Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Retaliation seemed certain. When the US tightened its grip on advanced artificial intelligence technologies in January — blocking China’s access to advanced AI chips and locking proprietary models behind trade barriers — the response appeared predictable. China would build its own walls, guard its breakthroughs and double down on secrecy.
Instead, China is doing something unexpected: it is giving away its most advanced AI models.
In recent weeks, Chinese tech groups including Alibaba, Baidu and Tencent have been flooding the market with powerful AI models. But in an industry where secrecy is the norm, the real shock is their openness — these models are free to download, modify and integrate.
The pace of China’s open-source AI push has been relentless. Since the debut in January of DeepSeek R1 — China’s answer to OpenAI’s o1 series — a wave of increasingly capable models has followed. Alibaba claims its latest AI reasoning model QwQ-32B rivals DeepSeek’s R1 and has performed well in official benchmark tests. Every few weeks, another arrives, pushing the boundaries of what open-source AI can do.
At first glance, this surge might seem like a statement that AI should be open to the world, not just a handful of companies. But in business and geopolitics, generosity is rarely without strategy. The real question is not why China is open sourcing its AI, it is why the world assumed it would not.
For now, most US tech groups treat AI like an exclusive resource, restricting access to their most powerful models behind paywalls. OpenAI, Google DeepMind and Anthropic limit full access to their most advanced AI models, offering them through plans such as paid subscriptions and enterprise deals. Meanwhile, the US government views open-source AI as a security risk, fearing that unregulated models could be fine-tuned into cyberweapons. US lawmakers are already pushing to ban DeepSeek AI software from government devices, citing national security concerns.
But Chinese tech groups are taking a very different approach. By open sourcing AI, they not only sidestep US sanctions but also decentralise development and tap into global talent to refine their models. Even restrictions on Nvidia’s high-end chips become less of an obstacle when the rest of the world can train and improve China’s models on alternative hardware.
AI advances through iteration. Every new release builds upon the last, refining weaknesses, expanding capabilities and improving efficiency. By open sourcing AI models, Chinese tech groups create an ecosystem where global developers continuously improve their models — without shouldering all the development costs.
The scale of this approach could fundamentally reshape AI’s economic structure. If open-source AI becomes just as powerful as proprietary US models, the ability to monetise AI as an exclusive product collapses. Why pay for closed models if a free, equally capable alternative exists?
For Beijing, this strategy could be a powerful weapon in the US-China tech war. US AI companies, built on monetisation through enterprise licensing and premium services could find themselves in a race to the bottom — where AI is abundant, but profits elusive.
Of course, this comes with trade-offs. If AI is freely available, nothing will stop foreign companies from taking China’s models, refining them and outcompeting Chinese companies. Over time, companies such as Alibaba, Baidu and Tencent may face the same pressures as US counterparts — forcing them to restrict access to protect intellectual property and generate revenue.
Beyond market dynamics, Beijing may have its own reasons to rethink this approach. The Chinese government, which prioritises control over key technologies, may also push for stricter AI regulations to manage misinformation, maintain oversight and ensure compliance with state policies.
But for now, open-source AI remains China’s best bet — a way to compete without access to the best chips or the advantage of an early lead.
The timing of the open source rush is no coincidence. It is a response to a closing window. With US chips and AI technology restrictions set to tighten under President Donald Trump and proprietary AI models becoming entrenched, China’s most effective strategy is speed and scale. To flood the market, to shift the balance before AI monopolies emerge.
If OpenAI, Google and Microsoft have already won the AI race as we know it, then China’s best move would not be to compete — it would be to make winning meaningless.
june.yoon@ft.com
