Citigroup: Alibaba Cloud Becoming the 'Chinese Google', Half of Revenue to Come from 'Selling Models' in Five Years

When a cloud computing company begins transitioning from chip manufacturing to model development and from selling computing power to selling "intelligence," a fundamental shift in revenue structure is taking place.

On May 11th, Citi Research released a comprehensive 60-page report, establishing a systematic forecasting framework for Alibaba Cloud's Model-as-a-Service (MaaS) business for the first time, and reiterating Alibaba as "China's top AI investment target," maintaining a buy rating, with an ADR target price of $205 and a Hong Kong stock target price of HK$204.

Analysts Alicia Yap and Nelson Cheung believe that Alibaba Cloud is building a vertically integrated AI system from chips to applications, highly similar to Google's "from TPU to Gemini" full-stack strategy.

More than half of the revenue will come from "selling models" in five years

Citi forecasts that Alibaba Cloud's AI-related revenue will increase from 24 billion yuan in FY2026 to 585.5 billion yuan in FY2031, with a 90% annual compound growth rate, and its share of total cloud revenue will jump from 15% to 70%.

The fastest growing is the MaaS business. Citi predicts that MaaS revenue will grow at an annual compound growth rate of 235%, from about 1 billion yuan in FY2026 to 438.6 billion yuan (approximately $62.6 billion) in FY2031, accounting for 53% of Alibaba Cloud's total revenue. In other words, by 2031, more than half of Alibaba Cloud's revenue will come from "selling model calls," rather than traditional servers, storage, and bandwidth.

Overall, Citi expects Alibaba Cloud's combined total revenue to grow at an annual compound growth rate of 39%, from 159 billion yuan in FY2026 to 833.4 billion yuan in FY2031, with revenue from external customers approaching the management's target of $100 billion in five years.

From chips to models: Why it's called "China's Google"

Citi cites NVIDIA CEO Jensen Huang's "five-layer AI cake" theory, pointing out that, aside from the energy layer, Alibaba has a presence in the other four layers of chips, infrastructure, models, and applications.

At the chip level, T-Head's (PingTouGe) main AI chip, Zhenwu 810E, is cited as being "comparable to NVIDIA H20" with a cumulative shipment of 470,000 units, serving over 60% to more than 400 external enterprise customers. At the infrastructure level, Alibaba Cloud operates 94 availability zones across 29 regions worldwide, holding a 33% share of the China IaaS public cloud market, ranking first. At the model level, the Qwen series has over 600 million cumulative downloads on Hugging Face, with more than 170,000 derivative models; the recently released Qwen3.6-Plus set a record on the OpenRouter platform by processing over 1 trillion tokens in a single day. At the application level, Alibaba is deeply integrating Qwen into Taobao, enabling users to browse, compare, and shop through AI conversations, covering over 4 billion items.

Explosive growth in tokens, price increase signals a return to pricing power

The underlying logic for the growth of MaaS is the explosive consumption of tokens. According to data from the National Data Bureau, as of March 2026, China's daily AI token usage exceeded 140 trillion times, growing over 1400 times from the beginning of 2024; in February 2026, China's weekly token usage surpassed that of the United States for the first time, becoming the world's largest token consumer.

The surge in demand has led to improved pricing power. Alibaba Cloud has increased the prices of AI computing products by 5% to 34% in April 2026, a move that Citi views as a positive sign of strong demand. Concurrently, Alibaba's management disclosed that the token consumption on the MaaS platform increased sixfold over the past three months.

Six years of capital expenditure of 822.9 billion yuan, "a once-in-a-century opportunity"

Citi forecasts that Alibaba's cumulative capital expenditure from FY2026 to FY2031 will reach 822.9 billion yuan, equivalent to an average of 32% of cloud revenue. However, as economies of scale take effect, the proportion of capital expenditure to cloud revenue will decrease from 83% in FY2026 to 17% in FY2031, presenting a significant room for margin