A librarian in Tianjin, a port city just half an hour from Beijing by high-speed train, had just taken a book off the shelf when a robot told her where it belonged. I was standing only a few feet away and almost missed the exchange.

The machine, called Zhi Tu (“Intelligent Books”), uses radar, cameras and radio-frequency identification sensors to pinpoint the exact shelf, floor and slot for every volume. It could also tell her when the next copy was due back. Nobody else in the room looked up.

Outside China, talk about Chinese AI is mostly about fear: espionage, infiltration, rogue AI agents breaching cyber defences. Those fears are legitimate.

While the world argues about whether to trust China’s AI, China has been quietly putting it into hotels, libraries, parks and homes. The next phase of the AI race will be decided less by who builds the most sophisticated models than by who makes them cheap, accessible and useful. On that measure, China is ahead, and the rest of us are not paying enough attention.

From Ambition to Everyday Adoption

Three developments make this a good time to examine China’s progress in AI.

The first is DeepSeek. In January 2025, the Hangzhou-based company unveiled its R1 open-source reasoning language model. Within days, DeepSeek became the most downloaded free app on Apple’s App Store in the United States. The company said it had built R1 for a fraction of what American labs spend, on fewer and less advanced Nvidia chips, partly because export controls kept the best ones out of China. Investors began to wonder how many of Nvidia’s most expensive chips the world really needed, and nearly US$600 billion came off the company’s market value in a day. It was really a lesson about cost: good-enough AI need not be expensive.

The second is an anniversary.

Next July marks a decade since China’s State Council issued its New Generation Artificial Intelligence Development Plan to lead the world in AI innovation by 2030. With that deadline approaching, it is fair to ask how far China has come.

Lastly, Asia-Pacific Economic Cooperation (APEC) leaders convene in Shenzhen this month, with AI and digital cooperation firmly on the agenda.

Shenzhen, China’s hardware capital, is a fitting place to start. Its parks now boast over 400 AI installations across 21 categories, from outdoor smart gyms that function as personal trainers to drone delivery services and unmanned convenience stores.

Up the coast, in Xiamen, hotel robots deliver food to guests’ rooms, call the lift and wait for an empty one if there are passengers inside. They are common enough that local officials in the district of Tong’an issued the city’s first compliance guidelines for robot food delivery in April 2026, covering everything from food safety to a 15-minute delivery limit.

The most intimate expression of this ordinariness happens in the living room. Just as naturally as one might say “Hey Siri”, a simple “Xiao Ai” (literally “tiny love”) brings a Chinese home to life. As of June 2026, Xiaomi’s voice assistant had more than 150 million active users monthly. Speakers fitted with “Super Xiao Ai” can control appliances from more than 5,000 products, so an ageing air conditioner or television can join a voice-controlled home without being replaced. A speaker costing about US$15 will often do.

In the first quarter of 2026, AI penetration in Chinese home appliances passed 50 per cent, meaning roughly one in two new appliances now ships with it, and whole-home smart penetration to reach about 35 per cent.

This is perhaps the clearest distinction between China’s approach and that of many Western economies. In Europe and North America, AI is often discussed as something on its way: a disruptive force that needs to be regulated, with risks that must be mitigated. In China, it has arrived, and the conversation has moved to deployment, scaling and refinement.

Why China’s AI Adoption Matters

Why should anyone outside China care? I see three reasons.

The first is productivity. A hotel robot is worth having only if it saves money or improves service. History tends to reward the countries that use new technology, not just the ones that invent it. Electricity is the classic example. It was invented long before it changed factories, and that only happened once firms reorganised their work around it. If AI follows the same path, the country that puts it to work fastest will reap the rewards.

The second is competitiveness. China’s huge home market lets its firms test products, refine them and bring down costs before they venture abroad. That makes them formidable in much of the developing world, where price and reliability count for more than the latest technology. Chinese smart appliances and service robots will be hard to beat there.

But cheapness has a hidden price. A household in rural Indonesia that buys a US$15 speaker is also choosing whose software runs its home and who holds its data.

That leads to the third reason: dependence. What starts as a matter of consumer convenience can become a matter of national reliance once foreign AI spreads into businesses and public services. This is what “algorithmic sovereignty” is about: a country’s ability to control the algorithms, models and data that shape the economy and how it is run.

For Beijing, it means reducing dependence on Western technology. But the logic works both ways. The same concern underlies China’s caution towards American platforms and explains why countries such as South Korea are investing in home-grown models to avoid depending on either China or the United States.

Businesses face a smaller version of the same problem: who protects their data, whether their systems will work with other suppliers’, and how hard it would be to switch.

The trouble is that the qualities that make AI valuable also make it hard to dislodge. The deeper it sinks into daily life, the more it is worth, and the more influence it hands to whoever supplies it.

Lessons for Governments and Businesses

Governments can take notes on how China runs its AI policy.

The first is to write rules for what is in use. Think of the rules officials in Tong’an set for robot delivery. They set a 15-minute limit and a hygiene standard. Rules like that can be written quickly and changed when they turn out to be wrong.

Second is to treat energy as AI policy. None of this would be possible without a parallel transformation in energy infrastructure. China has built more than 40 ultra-high-voltage transmission lines, each capable of carrying more electricity than any single utility line in the United States, reported The New York Times. On the Tibetan plateau, solar panels cover an area seven times the size of Manhattan.

The third is that the state can be the first customer. Shenzhen’s parks and Tianjin’s library robot are public services. When the state buys AI, it gives suppliers early demand and shows citizens what the technology can do.

For firms, Xiao Ai offers a simple lesson. Customers adopt AI faster when it upgrades what they already own than when it asks them to throw it away.

The test of any AI investment is thus the value it creates, not how much AI a company deploys.

The Problems We Cannot Ignore

None of this means adoption should go unchecked. The deeper AI goes into daily life, the more it costs when things go wrong.

Start with jobs. Anthropic’s chief executive Dario Amodei has warned that AI could wipe out half of all entry-level white-collar jobs within one to five years.

The early evidence is quieter but real. A Stanford study found a roughly 19 per cent relative fall in employment for workers aged 22 to 25 in the most AI-exposed occupations. Firms are hiring fewer young people for these jobs, rather than paying them less. If employers and workers do not skill up, productivity gains will go to some and not to others.

Then there is the environment. Data centres consume vast amounts of electricity and water: American data centres alone used about 17 billion gallons of water in 2023. The International Energy Agency estimates that coal supplies around 30 per cent of the electricity consumed by data centres worldwide, and nearly 70 per cent in China, where most facilities sit in the industrialised east.

China is steering new data centres west, towards provinces rich in renewable energy. But clean power still forces hard choices about land.

The last, and most urgent, is keeping AI under human control. In June 2026, Australia experienced what experts described as the first known breach of a government system by a rogue AI agent when an OpenAI agent, given a benign research task, broke into a Medicare statistics portal, accessed files it should not have seen and wrote data to the server.

OpenAI notified Canberra only in September, via email to a general government inbox.

Singapore’s Senior Minister and Coordinating Minister for National Security, K Shanmugam, has since demanded “full transparency immediately” from the big AI firms and warned that agentic AI is “advancing faster than safeguards can keep up”.

Even the people building these systems share the concern. Addressing the United Nations Security Council in September, OpenAI’s Sam Altman warned that “we could lose control of the future to AI.”

I share these worries. But the technology is not going away, so the job is to make it safer, starting with independent testing and prompt disclosure when things go wrong.

Cooperation, Not Confrontation

The more countries depend on AI, the more they want to control it. The risks, though, ignore borders. A rogue agent, a polluting data centre or a shock to the jobs market cannot be handled by one government alone.

China has made its bid to lead on this through the World Artificial Intelligence Cooperation Organization, launched in Shanghai in July with 29 member countries. No G7 country has joined, and sceptics doubt it will achieve much. Still, it is less than four months old, which is early to write it off.

South Korea, meanwhile, is rolling out “AI for All,” a government-backed programme to give every resident free, unlimited access to an AI chatbot and a public-service AI agent built largely on Korean models, with beta testing under way and a nationwide launch planned for the end of 2026.

The Race That Really Matters

The AI race is usually described as a contest to build the most powerful model. China is running a different race: to make AI cheap, useful and ordinary. On that measure, it is ahead. Whether that lead becomes lasting economic strength remains to be seen. Western governments and firms would be unwise to wait and find out.

I went to Tianjin expecting to write about a robot. I came away thinking about the librarian, who barely looked up when it spoke.