Alibaba Earnings Call: AI ARR Tops RMB49.5 Billion as Cloud Capex Targets Three-Year Payback

Key Takeaways

• Alibaba’s AI-related products surpassed RMB49.5 billion ($7.3 billion) in annualized revenue, representing 35% of Alibaba Cloud’s external revenue.
• AI Cloud and Compute Services revenue rose 45% year over year, while adjusted EBITA increased 133% and the segment’s margin reached approximately 12%.
• Alibaba’s Zhenwu M890 AI chip is entering broader commercial deployment after the Zhenwu product family attracted more than 650 external customers.
• Management expects AI computing investments to recover their cost in roughly three years, potentially shortening to 2.5 years as margins improve and proprietary chips replace third-party hardware.

Alibaba Group said its artificial-intelligence business is moving beyond heavy investment toward a model capable of funding more of its own expansion, as cloud revenue accelerates and proprietary chips enter wider commercial use.

During the company’s fiscal first-quarter 2027 earnings call on August 20, Chief Executive Eddie Wu said the annualized revenue run rate, or ARR, of AI-related products had exceeded RMB49.5 billion, equivalent to roughly $7.3 billion. Those products now account for 35% of Alibaba Cloud’s external revenue and carry significantly higher gross margins than the cloud portfolio average.

For the quarter ended June 30, AI-related product revenue reached RMB12.38 billion and delivered triple-digit year-over-year growth for the 12th consecutive quarter. Alibaba’s AI Cloud and Compute Services revenue increased 45% to RMB48.44 billion, its fastest expansion in 22 quarters. Adjusted EBITA for the segment more than doubled to RMB5.63 billion, lifting the margin to around 12%. Alibaba’s official results show that AI demand is contributing not only to sales growth but also to improving cloud profitability.

AI Commercialization Gains Momentum

Alibaba said customers are adopting services across the entire AI stack, including model APIs, AI agents, training and inference capacity, cloud infrastructure and proprietary processors.

Annualized revenue from model and application services, including model-as-a-service products, has exceeded RMB16 billion. Management expects demand for computing capacity to continue outpacing available supply, supporting utilization rates and pricing across Alibaba Cloud’s infrastructure.

The results indicate that Alibaba’s AI strategy is beginning to produce a measurable commercial return. Rather than relying solely on access to the Qwen model family, the company is earning revenue at multiple layers, from hardware and data centers to model services and enterprise applications.

Alibaba reported total quarterly revenue of RMB268.95 billion, up 9% from a year earlier. However, net income attributable to shareholders fell 76% to RMB10.54 billion as the group increased AI spending and recorded lower investment-disposal gains. Capital expenditure rose 75% to RMB67.68 billion, highlighting the near-term cost of the strategy. The company’s quarterly figures nevertheless showed that cloud growth remained the strongest part of its business.

T-Head Chips Move Into Commercial Scale

Alibaba is also expanding the commercialization of chips developed by its semiconductor unit, T-Head.

The company said more than 650 external customers across over 20 industries were using its Zhenwu processors as of early August. Its latest Zhenwu M890-based SuperNode instance has now launched on Alibaba Cloud, with supply expected to increase throughout the second half of the year.

Alibaba said the M890 system can support inference workloads for models containing more than two trillion parameters. Models including Kimi K3 and Alibaba’s Qwen3.8-Max are already being served through the infrastructure.

T-Head has developed a broader portfolio covering AI accelerators, central processing units and networking chips. That internal supply chain could become increasingly important as Alibaba expands data-center capacity while managing the cost and availability of externally sourced processors.

Management also said the delivery time for large-scale AI data centers had been reduced to around 100 days. Meanwhile, the Qwen model family has surpassed three billion global downloads and generated more than 300,000 derivative models, giving Alibaba a large developer ecosystem that could eventually convert into additional cloud demand.

Alibaba Sees a Three-Year Return on AI Infrastructure

Chief Financial Officer Toby Xu stressed that Alibaba’s AI business requires capital expenditure before revenue can be generated. Model subscriptions, API usage, GPU rentals and training or inference services all depend on computing infrastructure being installed first.

Alibaba has spent approximately RMB190 billion under the RMB380 billion three-year AI and cloud investment program announced in February 2025. Management said progress remains broadly in line with the plan, although quarterly spending can fluctuate because of hardware delivery schedules, CPU purchases and changing component prices.

Based on current utilization and margins, the company estimates that AI computing assets can recover their initial investment in about three years. Because the equipment generally remains useful for longer, those assets could continue generating positive free cash flow after the payback period.

Xu noted that Nvidia V100 processors purchased in 2018 and A100 chips acquired in 2020 remain close to full utilization. This suggests the economically productive life of AI hardware can extend beyond conservative depreciation assumptions.

Management believes the payback period could decline toward 2.5 years—and potentially closer to two years—if AI product margins continue rising. Greater use of T-Head chips could further improve profitability by reducing reliance on more expensive commercial processors.

The spending is still exerting pressure on cash generation. Alibaba recorded a quarterly free-cash-flow outflow of RMB44.67 billion, compared with an outflow of RMB18.82 billion a year earlier, primarily because of increased cloud infrastructure expenditure.

APIs Are Only an Interim AI Business Model

Wu said the current reliance on API calls should be considered a transitional stage rather than the final form of AI monetization.

As AI systems become more capable, Alibaba expects them to participate directly in research, product development, customer service and corporate operations. In that scenario, the economic value generated by AI could exceed the revenue currently captured through model-access fees.

Alibaba’s strategy is therefore designed around a closed commercial loop: expanding AI demand supports cloud usage, cloud growth justifies additional infrastructure, and proprietary chips lower the cost of supplying that capacity.

Quick Commerce Targets Profitability in Fiscal 2029

Outside AI, Alibaba reorganized its commerce operations under the Alibaba E-commerce Group. The division generated RMB205.86 billion in quarterly revenue, up 4%.

China Quick Commerce revenue climbed 45% to RMB53.30 billion, driven by Taobao Instant Commerce and Freshippo. Management said order economics improved and losses narrowed during the quarter.

Alibaba expects non-food quick-commerce transaction volume to exceed food orders in the next fiscal year. The company is targeting overall profitability for the business in fiscal 2029 and believes it could eventually contribute about 30% of total platform gross merchandise value.

Internationally, AliExpress achieved an operating profit during the quarter, although tariffs and geopolitical uncertainty continued to affect the broader international commerce business.

The Alibaba earnings call ultimately presented AI as both the company’s largest investment requirement and its clearest growth engine. Faster cloud expansion, higher-margin AI products and wider deployment of T-Head chips support the case for improving returns. The remaining test is whether Alibaba can sustain that momentum while limiting cash outflows and converting its infrastructure spending into durable free cash flow.