A week ago in Beijing, China's leading AI executives made an unusual admission: they are losing. Justin Lin of Alibaba put a number on it—less than 20% odds that any Chinese company catches OpenAI or Anthropic in the next three to five years. Tang Jie, from Zhipu, was blunter: "The gap could actually be widening," he said at the AGI-Next summit. This came the same week two Chinese startups had just hauled in $1.2 billion in fresh IPO capital.
The paradox cuts deep. China is losing in what it thinks matters—technological frontier, foundational research, the raw ability to build advanced models. Yet it is winning decisively at what actually matters: the global market.
Geographic Expansion
A Microsoft report published that same week tells a different story. DeepSeek—the model the West wrote off as competent but unremarkable—controls 89% of China's AI market. Russia: 43%. Iran and Syria: around 25%. Much of Africa: 11–14%.
Why? Because DeepSeek is free. It ships pre-loaded on millions of Huawei phones. And it is open-source, meaning anyone can modify it. For the billion people in regions where OpenAI requires a Western credit card they do not have, DeepSeek is not an alternative. It is the only option.
Scenario 2030
By 2030, Microsoft estimates, two-thirds of the world's population could be using Chinese AI models daily, whilst the global technology elite remains tethered to American ones.
In North America and Europe, DeepSeek barely registers today. But that may shift. China has already invited independent researchers to examine DeepSeek's performance in a paper now published in Nature—a strikingly transparent move for a startup, and one that opens Western doors it previously could not reach.
And Europe?
The European Union is not facing this scenario squarely. It continues pursuing independence from the United States, but its fight is centred on software. The real power—who controls the silicon—sits in Asia.
Engineer and AI hardware architect Bernardo Kastrup puts it plainly: Europe assumes sovereign AI means building models aligned with European values. It does not. Value alignment is not encoded in software. It emerges from who controls the infrastructure where models are trained—the hardware, the data, the computational foundation.
There is a path forward, though. European AI sovereignty across the entire value chain is achievable by 2030. Not in fantasy. In fact. But it requires something rare in Brussels: genuine continental coordination, abandoning the national silos that have neutered every attempt at European technological independence. It is the only defence of our values at a historical moment when, uniquely in recent decades, they are under pressure from both East and West. A conclusion I share—(and subscribe to)—with Kastrup.
This article is republished from Futuribles. Here's the original article in Spanish and English.
