- Alibaba chairman Joe Tsai said open-source artificial intelligence is Europe’s best path to technological independence from the United States and China, speaking at the Wave by Vento conference in Turin. He argued no country would want to rely entirely on another’s technology, since a change of government or circumstances could see access cut off.
- He urged Europe to build its own computing infrastructure to train models and run inference domestically, and said he is bullish on the region’s neoclouds, the specialised cloud firms operating data centres equipped with AI chips.
- Tsai said Europe is not making the most of its AI researchers, who are heavily recruited by US and Chinese labs, and that the region should train models on proprietary industrial data. He noted that Chinese factories hold valuable manufacturing data useful for training, and that many European countries retain large industrial bases.
- He said China has deliberately focused on open-source models and published research to drive global uptake, contrasting that with the proprietary approach of US firms including OpenAI and Anthropic, and remarked that the American closed-source labs no longer write papers because they do not want to share.
What Happened?
Alibaba shares traded at 106.72, up 2.32%.
Why It Matters?
Read the recommendation alongside who is making it. Alibaba is among China’s largest publishers of open-weight models, so the chairman is advising Europe to adopt an ecosystem in which his own company’s models are a leading option, while framing it as a route to independence from China. That does not make the argument wrong, and the dependency concern he describes is real, but the structure of the advice deserves to be visible to readers. The strongest part of his case is one he only half makes. Open weights, once downloaded, cannot be revoked. A proprietary model accessed through an API can be switched off by a government or a vendor, whereas a model whose parameters sit on European servers survives any political rupture. That is the genuine argument for open source as a sovereignty strategy, and it is more durable than arguments about cost or capability. It also cuts against his own framing, since it applies regardless of whether the weights originated in China or the United States. The industrial data point is the most actionable thing here. Europe’s comparative advantage is not in frontier model research, where it loses talent to better-funded US and Chinese labs, but in the manufacturing data its industrial base generates. Training specialised models on proprietary industrial datasets is a strategy that plays to an existing strength rather than attempting to catch up in a race already underway. The compute recommendation runs into a hard constraint he does not address. Building sovereign AI infrastructure requires substantial public and private capital at a moment when French yield premiums over German debt sit at euro-crisis levels, EU officials are warning member states about borrowing, and governments are cutting energy taxes rather than funding new programmes. The advice is sound and the fiscal room to act on it is narrow. Note also that other voices read Chinese open models as a risk rather than an opportunity, including former SEC chair Gary Gensler, so this is a contested question rather than a settled one.
What Next?
Watch whether European governments commit capital to domestic compute infrastructure, since that is the test of whether sovereignty rhetoric becomes policy, and the fiscal environment makes it difficult. European neocloud capacity and funding are the measurable indicator. On the open-source question, track adoption of Chinese open-weight models by European enterprises, which would show whether the dependency argument persuades buyers or whether security concerns dominate. Tsai claim that US labs have stopped publishing is verifiable and worth monitoring, since the rate of published research determines how quickly capability diffuses beyond the leading labs. The industrial data thesis would be validated by European manufacturers partnering with model developers on proprietary datasets.
Affected Tickers and Coins: BABA, NVDA, NBIS
Source: Bloomberg
