Alibaba has shifted its strategic focus on AI development, moving from pursuing an open-source ecosystem to commercial monetization. Core members of the Qwen team, including Lin Junyang and Hu Binyuan, left due to strategic disagreements. Former Alibaba Cloud CTO Zhou Jingren took over, and CEO Wu Yongming established the "Alibaba Token Hub" and formed an AI Strategy Committee, clearly aligning model development with cloud business revenue goals. The company has made a decisive turn, prioritizing MaaS and commercialization.

This transformation logic is a very correct choice at present. Although Qwen's open-source efforts have gained praise from developers worldwide, the model itself does not generate profit. The main source of Alibaba Cloud AI revenue still comes from selling GPU computing power. MaaS accounts for a small share and yields thin margins. Even with top-tier model capabilities and large GPU resources, the profits are modest, which is unsustainable and destined to fail commercially.

After the reform, Alibaba's AI strategy has become highly aligned with ByteDance's approach. Doubao has always been closed-source from the start. Volcano Engine was built around AI monetization from the beginning. It leads globally in token call volume and has pioneered token monetization, even though its model capabilities are not the top, its monetization rate is not inferior to Qwen.

The cost of this transformation is that Lin Junyang, the soul of Qwen's open-source ecosystem, left. His departure could shake community confidence and trigger a chain of talent loss. More critically, competitors like MiniMax and Zhipu have already surpassed Qwen in code generation, putting pressure on model capabilities. At this point, switching to closed-source could backfire if the product strength isn't enough, as customers will turn to competitors. Meanwhile, ByteDance's Volcano Engine is growing rapidly, having already preempted the cloud sales driven by token consumption. Its gross profit from cloud services surpasses Alibaba Cloud's GPU sales by a large margin.

Therefore, looking at Alibaba's future AI direction, it is highly likely they will choose a path of "closed-source flagship models tightly integrated with cloud services," similar to Microsoft's Azure + OpenAI model or Anthropic + AWS. Proprietary models can generate higher gross margins, rather than wasting resources on selling GPU computing power.

At the same time, AI applications will be prioritized in e-commerce scenarios. Open-source efforts will not be completely abandoned; Alibaba can emulate Google—keeping top-tier models closed-source and for internal use, while open-sourcing smaller models. The recent explosion of Agents has fully unlocked AI monetization channels. The token consumption driven by Agent AI far exceeds traditional chat. Alibaba's corporate DNA lies in B2B adoption rather than C-end users. If they can establish a foothold on the enterprise-level Agent platform, the MaaS ceiling will significantly rise, and profit margins will improve markedly.

To give a more relatable example: computing power is like a gold mine. Originally, Alibaba leased tools and mines to others and also open-sourced mining methods, earning rent and tool fees from the mines. But when gold prices rose, Alibaba decided to go into mining themselves.

In summary, the direction is correct. Success or failure depends on whether model capabilities can be rebuilt to restore competitive advantage. Currently, the situation is somewhat similar to Kimi's past—it's a matter of doubling down on R&D. Kimi's transformation involved halting deployment and shifting from closed-source to open-source models. But for large companies, maintaining flagship closed-source models is actually the right path, as they have enough computing power to monetize and tokenize effectively. Additionally, whether Alibaba's e-commerce and cloud synergy can truly complete the business closed loop—and then sell this solution to B2B clients—is crucial.

At this point, one must admire ByteDance. From the start, they had a clear plan: avoid blindly following the open-source trend, instead stockpiling tokens, developing proprietary applications, and self-using tokens. Once the system was running smoothly, they sold this entire solution through Volcano Engine to B2B clients, completing the business closed loop. The AI transformation and organizational restructuring are simply a reflection of ByteDance's pre-set strategic blueprint.