Alibaba Qwen: the most downloaded AI model, but without profitability

The rise of Qwen

Alibaba launched its Qwen model family in 2023 and published them with open weights almost immediately. That made them a perfect alternative for local use and fine-tuning in all kinds of scenarios. In January 2026, Qwen was already the most downloaded open AI model in the world, with nearly one million daily downloads according to Hugging Face data (Source: AI Base). This milestone is no coincidence: since its appearance, Qwen has accumulated over 100 million cumulative downloads, surpassing models like Meta's Llama and Mistral. Alibaba's strategy of offering models in multiple sizes (from 1.8B to 72B parameters) and multilingual support, especially in Chinese and English, has been key to its global adoption. Additionally, the Qwen ecosystem includes tools like Qwen-Agent and Qwen-VL, which facilitate its integration into enterprise applications. However, success in downloads has not translated into significant revenue.

Popularity ≠ revenue

In the first quarter of 2026, Alibaba indicated that it had obtained $1.3 billion in AI-related revenue, barely 4% of its total turnover. The figure falls far short, especially considering that Alibaba plans to invest $55 billion in AI infrastructure by the end of 2027 (Source: New York Times). To put it in context, Alibaba's AI revenue represents less than 0.5% of its market capitalization, while companies like Microsoft generate more than 10% of their revenue from Azure AI and Copilot. Additionally, Alibaba Cloud, which hosts the Qwen models, reported only 3% revenue growth in the last quarter, well below AWS's 20%. Direct monetization of Qwen is almost nil: the models are free to download and use, and Alibaba only charges for API usage on its cloud, but at very competitive prices to gain market share. This has led to extremely tight margins.

Uneasy investors

Alibaba's shares have fallen 37% on the Hong Kong stock exchange this year. Investors are clearly concerned about the lack of return on this AI bet, something also observed in the US market. Alibaba's strategy resembles Google's with Android: seeking to dominate the open ecosystem and then monetize indirectly, but the path is uncertain. Android took nearly a decade to generate significant revenue through advertising and services, and Alibaba faces a more complex scenario: AI requires massive investments in GPUs and data centers, with operating costs exceeding $10 billion annually. Additionally, the price war in China, led by DeepSeek and Baidu, has reduced AI API margins to nearly zero. Alibaba also competes with Tencent and ByteDance, which are aggressively investing in their own models. Regulatory uncertainty in China, with potential restrictions on exporting open models, adds more pressure.

What consequences will it have?

If Alibaba fails to monetize Qwen, it could be forced to reduce its investments or change its open-source strategy. A plausible scenario is that Alibaba limits access to its largest models (Qwen-72B) only through its cloud, while keeping the small ones open to attract developers. Meta already did this with Llama 3.1, offering free versions but charging for commercial use. Another option is to follow Red Hat's model: sell support and enterprise services around Qwen. However, the AI support market is still nascent. Its competitors, like DeepSeek or Baidu, also face similar challenges. DeepSeek, for example, has raised $5 billion but is not yet profitable. The market watches closely whether the open AI bet can generate sustainable long-term revenue. History shows that open-source software can be profitable (as demonstrated by MongoDB or Elastic), but it requires time and a mature ecosystem. Alibaba has the advantage of its huge user base in e-commerce and cloud, but AI monetization remains elusive. If it does not achieve results in the next two years, investors could force a strategy change, which could slow down open AI innovation globally.