From commodity to currency: How Alibaba is monetizing the AI moment

Structural overhaul, price hikes, and a dominant open-source model family signal that Alibaba is done playing infrastructure catch-up–it wants to own the AI stack

Alibaba has spent the last several years building the infrastructure. Now comes the bill. Cloud revenue at the Hangzhou-based tech giant grew 36% year-on-year in the December quarter, reaching RMB 43.3 billion (approximately US$6.2 billion), with AI-related product revenue logging its tenth consecutive quarter of triple-digit growth.

Morgan Stanley and HSBC analysts, as reported by the South China Morning Post, are projecting growth to accelerate to around 40% in the March quarter, driven by a combination of surging token usage, price increases across cloud services, and the most sweeping internal reorganization Alibaba has undertaken in years.

The momentum is not accidental. It is the result of a deliberate pivot, from treating cloud as a utility to treating AI as a product.

The price of demand

In late March, Alibaba Cloud announced it would raise prices for services running on its in-house T-Head AI chips by between 5 and 34%, effective April 18. Cloud Parallel File Storage pricing was simultaneously increased by 30%. The move was framed as part of a broader industry trend–global cloud providers have been recalibrating fees as AI workloads strain capacity and, frankly, as market power begins to shift back toward suppliers–but the timing and scale of Alibaba's increases say something more deliberate about where the company believes it stands.

Morgan Stanley analysts, in a research note last week, attributed the anticipated growth acceleration to "a robust surge in token usage." HSBC analysts added that price increases, expansion among Chinese enterprise clients, and the growing contribution of in-house chip infrastructure were expected to collectively improve long-term cloud margins.

These are not opposing forces; they reinforce each other. Higher token consumption justifies pricing power; pricing power funds model development; better models drive more token consumption.

The loop is closing.

Restructuring around the AGI bet

The organizational changes Alibaba has made in recent weeks are as telling as the financial figures. In mid-March, the company formed the Alibaba Token Hub (ATH) business group, consolidating five previously separate AI units–Tongyi Lab, the MaaS business line, Qwen, Wukong, and AI Innovation–under CEO Eddie Wu's direct leadership. The mandate, as Wu put it in an internal memo reviewed by Bloomberg, is to "create tokens, deliver tokens, and apply tokens."

That framing is worth sitting with. Alibaba is not describing this as an AI strategy. It is describing it as a token economy, one in which the fundamental unit of value is the inference output of a model, and where the company's role is to produce and monetize that output at scale.

Wu's accompanying internal letter was explicit about the stakes: "We are standing at the threshold of an AGI inflection point. Billions of AI agents are poised to take on an ever-greater share of digital work, each powered by tokens generated by models, and these agents will increasingly become the primary interface between people and the digital world."

The restructuring carries weight beyond the philosophical. It arrived after notable internal friction–three senior AI executives departed Alibaba earlier this year, including Lin Junyang, who left as head of the Qwen division in March. Shortly after, the company also announced a separate CEO-led Alibaba Group Technology Committee, bringing together senior technical leadership with Zhou Jingren as chief AI architect and Fei-Fei Li appointed as CTO of Alibaba Cloud.

The dual-structure approach–one body focused on commercial AI execution, another on technical governance–suggests Alibaba is trying to move faster without losing architectural coherence. Whether the restructuring delivers is a question for the next few earnings calls. T

The first real test came on March 19, when the December quarter results showed that 36% cloud growth figure. The question is whether Wu's centralized control can push that trajectory higher still.

Qwen: The open-source flywheel

Alongside the infrastructure and organizational plays, Alibaba has quietly built something that is now difficult to ignore in the global AI conversation: dominance in open-source model downloads.

As of March, the Qwen model family had reached approximately 942 million cumulative downloads, with 153.6 million in February alone, more than double the combined total of the next eight players, including Meta's Llama and DeepSeek, according to research by Interconnects AI citing Hugging Face data. Alibaba Cloud captured over 50% of global open-source model downloads following the release of the Qwen 3.5 series in February.

The Qwen 3.5 flagship model–a 397-billion parameter mixture-of-experts architecture with 17 billion active parameters–is reportedly up to eight times faster and around 60% cheaper than its predecessor, making it highly attractive to developers building cost-sensitive enterprise applications. It competes on benchmarks against GPT-5.2 and Claude Opus 4.5, though the more practically significant detail is the pricing: at approximately US$3.60 per million tokens on OpenRouter, it represents a fraction of the cost of many Western alternatives.

This is not a coincidence of open-source idealism. It is a deliberate funnel. Free and widely adopted open-source models build developer mindshare and ecosystem depth; enterprise adoption of those models drives traffic back to Alibaba Cloud's managed services, where the actual monetization occurs. The more developers build on Qwen, the more they eventually pay for inference infrastructure, fine-tuning services, and the agentic platform layer Alibaba is now assembling through Wukong.

Airbnb CEO Brian Chesky has said publicly that his company "heavily relies on Qwen." NVIDIA's Jensen Huang has acknowledged the model family's growing dominance in the open-source space. Over 100,000 derivative models have now been built on top of Qwen.

The US$100 billion target and what it requires

Last month, Alibaba stated a target that would have sounded implausible two years ago: US$100 billion in annual external revenue from its combined cloud and AI businesses within five years. For context, the Cloud Intelligence Group posted approximately US$24.8 billion in annualized revenue based on its most recent quarterly figures.

Getting to US$100 billion requires compound annual growth well above current rates, sustained over half a decade, while competition in China's cloud market from Tencent, Huawei, and ByteDance remains intense, and US export controls continue to constrain access to advanced chips.

The path, as Alibaba has now clearly defined it, runs through three parallel tracks: infrastructure pricing power supported by in-house silicon; enterprise AI penetration through Qwen and the Wukong platform; and the monetization of a developer ecosystem that is, by volume, already the largest in open-source AI.

Citic analysts said last week they expect "acceleration in cloud revenue to continue and faster-than-expected loss reduction from Shangou"––Taobao Instant Commerce––to support a broader rebound in China e-commerce earnings. HSBC analysts noted that if the instant commerce business narrows losses as expected, freed-up capital can be redirected toward AI model training and Qwen user acquisition, compressing the timeline between adoption and revenue.

What this means for enterprise buyers

The price increases are real, but context matters. A 34% rise on AI chip services that already undercut most Western alternatives still leaves Alibaba Cloud competitive, especially across APAC, where procurement decisions live and die on cost-per-token math.

The more pointed signal is Wukong. As the only newly created unit within Token Hub, it marks where Alibaba's ambitions are actually aimed: not the model layer, not the infrastructure layer, but the application layer, where enterprise contracts get signed. That's the layer where the real competition in APAC enterprise AI is about to be fought.

Alibaba isn't just selling cheaper compute anymore. It's coming for the workflow.