Key takeaways
- Wu Yongming (吴泳铭), one of Alibaba's 18 co-founders, returned as Group CEO in 2023 and chairs its Cloud Intelligence Group.
- He committed over 380 billion yuan (≈ US$53 billion / HK$414 billion) to AI and cloud infrastructure over three years—more than Alibaba spent in the prior decade.
- Under him, Alibaba open-sourced 300+ Qwen (通义千问) models; downloads passed 600 million and derivative models exceeded 170,000.
- He reframed strategy around "ASI" (super artificial intelligence) and "AI as the next operating system."
- Honest gap: the consumer-app pivot (from Quark to Qianwen) is recent and unproven, and heavy capex pressures margins.
The co-founder who came back to run the AI bet
At the 2025 Apsara Conference, the man on stage was not Jack Ma but a soft-spoken engineer who had been in the room at Alibaba's founding in 1999.
Wu Yongming (吴泳铭), nicknamed "Wu Ma" internally, is one of Alibaba's original 18 founders and now serves as Group CEO and Chairman of the Cloud Intelligence Group. After years away from the spotlight, he took the CEO role in September 2023 and declared two priorities: "user first, AI-driven."
He has since reorganized the company around AI, moving the Tongyi (通义) app team, merging hardware efforts, and pushing every business unit to embed large model (大模型) capabilities into its products.
The US$53B infrastructure wager
In February 2025, Wu announced Alibaba would invest more than 380 billion yuan (≈ US$53 billion, about HK$414 billion) in AI and cloud hardware over three years. That sum exceeds Alibaba's total AI-and-cloud spend across the previous ten years, according to Xinhua's reporting.
The thesis is simple to state and hard to execute: build the "compute highway" first, then let applications run on it—what he calls "AI as the next operating system." Alibaba Cloud revenue has responded: AI-related product revenue grew triple digits for nine straight quarters into 2025.
Wu has also sketched a long-range energy plan: compared with 2022, Alibaba's global data-center power consumption could rise roughly tenfold by 2032, a bet that AI compute will become a utility-scale load rather than a niche one. The scale of that commitment is what distinguishes his approach from rivals that are waiting for demand to arrive.
Open-sourcing Qwen as the "Android of AI"
Alibaba has open-sourced 300+ models under the Qwen (通义千问) family since the first release in 2023. Official figures cite over 600 million downloads and more than 170,000 derivative models—claiming the world's largest open-source model family.
Wu positions Qwen as the "Android of the AI era," giving global developers a free base to build on. He also outlined an "ASI" roadmap: from knowledge emergence to autonomous action to self-iteration, with AGI as merely the starting point rather than the finish line.
For developers outside China, the practical upside is licensing. Qwen's open weights can be fine-tuned and self-hosted without per-call fees, which has made it a default base model in many emerging markets where API dollars are scarce. That reach is part of why Wu treats openness as strategy, not charity.
Reading the ASI roadmap
Wu's framing splits the road to super-intelligence into three stages. The first, "intelligence emergence," is where models learn from vast human knowledge. The second, "autonomous action," is the current phase, where AI uses tools and writes code to assist people. The third, "self-iteration," is his endpoint: AI connected to the physical world that learns on its own and surpasses humans.
Whether that endpoint is reachable is contested, but the staging helps explain why he is willing to spend first and monetize later. If models truly become a layer every device runs on, owning the open base is worth a decade of infrastructure pain.
The consumer-app pivot
Not every move has been smooth. Alibaba first anointed Quark (夸克) as its consumer-AI entry point, then in late 2025 shifted to the Qianwen (千问) App, rebranding the former Tongyi App. Early reports claim Qianwen reached tens of millions of monthly active users within weeks, but those numbers are recent and unverified by independent trackers.
The pivot reflects a harder truth: owning the best model does not guarantee owning the user. Wu's larger bet is that an open, widely adopted Qwen ecosystem will matter more than any single app, because the operating system, not the first killer app, captures the platform.
Honest limitations
- The US$53B figure is Alibaba's stated plan; actual spend depends on demand and may be adjusted.
- Qwen download and derivative counts are company-reported and not independently audited.
- The consumer-AI app strategy shifted from Quark (夸克) to Qianwen (千问) in late 2025; early user-growth claims need time to validate.
- Heavy capital expenditure weighs on near-term profitability—a trade-off readers should weigh.
What readers can do now
- Developers can build on Qwen's open weights today; it is among the most permissive large model (大模型) families from a major Chinese cloud.
- Investors should watch Alibaba Cloud's AI-revenue growth as a real-economy signal distinct from consumer hype.
- Enterprises evaluating China AI should compare Qwen's open ecosystem against closed US models on cost and compliance.