In two years, Alibaba has gone from e-commerce giant in regulatory crisis a global powerhouse in open source AI. Its family of Qwen models has accumulated 400 million downloads and 140,000 derived models arising from it, figures that surpass any Western competitor, except one: Llama, from Meta.
Why is it important. Eddie Wu, CEO of Alibaba, has openly said that Qwen aspires to be “the Android of the AI era.” It’s not marketing: the company has released 357 models in less than two years, a pace that neither OpenAI nor Google maintain nor want to maintain in their public version.
The strategy copies Google’s manual with Android: giving away the operating system to dominate the infrastructure that supports it. Only this time the dominant player doesn’t come from Silicon Valley.
The context:
- Alibaba poured more than $800 million into the country’s top four AI startups —Moonshot, Baichuan, Zhipu, MiniMax— before realizing its own technology could lead the market.
- Now those investments have stopped because the bet is on home.
The road has been brutal. Between 2020 and 2022, Alibaba lost half of its stock market value due to the Chinese government’s regulatory offensive. DAMO Academy, its research arm, fired 30% of his staff.
Some key scientists such as Yang Hongxiacreator of M6; either Zhou Changtechnical leader of Qwen, left for other companies such as ByteDance. The brain drain left a crater. Even so, Alibaba has managed to get back on its feet.
The unexpected turn. In January 2025, DeepSeek launched R1an open source reasoning model that rivaled the o1 by OpenAI then. It had rapid and global adoption, and Alibaba seemed to lose steam. Joe Tsai, president of Alibaba, admitted the hit: “We said, ‘How is it possible that they got ahead of us?'”
The response was drastic. On the first day of the Chinese New Year—a sacred holiday in China—the AI team canceled his vacation and launched Qwen 2.5 Maxsurpassing DeepSeek V3. The war was not over.
In figures. Alibaba has promised to invest 380 billion yuan ($53 billion) over three years in AI and cloud infrastructure. It’s more than it spent in the entire previous decade.
In the second quarter of 2025, AI-related revenue grew by triple digits for the seventh consecutive quarter. Alibaba Cloud increased sales 26% year-on-year. The stock is up more than 90% so far this year. a rocket.
The strategy. Free models generate demand for GPUs and training. That demand is monetized on Alibaba Cloud. It is the Microsoft-OpenAI model for Azurebut here Alibaba is an investor and beneficiary.
Unlike Amazon (without competitive open source models), Google (closed models) or Meta (without its own cloud), Alibaba unites open model, cloud and developer ecosystem. And it’s already the fourth world in cloud behind the Amazon-Microsoft-Google trident.
Of course, Alibaba opted for the wrong architecture while OpenAI scaled with GPT. When ChatGPT took off in December 2022, it had to admit the mistake.
- In August 2023 he opened the Qwen code just as Llama 2 crashed in Chinese. He filled that space immediately.
- In February 2025, Apple chose Alibaba as a partner for Apple Intelligence in China. Beastly validation of the then most valuable company in the world.
Months later, Wu detailed the roadmap: expand the context, from 1 million to 100 million tokensand scale from one billion to ten billion parameters. Quantifiable bets with assigned budget. ByteDance is the only rival that keeps the pulse.
Missing. The western market. Qwen dominates Asia but has not penetrated Europe or America like Llama has (and it is not like it has swept mobile phones like Android). Besides…
The big question. Can a Chinese company that has endured regulatory purges and talent drains become the global standard for open source AI?
What remains to be seen is whether the West agrees to rely on an AI operating system designed in Hangzhou.
In Xataka | ‘World models’ aim to be the next great revolution in AI: this is what robotics needs to look like movies
Featured image | Xataka
GIPHY App Key not set. Please check settings