Goldman Sachs has released a research report stating that it is encouraged by Alibaba-W (09988)'s progress on its exclusive full-stack AI capabilities, namely AI infrastructure (targeting expansion of global data center capacity to 20GW by 2032, supported by Alibaba Cloud and partners, with self-developed Pingtouge chips making significant contributions to computing power), AI models (the Qwen series roadmap aims to remain at the top level, adopting a multimodal approach, with the Qwen 4 series launching soon and Qwen 4.5/5 versions expanding to next-generation parameter scales of 5-10 trillion), and AI applications/ecosystem (focusing on the agent era's "Harness x Context x Model," connecting AI work agents with DingTalk through new "Qwen Office" and "Enterprise Context" to improve productivity and scalability).

The main points from Goldman Sachs are as follows: The bank maintains its cloud revenue year-over-year growth forecasts of 53%/55%/55% for the fiscal quarters ending September 2026/December 2026/March 2027, still above market consensus estimates (Visible Alpha). It expects investors will focus on the capital expenditure/funding sources supporting the 2032 20GW total capacity target, which the bank believes is within manageable scope through a combination of asset-light and asset-heavy models: 1) capital expenditure, with funding sources including stable e-commerce EBIT, growing cloud business profits, cash on the balance sheet, and future financing; 2) partnerships with data centers/telecom operators/other parties; 3) leasing, particularly overseas AI computing power.

Synergistic Capacity/Capital Expenditure/Cloud Business Growth Targets. The bank estimates that achieving the 20GW capacity target by calendar year 2032 implies adding 1-2GW per year over the coming years from a base of approximately 5-6GW at the end of 2025 (Goldman Sachs estimate), meaning Alibaba's annual capital expenditure in the coming years will reach RMB 200-300 billion, assuming capital expenditure of approximately USD 20-25 billion per GW, based on a model primarily using Pingtouge self-developed chips combined with asset-light and asset-heavy plus leasing (operating expenditure) computing power. The bank believes that the unit GW cost of self-developed Pingtouge chips has unique advantages over domestic and international peers, and in the medium term, Pingtouge's contribution to Alibaba Cloud computing power is expected to rise from the current approximately 10% to 50%. Meanwhile, the company still expects external AI cloud revenue to reach USD 100 billion by calendar year 2030, with a medium-term EBIT margin of 20%, depending on the medium-term conversion of capital expenditure to revenue and the evolution of the internal/external revenue structure. The bank currently forecasts capital expenditure of RMB 209 billion/243 billion/260 billion for fiscal years 2027/28/29 respectively.