‑investing alongside them.

Chinese humanoids move to mass production

Chinese firms are surging ahead toward scale. The question for all humanoid companies is whether they can ship robots at volume and integrate them usefully into real scenarios.

According to a report earlier this year, six of the country’s 11 major humanoid‑robot manufacturers, including Unitree, AgiBot, Galbot, Engine AI, and Leju Robotics, launched mass‑production initiatives in 2024, aiming to produce over 1,000 units per year by the end of 2025.

Collectively, Chinese firms are projected to manufacture more than 10,000 humanoid robots in 2025, accounting for over half of global output. UBTECH, AgiBot, and Unitree captured roughly 60% of the domestic market in the first half of this year. Ironically, there is still substantial manual labor in key components for small‑batch humanoids.

Can humanoids integrate into real, useful scenarios?

Despite the optimism and capital inflows, many experts caution that humanoid robotics are not yet mature. Significant technical and commercial hurdles remain in the near term.

Recent procurement deals are notable. In June 2025, a China Mobile-affiliated platform awarded a CN¥124 million ($17.4 million U.S.) procurement split between AgiBot and Unitree. In July, UBTECH won a ¥90.5 million ($12.7 million) order from a Shanghai‑based EV startup. It was the largest known single‑company humanoid robot order worldwide at that time.

Noetix disclosed 2,000+ unit orders in H1 2025 (worth over ¥100 million or $14 million), mainly from K‑12 schools, universities, and vocational institutes.

As of mid‑2025, the vast majority of humanoids in China are prototypes or pilot units, not mature mass‑produced products. A large portion of deployments are demonstrations, performances, or data collection rather than mission‑critical tasks, with deliveries scheduled over the next two to three years.

Commercialization timelines (industry guidance)

Unitree said it expects visible gains in humanoid intelligence within the next one to two years. It said real commercialization will begin within three to five years, likely first in public service or select manufacturing tasks.

Inovance, the largest industrial automation company in China, is preparing for a viable market by 2030.

Investments shift toward VLA models

Attendance was strong at China’s World Robotics Conference. Credit: Georg Stieler

The crucial question is about software. When will flexible, semi‑autonomous robots become reality? The current trend targets physical AI plus motion control “moats” and favors upper‑body humanoids on wheeled bases – two arms, not necessarily legs.

Capital is gravitating to the “brains.” Nearly every July deal earmarked more than 40% for data‑engine generation and large embodied‑model training.

Vision-language-action (VLA) models from Spirit AI, XSquare, and Galaxea demonstrated task suites such as cloth folding, drivetrain assembly, bed‑making, and other handling tasks. While these systems may still trail Western leaders like Google Gemini Robotics or Physical Intelligence in breadth of generalization, data diversity, and benchmarked reasoning depth, the Chinese firms are moving faster in deployment and task diversity within their chosen niches.

Many Chinese companies demonstrated their physical AI at the World Robotics Conference in Beijing last month. XSquare founder Wang Qian argued that progress is outpacing expectations.

Models with GPT‑3‑like capability for the physical world could emerge within a year, with commercial use one to two years later and household adoption in three to five years.

Companies in China attack data bottleneck

China is attacking the embodied AI data bottleneck on several fronts:

• Large real‑world data factories, such as AgiBot’s and new hubs in Shanghai, Beijing, Shenzhen/Tianjin
• Human demonstration and simulation pipelines like those of Parsini and academia
• Regional consortia in Hubei and Guangzhou
• Open datasets including AgiBot World and RoboMIND

Together, they form a national data engine for multi‑skill, multimodal robot learning. As centers interconnect via data exchanges, throughput is rising: AgiBot’s ~50k/day capture is no longer unique; peers target similar output, and Parsini aims an order of magnitude higher.

Coupled with cloud and chip partnerships, this surge marks an inflection toward broad real‑world deployment – potentially at a scale that outpaces other countries.

Beijing provides policy support for physical AI

Beijing is throwing its weight behind “new productive forces” in physical AI. “Embodied intelligence” was mentioned for the first time in the premier’s annual work report in March.

Also that month, during the Two Sessions, the National Development and Reform Commission (NDRC) announced a ¥1 trillion ($138 billion) fund for “hard tech.” This includes robotics and AI, chips and advanced equipment, quantum computing, clean energy, and other emerging technologies and will use market‑based equity with lifecycles up to 20 years.

At the second Embodied AI Conference hosted by the Chinese Academy of Sciences together with leading universities in March, delegates adopted a white paper with a three‑stage roadmap:

1. 2025 to 2027, foundation: Shared datasets and open middleware
2. 2028 to 2030, scaling: Deployment in factories, logistics, elder‑care pilots
3. Post‑2030, generalization: Mass‑market humanoids and composite robots

Regulatory scaffolding is emerging. The Ministry of Industry and Information Technology’s (MIIT) Intelligent Data Collection Standard 1.0 (11/2024) proposes a unified, enforceable framework for synchronizing, formatting, labeling, and quality‑grading multimodal robot training data. It is aimed at interoperability and privacy compliance.

At the municipal level, Shanghai has issued China’s first plan for embodied intelligence, pairing broad R&D support with shared infrastructure – compute, testing, pilot production, and financing. The city also added demand‑pull tools such as vouchers, demonstration projects, and modest sales incentives to move physical AI from lab prototypes to scaled production.

Nationally, the“AI+” plan, announced last month, seeks to mainstream intelligent systems across China’s economy. It has set adoption targets above 70% by 2027 and 90% by 2030 on the way to an “intelligent economy” by 2035. This creates stable demand and infrastructure support for AI in robotics, smart vehicles, healthcare, and upgraded infrastructure.

Is Chinese AI investment a bubble?

Critical observers have asked how many industries China can sustain with heavy state support, and for how long.

When asked why investment in humanoids is booming, one insider at a leading manufacturer put it bluntly: Capital is rotating into this sector because several of the previously “hot” plays failed to make money.

Constraints are real: Local governments that underwrite much of the support carry heavy debt burdens. Subsidy-driven dominance is provoking trade pushback abroad, and some target sectors might be structurally harder to catch up in.

At the same time, Beijing views physical AI capabilities as essential to future productivity, demographic headwinds, and national security. Past episodes show China can absorb significant inefficiencies and still achieve strategic outcomes, albeit at high cost. Pockets of overbuild and consolidation are likely, but it would be a mistake to underestimate China’s resolve.

Although the U.S. invented the industrial robot, it no longer has a scaled, homegrown industrial robot OEM. Foreign vendors are only now adding assembly capacity. There is a generational gap in manufacturing know-how. Policymakers want to reshore production, but skills shortages, high costs, a thinned-out supplier base, and a shortfall in core components are real hurdles.

The U.S. still leads in software and core AI, and its universities attract top talent. For Chinese humanoid makers, NVIDIA remains the central compute supplier for now. Its toolchains and training GPUs are the default worldwide.

Physical Intelligence and Google’s Gemini Robotics projects are viewed as benchmarks for end-to-end VLA models. U.S. humanoid efforts from Figure AI, Agility Robotics, and Apptronik look more technologically ambitious than most Chinese counterparts.

In theory, the U.S. allocates capital more efficiently. The global division of labor has raised living standards, but it also hollowed out domestic manufacturing.

Smart robotics firms arbitrage China’s cost and speed. Broad countervailing duties, while sometimes warranted, would raise the cost of automation and slow innovation. Meanwhile, dependencies have multiplied: rare-earth processing, battery cells for drones, and the concentration of advanced foundry capacity in Taiwan are well-known vulnerabilities.

In this context, the Association for Advancing Automation (A3) has warned that the U.S. risks losing its edge in physical AI. Without leadership in the physical manifestation of AI, A3 argues, the U.S. could lose not only the robotics race but the AI race itself.

What the U.S. should do to compete with China

To address these concerns, A3’s National Robotics Strategy, released in March, offered recommendations:

• Establishing a central robotics office in the federal government to coordinate strategy
• Offering tax incentives for automation adoption
• Making the government itself a lead customer of robotics
• Expanding STEM workforce training
• Boosting R&D funding
• Updating standards for AI-driven machines

Will that be enough to counter China’s momentum? Market economies prize efficiency and near-term returns. As highlighted above, China’s state-led model tolerates prolonged inefficiency and even years of losses to secure strategic breakthroughs.

What kind of U.S. industrial policy makes sense in this context? Silicon Valley grew out of defense procurement. The CHIPS Act and Inflation Reduction Act showed both what is possible and where policy design can improve. The goal must be better system design with market discipline, not central planning.

Rebuilding the industrial base is the core. Robotics companies need customers, sparring partners, and supply-chain synergies. An energy policy that delivers abundant, reliable, and affordable power is a structural advantage — especially relative to Europe’s higher industrial energy costs.

Capital will matter. The U.S. brings decades of venture capital experience, unmatched around the globe. However, capital-intensive fields like robotics often lost out to fast-scaling software plays over the past 20 years.

The LLM breakthrough and its transfer into the physical world appears to be changing that. Alongside Tesla and Alphabet, Amazon and Meta are building AI-enabled robots.

Google co-founder Larry Page is reportedly exploring AI-reinvented manufacturing at Dynatomics. Sequoia recently coined the term “cognitive assembly line.” The relative lack of legacy infrastructure might turn out as an advantage during the next technological revolution.

The country that attracts the brightest and most ambitious people will lead the world. The U.S. should therefore keep its openness to people and markets. In particular, as China eases skilled immigration rules and promotes its tech stack as an affordable alternative for the Global South.

Finally, rebuild critical hardware supply chains for actuators, precision reducers, high-efficiency drives, sensors, power electronics, batteries, and precision gears. With intense competition and local-content rules in China, leading European and Japanese automation suppliers are reassessing footprints and are open to North American co-location where demand and policy clarity exist.

The U.S. should court them with fast-track permitting, site-readiness, and mission-driven procurement to localize production and close the capability gap.

About the author and RoboBusiness

Georg Stieler advises some of the world’s largest robotics companies. He spent more than 10 years living in China and now splits his time between Switzerland and the People’s Republic of China. Thanks to multi‑year, close collaboration with AI startups in Silicon Valley, he is also deeply familiar with its culture.

At RoboBusiness 2025, Stieler will be participating in a panel on “Closing the Robotics Gap With China” along with Jeff Burnstein, president of A3, and Eric Truebenbach, managing director at Teradyne Ventures. He will also present in a session on the “Global Implications of China’s Robotics Push.”

RoboBusiness, the premiere event for developers and suppliers of commercial robots, will be on Oct. 15 and 16 in Santa Clara, Calif. The event will feature tracks on enabling technologies, humanoids (new), business, design and development, and field robotics, as well as the Physical AI Forum.

This year’s conference will also include more than 60 speakers, a startup workshop, the annual Pitchfire competition, and numerous networking opportunities. Over 100 exhibitors on the show floor will showcase their latest enabling technologies, products, and services to help solve your robotics development challenges. Registration is now open.

Keep track of your essentials with the Apple AirTag 4 Pack, the ultimate tracking solution for your belongings. With over 5,972 ratings and a stellar 4.7-star average, this product has quickly become a customer favorite. Over 10,000 units were purchased in the past month, solidifying its status as a highly rated Amazon Choice product.

For just $79.98, you can enjoy peace of mind knowing your items are always within reach. Order now for only $79.98 at Amazon!

Help Power Techcratic’s Future – Scan To Support

If Techcratic’s content and insights have helped you, consider giving back by supporting the platform with crypto. Every contribution makes a difference, whether it’s for high-quality content, server maintenance, or future updates. Techcratic is constantly evolving, and your support helps drive that progress.

As a solo operator who wears all the hats, creating content, managing the tech, and running the site, your support allows me to stay focused on delivering valuable resources. Your support keeps everything running smoothly and enables me to continue creating the content you love. I’m deeply grateful for your support, it truly means the world to me! Thank you!