Why China May Win The AI Race And How The US Can Still Fight

The uncomfortable argument gaining ground in Silicon Valley is that China may be better positioned to win the AI race than many Americans want to admit. That does not mean the United States has already lost. It means the old assumption that American frontier labs will naturally stay ahead is no longer safe.

The AI race is no longer only about who has the smartest closed model this month. It is about who can build models cheaply, deploy them widely, generate enough electricity, train enough engineers, control enough hardware supply and win developer trust across the world. On those measures, China has real advantages.

The Warning From Tech Leaders

Hugging Face CEO Clement Delangue recently argued that China is winning the AI race because Chinese developers are advancing through open-weight models and broader collaboration, while the US is building more in corporate silos. Business Insider reported that Delangue warned China could soon dominate not only open models but possibly the frontier itself.

Elon Musk has made a similar point from the compute and power side. Earlier this year, Musk argued that China could far exceed the rest of the world in AI compute because of its ability to scale electricity generation. Business Insider noted his view that power infrastructure, not only chips or algorithms, will become the next bottleneck. Musk has also suggested that Google may win in the West while China wins on Earth, a line widely circulated after his March comments.

Nvidia CEO Jensen Huang has also pushed back against panic over Chinese models. In an Axios interview, Huang said Chinese open-source models are excellent and should be used rather than banned. That is an important statement because Nvidia sits at the centre of the AI compute economy.

China Is Winning The Open-Weight Layer

The clearest Chinese advantage right now is open-weight AI. DeepSeek, Alibaba Qwen, Moonshot Kimi, Z.ai GLM and other Chinese model families have become serious global options for developers. These are not weak copycat systems. They are increasingly capable, cheaper to use and often easier for developers to test or adapt.

Kimi K3 is a useful example. We have covered how Kimi K3 turned into a broader AI market headache, how it topped a frontend coding leaderboard and how GPU demand forced Moonshot to pause new subscriptions. Whether every benchmark should be trusted is another question, but the market signal was clear: Chinese open-weight models are no longer background noise.

Alibaba is also leaning hard into Qwen. The Verge recently reported that Alibaba Qwen3.8-Max is being positioned against top American models and that Alibaba is returning to an open-weight release strategy. Qwen already has a large developer presence through Hugging Face and GitHub, and that matters because AI ecosystems are built through developer adoption, not only press releases.

This open-weight strategy gives China distribution. A closed American model may be better in a benchmark, but if a Chinese model is good enough, cheap enough and downloadable, many developers in Africa, Asia, Latin America and Europe will test it. In the long run, ecosystem reach can matter as much as raw model rank.

Chinese models are also changing the cost conversation. DeepSeek showed the world that strong AI could be built and served more cheaply than many assumed. Kimi, Qwen and GLM extend that pressure. If Chinese models keep undercutting expensive American APIs while remaining capable enough for business workloads, the US frontier-lab business model becomes harder to defend.

This is not only about saving money. Cheaper models make AI adoption easier for small businesses, schools, public agencies and developers in emerging markets. If China becomes the source of affordable, adaptable AI, it can win influence even in countries that still prefer American brands.

That is why the US should not dismiss cheap models as inferior models. In most markets, good enough and affordable beats perfect and expensive. The history of technology is full of examples where lower-cost tools won the world while premium tools won headlines.

Compute, Chips And Power Are Moving In China’s Direction

The US still leads in advanced chips, especially through Nvidia. But China is trying to reduce that dependency quickly. Reports have described Chinese efforts to build national AI data-centre capacity around domestic chips and to push state-backed infrastructure away from foreign hardware. A Tom’s Hardware summary said China has drafted a major AI data-centre grid plan targeting heavy use of homegrown silicon.

Export controls were meant to slow China down. They have done that in some areas. But they may also have forced China to become more disciplined about efficiency, domestic hardware and open ecosystems. A recent academic paper titled U.S. Policies Unintentionally Accelerated China’s Open AI Ecosystems argues that US controls raised costs for Chinese AI development while increasing the strategic value of open and locally adaptable systems.

China energy position is also important. AI is becoming an electricity business. Training and running frontier models requires huge power supply, data-centre construction and grid planning. Al Jazeera has reported that China cheap and abundant electricity may become a major AI advantage. Musk has been making the same point in his own way.

China could win because it is attacking the AI race from several directions at once. It has strong open-weight models, aggressive state support, a huge engineering base, fast industrial deployment, cheaper energy and a willingness to build domestic alternatives when blocked from US technology.

It also has a different commercial route. American AI companies are often trying to sell premium access to closed frontier intelligence. Chinese AI companies are increasingly spreading capable models across developer ecosystems at lower cost. That can build dependency and influence even if China does not always have the single best model on a given day.

There is also a psychological factor. The US sometimes behaves as if leadership is something to be defended. China behaves as if leadership is something to be seized. That difference shows up in infrastructure, manufacturing, subsidies, energy planning and open-model distribution.

Why The US Still Has A Chance

The US still has major advantages. It has Nvidia, the strongest frontier labs, deep capital markets, world-class universities, cloud giants, a culture of startup creation and global trust in many enterprise markets. OpenAI, Anthropic, Google DeepMind, Meta, xAI, Microsoft, Amazon and Nvidia are not weak players. The US can still win, but only if it stops treating export controls as a complete strategy.

The first solution is to build a serious American open-weight ecosystem. The US needs more strong, safe, commercially usable open-weight models that developers around the world actually want. Meta Llama helps, but the US needs more than one serious open-weight pillar. It needs universities, startups, public labs and cloud providers working in a more coordinated way.

The second solution is to make AI infrastructure easier to build at home and with allies. That means faster power approvals, more transmission, more data-centre capacity, more chip packaging capacity, and financing for trusted AI infrastructure in Africa, Asia, Latin America and Europe. If the US only sells restrictions while China sells usable systems, China will win many markets by default.

The third solution is smarter regulation. The US should not ban open models simply because some come from China. That would push developers elsewhere and slow American learning. It should instead create capability-based safety rules, clear disclosure standards, model-evaluation infrastructure and procurement guidance that lets companies use open models safely.

The fourth solution is price. American AI must become cheaper. If the best US models remain expensive black boxes while Chinese models are affordable and adaptable, developers will vote with their budgets. The US must compete on access, not only performance.

The Real Race Is For Developers And Deployment

The AI race will not be won only in benchmark tables. It will be won in banks, hospitals, classrooms, factories, government offices, call centres, coding tools and small businesses. It will be won by the models people can afford, trust, customise and deploy.

China is likely to win if the contest becomes open distribution plus cheap deployment versus closed excellence at high cost. The US can still win if it combines frontier research with open ecosystems, lower prices, trusted infrastructure and global partnerships.

The point is not to panic about China. The point is to take China seriously. Chinese AI is not waiting for permission. It is getting cheaper, stronger and more available. If the US wants to remain the centre of the AI world, it has to compete with that reality, not the older version of the race it was already winning.