Alibaba Group (BABA) Alibaba Cloud APSARA Key Takeaways

Equity Research 22 September 2026 | 4:48PM HKT

Alibaba Group (BABA): Alibaba Cloud APSARA - Key Takeaways; Powering & Scaling for the Agentic era

We highlight key takeaways from the Alibaba Cloud APSARA Conference (“云栖大会”) Ronald Keung, CFA | held in Hangzhou, across Chairman and CEO keynote speeches that focused on Goldman Sachs (Asia) L.L.C. powering and scaling towards Machine Intelligence/Artificial Super Intelligence (ASI) Timothy Zhao for the agentic AI era, an analyst meeting with Group Chairman and management, | and sessions hosted by respective AI Cloud & Compute and AI Labs & Applications Goldman Sachs (Asia) L.L.C.

management/business heads. We are encouraged by Alibaba’s progress across its Steve Qiu | unique AI full-stack, namely AI infrastructure (targeting to reach 20GW in total data Goldman Sachs (Asia) L.L.C.

center power consumption by 2032, to be powered by Alibaba Cloud & partners, Damian Xie | with its custom silicon T-head to contribute meaningfully to compute), AI models Goldman Sachs (Asia) L.L.C. (with Qwen’s roadmap in sustaining amongst the top-tier, a multi-modal approach, Iris Xiao Qwen4 family coming soon and Qwen4.5/5 to scale towards next 5-10T parameter | Goldman Sachs (Asia) L.L.C. sizes) and AI applications/ecosystems (focusing on Harness x Context x Model for the agentic era, with new QwenWork functionalities and ‘Enterprise Context’, connecting AI workforce agents with DingTalk for productivity and expanded capabilities).

With these, we maintain our above-street (vs. Visible Alpha Consensus Data) Alibaba Cloud revenue growth expectations of 53%/55%/55% year-on-year for Sep-2026/Dec-2026/Mar-2027 quarters. We anticipate investor focus around capex/funding sources in powering the 20GW power consumption target by 2032, which we estimate will be manageable, from a combination of asset heavy and asset light formats, 1) capex, funded by stable eCommerce EBITA, growing cloud profits, cash on balance sheet and future funding, 2) partnerships with data centers/telcos/others and 3) rental, esp. for overseas AI compute.

Reconciling power consumption/capex/cloud growth: We estimate 20GW by calendar year 2032 will imply 1-2GW increase p.a. over the next few years, from c.5-6GW in end-2025 (GSe), translating to Rmb200-300bn annual capex for Alibaba over the next few years, assuming c.US$20-25bn capex per GW attributable to Alibaba from a T-head custom silicon dominated approach and mixture of asset heavy, asset light and rental (opex) compute. We believe its custom silicon T-head creates unique US$/GW advantages vs. domestic and overseas peers, where T-Head could contribute to half of Alibaba Cloud compute medium-term, from c.10% today. Meanwhile, the company continues to expect US$100bn external AI cloud revenue for calendar year 2030 and medium-term 20% EBITA margin, depending on the evolving capex-to-revenue conversion in the medium-term and internal/external revenue mix. We are currently forecasting Rmb209/243/260bn capex for

Goldman Sachs does and seeks to do business with companies covered in its research reports. As a result, investors should be aware that the firm may have a conflict of interest that could affect the objectivity of this report. Investors should consider this report as only a single factor in making their investment decision. For Reg AC certification and other important disclosures, see the Disclosure Appendix, or go to Analysts employed by non-US affiliates are not registered/qualified as research analysts with FINRA in the U.S.

Goldman Sachs Alibaba Group (BABA)

n 20GW by 2032: AI cloud as the supergrid for Machine Intelligence, where Alibaba expects Alibaba Cloud to reach 20GW in power consumption by 2032 to be powered by Alibaba Cloud & partners. The company expects continued strong compute demand from intelligent driving, embodied AI and world models. Fastest drivers of AI Cloud growth have been from frontier AI lab customers, physical AI and overseas markets, see Exhibit 2. n Qwen’s roadmap: A clear path to Artificial Super Intelligence (ASI), with Qwen4 coming soon and Qwen4.5/5 to scale towards 5-10T parameter model sizes, alongside multi-modal understanding/generation. 1)…