There is a strange echo from the Cold War in the rivalry between USA and China over artificial intelligence. The rockets have disappeared. Data centres have taken the place that launch pads once occupied. And the people pushing this frontier forward are not astronauts but chip engineers, AI researchers and software developers working on machines that can reason through problems, write code, use digital tools and carry out tasks with surprisingly little supervision.
Yet the underlying contest feels familiar.
When Sputnik went into orbit in 1957, the shock ran deeper than the sight of a Soviet satellite circling Earth. It suggested that the USSR had developed the scientific expertise, industrial machinery and rocket technology needed to reach space, a capability many Americans had not expected Moscow to achieve so quickly. The United States responded with enormous investment in science and technology. The race eventually carried two superpowers to the Moon.
The AI competition has a similar psychological structure. America entered the generative-AI era with a commanding lead. Silicon Valley had the leading laboratories, enormous pools of private capital, sophisticated semiconductor companies and access to the world's most advanced AI accelerators. Nvidia's GPUs became the workhorses of the AI revolution.
Then China surprised everyone.
DeepSeek's R1, released in early 2025, was a Sputnik moment. It demonstrated that a Chinese laboratory could produce a highly capable reasoning model despite operating under restrictions on access to America's most advanced chips. More importantly, Chinese researchers were showing that brute-force computing was not the only route to better AI. Architecture, training techniques, inference efficiency and clever use of compute mattered enormously.
The surprise was not that China could build a chatbot. China had been investing in artificial intelligence for years. The surprise was how quickly its researchers moved toward the frontier while facing a significant hardware disadvantage.
That disadvantage is crucial. Modern AI is not simply software. It is an industrial supply chain.
At the heart of modern AI is computing power. That means huge numbers of GPUs and other specialised processors. Building and running them depends on advanced chip fabrication, high-bandwidth memory, sophisticated packaging, fast networking and vast amounts of electricity and cooling. None of this exists in isolation. It depends on a long chain of chipmakers, engineers, cloud providers, data centres and investors.
America has formidable advantages across this chain, particularly in advanced AI chips, capital and frontier-model development. China has been forced to compensate for restricted access to cutting-edge hardware by squeezing more efficiency from its available compute while accelerating domestic alternatives. Huawei, for example, is developing its Ascend accelerator ecosystem and attempting to connect large numbers of processors into AI clusters.
This is where the analogy with the Space Race becomes even more interesting.
The United States and Soviet Union did not compete merely to build individual rockets. They competed to build an entire technological ecosystem around rocketry, satellites, nuclear science, communications and aerospace engineering.
AI is behaving in much the same way.
The contest is no longer just about who has the smartest large language model. It is about who can build the most powerful AI stack: chips at the bottom, networking and data centres above them, foundation models above that, and applications and autonomous AI agents at the top.
And Chinese companies are no longer spectators.
Alibaba's Qwen family has become one of the world's largest open-weight AI ecosystems. DeepSeek has developed sophisticated reasoning systems. Moonshot AI has pushed long-context and agentic capabilities. Zhipu AI, MiniMax and others are competing in the same arena. Chinese AI systems can now handle tasks that once seemed firmly within the territory of America's leading models. They can write and debug software, work through mathematical problems, digest large documents and interact with external tools.
An AI agent is quite different from the chatbot most people first encountered. A chatbot answers a question. An agent can break a complicated objective into steps, call external tools, browse information, write and execute code, inspect the result and continue working. That shift could make AI less like a search engine and more like a digital employee.
Here too, Chinese systems are beginning to stand surprisingly close to their American counterparts. The gap has not vanished in every category. Current assessments still place leading US frontier systems ahead in several areas, particularly at the absolute frontier. But the distance is no longer comfortable. Some researchers now regard the US-China performance gap as effectively closed in certain areas, while other evaluations still find leading Chinese models behind the frontier. The exact size of the gap remains disputed, but the broader direction is difficult to ignore.
China has also discovered something strategically valuable: cheap AI can spread faster than expensive AI.
Many Chinese laboratories have embraced open-weight models. Developers can download them, fine-tune them and build specialised applications without paying the full price of a proprietary system. That creates a feedback loop. More users create more applications. More applications produce more engineering experience. More experience improves the models. Improved models attract more users.
Alibaba's Qwen ecosystem illustrates this phenomenon particularly well. Qwen has accumulated a vast number of derivatives on platforms such as Hugging Face, showing how rapidly an open model can become an ecosystem rather than merely a product.
This is perhaps China's most intriguing strategic advantage. It does not necessarily need to beat America in every benchmark. It needs to make its technology ubiquitous.
And China possesses something America cannot easily reproduce: an enormous manufacturing base. If AI becomes deeply integrated with robotics, factories, autonomous vehicles and industrial machinery, China's physical economy could become a powerful testing ground for what is increasingly called physical AI.
America, meanwhile, retains extraordinary strengths in frontier research, capital, software and commercial AI ecosystems. Its technology companies are spending staggering sums on data centres and specialised computing. The United States therefore has no obvious reason to assume that China's progress means American leadership is disappearing.
The more accurate picture is a contest in which both sides possess different pieces of the puzzle.
That is why the Space Race analogy works, but only up to a point.
The Moon had a finish line.
AI does not.
There will be no single afternoon when one president stands before a flag and declares victory. The real prize may be much larger and much less visible: the ability to turn intelligence into economic productivity faster than the other side.
Whoever builds the cheapest powerful models, controls sufficient compute, secures energy and semiconductor supply, deploys millions of useful agents and connects AI to factories and robots could gain an advantage that compounds for decades.
The rockets of the Cold War carried humans beyond Earth.
The machines of this new AI Cold War may carry human civilisation into an entirely different economic era.
And this time, the finish line is nowhere in sight.
The author is one of the website's editors.