I think China and the US are trying to win two different games.
Most people think the AI competition is a simple race:
Whoever builds the biggest model, trains it with the most GPUs, and reaches AGI first wins.
That is largely the American AI narrative.
But China’s AI strategy appears to be based on a different assumption:
The winner may not be the one who creates the smartest model first. The winner may be the one who builds the most complete AI ecosystem and integrates AI into the real economy.
A recent article published by Qiushi, a major Chinese policy journal, offers an unusually clear window into how China’s AI policymakers view the competition.
The author, Yu Xiaohui, president of the China Academy of Information and Communications Technology, argues that the AI race has entered a new phase: the competition is no longer just about algorithms. It is about entire systems.
And that difference matters.
The US approach: build the smartest brain
The Silicon Valley approach is relatively straightforward:
Build the largest models.
Acquire the most advanced chips.
Spend billions on computing power.
Create the most capable AI systems, then sell access to the world.
Companies like OpenAI, Anthropic, Google, and others are essentially betting that intelligence itself is the ultimate competitive advantage.
The logic is:
The first country to achieve AGI will create an irreversible lead.
This is why some American AI leaders frame the competition almost like a geopolitical countdown: if democratic countries do not win soon, authoritarian countries might dominate the future of AI.
The focus is on the frontier.
The biggest model.
The most powerful computer cluster.
The next breakthrough.
China’s vision is different.
It is less about creating an AI “god” and more about creating an AI “infrastructure.”
China’s key concept: full-stack coordination
One of the most important ideas in Yu’s article is full-stack coordination.
AI competition, he argues, is not decided by one technology alone.
It depends on the interaction between:
- chips,
- computing architecture,
- algorithms,
- frameworks,
- operating systems,
- industrial applications.
Nvidia is the perfect example.
Nvidia is not dominant simply because it makes excellent GPUs.
Its real advantage comes from the ecosystem around those GPUs:
CUDA, developer tools, software libraries, and the huge community built around them.
A competitor cannot defeat Nvidia by making a slightly better chip.
It has to build an entire ecosystem.
This is why China faces a difficult challenge.
China now has dozens of large-scale computing clusters and enormous demand for AI computing. But many domestic chip companies still operate separately, each with its own technical standards.
Developers often have to spend significant time adapting AI models to different chips.
The problem is not only producing hardware.
The problem is turning hardware into an ecosystem.
Why China sees AI as a new computing paradigm
The deeper argument is that AI may represent the same kind of technological transition as the smartphone revolution.
Before the iPhone, companies like Nokia, BlackBerry, and Microsoft were not stupid.
They were actually leaders in mobile technology.
But they were building phones based on old assumptions.
Windows Mobile was basically desktop Windows squeezed onto a smaller screen.
BlackBerry brought email into a mobile device.
Nokia improved traditional phones.
Then Apple rebuilt everything around a different logic:
A low-power ARM chip instead of PC processors.
A new operating system designed for touch interaction.
A completely different software ecosystem.
The winners were not the companies that improved the old system.
They were the ones that created a new system.
China’s AI policymakers appear to believe something similar is happening now.
Today’s AI infrastructure was not originally designed for AI.
GPUs were created for graphics.
CUDA was created for parallel computing.
Linux was created for servers.
HTTP was created for web browsing.
They work, but they were not born for AI.
A future AI-native world may require different chips optimized for inference, different operating systems designed around AI agents, and different computing models.
From this perspective, China’s disadvantage may also become an opportunity.
Because China does not have Nvidia’s CUDA ecosystem, and access to that ecosystem is increasingly restricted, China has an incentive to build alternatives from the ground up.
Not a copy of Nvidia.
A different architecture.
The biggest difference: AI as a product vs AI as infrastructure
Another major difference is how the two countries imagine AI’s purpose.
The American model is largely focused on AI as a product:
A powerful model becomes a service.
Users subscribe.
Companies monetize intelligence.
China’s approach is closer to seeing AI as industrial infrastructure.
How can AI improve factories?
How can it reduce manufacturing defects?
How can it optimize energy use?
How can it predict equipment failures?
For example, AI vision systems can inspect products in factories faster and more accurately than humans. Industrial models can analyze years of production data and suggest better manufacturing parameters.
The question is not:
“Can AI replace humans?”
The question is:
“How can AI make the entire system more efficient?”
That is a very different philosophy.
The hidden battle: who pays the cost?
This is where the debate becomes more complicated.
A technology company making billions does not automatically mean society is benefiting.
The real question is:
Does the technology reduce the total cost of running society, or does it simply concentrate wealth while spreading costs?
China often uses high-speed rail as an example.
Many individual railway lines are not profitable by themselves.
But the broader system benefits:
Millions of people travel faster.
Companies relocate more efficiently.
Regional economies become connected.
The entire economy becomes more efficient.
In other words, the system becomes more organized.
AI could work the same way.
A country may not need the world’s most expensive AI model sitting in a few data centers.
It may need millions of businesses, factories, schools, and individuals using affordable AI tools.
The goal is not only creating the smartest AI.
The goal is lowering the cost of intelligence.
The social problem facing American AI
This also explains why AI enthusiasm in Silicon Valley does not always translate into public enthusiasm.
Many Americans are excited about AI breakthroughs.
But many ordinary people worry about job losses, rising electricity costs, and whether the benefits will actually reach them.
When communities see massive data centers consuming huge amounts of electricity while electricity bills rise, they naturally ask:
“Who is this technology really serving?”
That is not simply anti-technology sentiment.
It is a question about distribution.
A technology that makes a few companies richer but increases costs for everyone else creates social resistance.
A technology that improves the efficiency of the entire system creates acceptance.
The difference is not only technical.
It is institutional.
So the difference isn't really who has better AI today.
It's what each country thinks AI is ultimately for.
The US is largely focused on building the most powerful AI.
China is increasingly focused on building an AI system that can be deployed across the entire economy.
Whether one approach will outperform the other is still an open question. But they're clearly aiming at different definitions of success.