Europe's AI independence takes more than open models, despite Tsai's claim

What does open-source AI offer European developers?

The terms open-source and open-weight are not interchangeable guarantees of control. An open-weight model makes its trained parameters available; code and documentation may also be released, while the specific terms and materials vary by model.

Alibaba says it released its first Qwen and Qwen Chat models as open-weight offerings in August 2023, making code, weights and documentation freely accessible. That account describes those releases; it does not establish that every Qwen model has the same terms or that using one would make a European developer independent of outside providers.

For a European team, an available model can be a starting point for building an application or adapting a system to a task. But operating it can still require computing resources, hosting and technical support. The European Commission’s Open Source Strategy identifies sustainable funding, long-term maintenance, scaling and industrial deployment as challenges for the region’s open-source ecosystem.

That makes Tsai’s manufacturing-data argument a proposal, not evidence of a ready supply of usable training material. The SCMP reported that he sees factories’ data as an opportunity, but the reporting available here does not establish how much such data exists or what legal, commercial and technical conditions would govern its use.

Is open source Europe’s only route to AI independence?

The European Commission’s plans describe a broader approach. Alongside support for open technologies, they include computing and data infrastructure, skills, adoption in strategic industries and investment. The Commission says open technologies can reduce reliance on providers outside the EU, while its broader AI agenda includes AI factories and gigafactories and a proposed Cloud and AI Development Act intended to reduce strategic dependencies.

Those are policy aims, not proof that Europe has secured independence or that its initiatives will succeed. But they show that European institutions are pursuing multiple levers; the supplied evidence does not establish Tsai’s claim that open-source AI is the sole path.

Existing infrastructure is one part of that effort. EuroHPC says it is overseeing 19 AI factories and 13 AI Factory Antennas, which it describes as offering free, tailored support to startups and small and medium-sized businesses. Those programs provide support and access, not proof that Europe controls the full AI supply chain.

The next step is still prospective. EuroHPC’s call seeks proposals for up to seven AI Gigafactories, large facilities intended to combine computing, storage, networking, cloud access and AI support. The organization expects to select projects in early 2027, with operations to begin within 18 months of selection. It also expects more than €20 billion in private investment across the EU associated with the plan; that is anticipated investment, not money already spent or facilities already operating.

The difference between plans and delivered capacity is central to judging Europe’s options. Open models may let European developers choose and adapt software, while factories and gigafactories address the computing resources needed to develop and run AI. Neither the existence of a model nor an infrastructure proposal establishes that developers can rely on European suppliers for every part of the process.

What is Alibaba’s role in the argument?

Alibaba is relevant both because Tsai chairs the company and because it develops Qwen models. Its account of the first Qwen releases shows that it has offered some models with weights, code and documentation available. It does not show that Tsai was speaking for the Qwen team, make his remarks a product announcement or provide an independent assessment of Qwen’s suitability for European industrial use.

For European buyers, the distinction is practical. A model developed outside Europe might offer another option, but its openness alone does not settle where it is hosted, what computing it relies on, who maintains it or which terms apply. The Commission’s stated goal of giving public administrations more choice and reducing lock-in is a policy objective, not evidence that open-source procurement has already achieved those outcomes at scale.

Tsai’s comparison with China also remains an argument rather than a demonstrated template. The reported rationale points to China’s open-model push and Europe’s industrial base, but the available account does not show how the two regions’ conditions compare or establish that the same approach would produce technological independence in Europe.

The test will be whether Europe’s plans deliver durable capacity alongside access to models: computing and data infrastructure, reliable maintenance, and routes for businesses to put AI into use. The gigafactory projects have not yet been selected, and their practical services and impact remain to be seen. For now, open-source AI is one tool in Europe’s stated strategy, not a proven substitute for the other dependencies Tsai’s claim leaves unresolved.