Artificial intelligence is beginning to look like a profitable business for Alibaba. It is also proving to be an expensive one.
The Chinese technology giant’s latest earnings offer a glimpse of both realities. In the quarter ended September 30, revenue from its AI cloud and compute business rose 53% year-on-year to RMB 61 billion (USD 9.1 billion), accelerating from 44.9% growth in the preceding quarter. Adjusted earnings before interest, taxes, and amortization (EBITA) more than doubled to RMB 8.8 billion (USD 1.3 billion), bringing the segment’s margin to 14.5%.
The figures, released November 19, suggest that years of investment in AI are beginning to produce more substantial commercial returns. But they also reveal a growing divide within Alibaba’s AI business. The infrastructure that other companies pay to use is becoming increasingly profitable. The models and applications that Alibaba is developing for itself remain costly to operate.
The company has spent recent years assembling what it calls a full-stack AI ecosystem. Rather than relying on other suppliers, Alibaba wants to design the chips, provide the computing power, develop the models, and build the applications that people and businesses ultimately use.
At the foundation of this effort is T-Head Semiconductor, Alibaba’s chip design arm. Its processors are designed to handle various computing workloads, including the demanding calculations required by AI systems. The company is developing its Zhenwu series of AI chips, with the next-generation V900 scheduled for commercial release in the first quarter of 2027.
T-Head may soon become a more visible business in its own right. On November 10, Alibaba disclosed that the chipmaker had confidentially filed for a listing in Hong Kong. The proposed spinoff would give T-Head independent access to capital markets and allow investors to value its business separately, although Alibaba expects to retain control.
Above the chips sits Alibaba Cloud, which supplies the processing power, storage, and software services that businesses need to train and operate AI systems.
Then comes Qwen, Alibaba’s family of AI models, which underpins everything from enterprise software to consumer assistants. QwenWork, for instance, is designed to automate business tasks, while the Qwen app offers assistance with shopping, travel, and everyday activities.
The idea is to make these businesses reinforce one another. Customers pay to use Alibaba’s computing infrastructure, while its models and applications encourage more people to use that infrastructure. The more AI becomes part of everyday business, the more computing power those businesses may need.
For now, the clearest returns are coming from the cloud.
Revenue from external customers in Alibaba’s “AI cloud and compute services” segment increased 64% year-on-year. Recurring revenue, which excludes compute equipment solutions, grew even faster, at 67%, indicating greater demand for cloud services purchased through subscriptions or usage-based arrangements.
Revenue from AI-related products reached RMB 17.9 billion (USD 2.7 billion).
Demand is also extending beyond companies developing large AI models. Alibaba said businesses in automotive, financial services, biotechnology, and embodied intelligence are expanding their use of its infrastructure.
Move further up Alibaba’s technology stack, however, and the economics become less encouraging.
Its “AI labs and applications” segment generated RMB 4.4 billion (USD 657.5 million) in quarterly revenue, up 51% year-on-year. Yet its adjusted EBITA loss widened to RMB 10.5 billion (USD 1.6 billion), from RMB 3.4 billion (USD 508.1 million) a year earlier.
Alibaba attributed much of the increase to continued investment in AI and higher inference costs associated with the Qwen app.
This creates an uncomfortable feature of the AI business. More users are not necessarily an immediate financial advantage. Every interaction requires computing resources, and those costs can accumulate well before the company establishes a reliable way to make money from its users. For Alibaba, the same growth in AI adoption that benefits its cloud business can therefore weigh on the profitability of its applications.
Group CEO Eddie Wu has argued that bringing chips, infrastructure, models, and applications together allows advances in one area to strengthen the others. It is a strategy that offers Alibaba greater control over its technology, but requires substantial investment before all its components can produce meaningful returns.
In the September quarter, Alibaba spent RMB 50.5 billion (USD 7.5 billion) on capital expenditure, 60% more than a year earlier, though less than the RMB 67.7 billion (USD 10.1 billion) recorded in the June quarter.
Free cash flow remained negative, with an outflow of RMB 34.5 billion (USD 5.2 billion), compared with RMB 21.8 billion (USD 3.3 billion) a year earlier. Alibaba attributed this primarily to spending on cloud infrastructure, partly offset by lower losses in quick commerce and increased customer prepayments for cloud services.
There is, meanwhile, another consideration that cannot be addressed simply by spending more money: access to technology.
Alibaba has become entangled in the controversy surrounding Nvidia and Singapore-based cloud computing company Megaspeed, which has faced allegations concerning the possible diversion of advanced American chips to China.
On November 12, The Straits Times reported that Singapore corporate filings showed Alibaba subsidiary Apex Enterprise Solutions had provided financing to Speedmatrix, a Malaysian company linked to Megaspeed. Separately, Reuters reported the same day, citing people familiar with the matter, that Alice Huang, a former Megaspeed executive linked to the Nvidia chip controversy, had business and investment ties with Alibaba and other Chinese technology companies, including Tencent.
A week later, on November 19, Reuters reported that Alibaba was named in a proposed shareholder class action in New York. The lawsuit alleged that the company had failed to adequately disclose risks arising from its Megaspeed connections and potential violations of US export controls. The allegations have not been established in court.
The dispute comes at a time when Washington is seeking to make advanced American computing technology harder for Chinese companies to obtain.
On November 16, the US announced new export controls targeting certain overseas data center arrangements that allow companies in China and other restricted jurisdictions to access advanced computing power without importing the chips themselves.
During Alibaba’s November 19 earnings call, Wu acknowledged that the restrictions could affect its products and services in 2027. He also warned that 2027 and 2028 could be particularly difficult years for computing capacity in China, as demand runs up against constrained supply.
That gives Alibaba’s investments in its own chips and domestic computing infrastructure significance beyond financial returns: what might otherwise appear to be an expensive attempt to control more of the AI business could transpire as a way to reduce dependence on technology whose availability is increasingly subject to political decisions.
Note: RMB figures are converted to USD at rates of RMB 6.69 = USD 1 based on estimates as of November 20, 2026, unless otherwise stated. USD conversions are approximate and, where appropriate, rounded for ease of reference. They may not fully match prevailing exchange rates.