Faced with a future where AI plays an essential role in most manufacturing processes, where would a European company want to get its data? Alibaba president and co-founder Joe Tsai is clear: perhaps not in a closed API provided by an external company, but in an open model that can be trained to measure, as he said at the Wave event in Italy.
In fact, Tsai goes a step further: she believes that this is the only way for Europe to achieve technological independence. And he makes it clear: According to him, while Chinese AI companies share their innovations and progress, US labs do not.
Why is it important?. Because today and despite warnings from Mistral CEO Arthur Mensch, about the two-year margin that Europe has to build its own AI infrastructure and various efforts by European administrations and its business ecosystem, the reality is that it continues to rely on technology from the US and China. And events like the international blockade of Anthropic’s most powerful models by the US over the summer have already shown that a government can turn off the tap on demand.
Since then, companies like Siemens or Renault simply combine models from different sources. But the problem doesn’t end there: China and the US control about 90% of the world’s computing infrastructure for AI, so even if it wanted it, Europe lacked the capacity.
Towards Sovereignty (Anticlockwise). On June 3, Brussels presented a package of measures aimed at reducing its external dependence on technology that included an economic injection to accelerate the construction of data centers, artificial intelligence, semiconductors and put open source on the table as an option. Of course, if we talk about open source, as recalled last September’s Mozilla report collected by Tom’s Hardware, China is the one with the leading voice.
Tsai’s proposal. What Joe Tsai proposes is fine-tuning in practice: download an open model of weights, run it on your own infrastructure, and train it with internal data, so that sensitive information doesn’t leave the company.
Coincidence or not, Alibaba released the weight of Qwen3.8 in August and its best version has 2.4 trillion parameters and about 95 billion assets per token, so it needs enough infrastructure to run, so for most companies the smaller version (Qwen3.8-27B) will be more practical, which allows commercial use and thanks to the license.
He is not neutral. It’s worth remembering: it’s an interested party because Alibaba is behind the Qiwen Open AI model and, in addition, a few weeks ago it presented the new Zhenwu V900 chip (it will start production in 2027) and announced future models. However, Alibaba has not always chosen open source: the opening came with Qwen3.8 and it just didn’t happen, since its compatriots and “rivals” Moonshot or Dipsic are already releasing big open models.
Open models have advantages over closed ones in terms of control and flexibility, which at first seems like a good idea. Of course, if the models are Chinese, this means going to others depending on third parties, who have the power to change the license or keep the best ones. It’s a patch, but it doesn’t solve Europe’s real problem: it lacks affordability and its own model.
In Xataka | To become technically “independent” of the United States, the European Union already has a plan: four desperate measures
In Xataka | Europe wants to rely less on the US for AI chips. Germany and the Netherlands have a plan: use AI to design them
cover alibaba
