Alibaba Group Holding Ltd (09988): Bernstein sees Alibaba's AI cloud build-out extending its growth runway and reiterates Outperform.

Overview

This quick take examines how Alibaba's Apsara strategy has shifted toward AI-serving infrastructure, enterprise applications and context capture. Bernstein maintains Outperform, viewing Alicloud's AI-compute opportunity as a growing offset to weak China consumption and pressure on Taobao and Tmall.

Core views

Alibaba's 20GW AI data-centre capacity target for 2032 was the central message from Apsara and the first time the company has given guidance in gigawatts. Assuming 25-30% of capacity is reserved for internal use, Bernstein calculates that management's revenue-per-GW guidance implies a US$170-180bn revenue target by 2032. Relative to Alibaba's previous US$100bn revenue target for 2030, Bernstein interprets the guidance principally as an extension of the high-growth runway rather than a material lift to the nearer-term trajectory. Given tight compute supply, the firm believes US$12bn of revenue per GW could prove conservative. The major constraint is capital intensity. Applying global greenfield norms of roughly US$30bn of capex per GW to 2.5-3GW of annual incremental capacity would imply more than RMB500bn of annual capex, far above Alibaba's RMB200-300bn and rising operating cash flow. Alibaba has roughly 3-4GW of existing compute, based on Bernstein's triangulation of 2GW in 2022 and the subsequent Alicloud capex ramp. Management argues that selling T-head chips to partners and leasing compute back from them will reduce Alibaba's direct capital burden. Bernstein considers reaching 50% in-house silicon, subject to fab and memory capacity, and deploying it through this chip sale/lease-back model a necessary condition for reaching 20GW by 2032. Telecom operators and insurers were identified as potential asset-holding partners. Fully prepaid sub-one-year compute contracts should also ease cash-flow pressure, making deferred revenue an important indicator. Apsara placed less emphasis on frontier-model launches, with Qwen4 still described as coming soon, and more on serving infrastructure and AI applications. Alibaba presented agentic harnesses and proprietary enterprise context as foundations for long-horizon task completion. While Workbuddy has an early lead in harness adoption, Bernstein describes Alibaba's Qoder and Qwen Work proposition as distribution-led: it combines Alicloud's go-to-market scale, enterprise-data capture, skills trained from collaborative work context, the Qwen model family, DingTalk, and devices including Quark glasses and smart earpods. Bernstein frames this as a full-stack AI proposition: silicon and hyperscale compute can lower serving costs; Qwen models supply the model layer; and Alicloud's distribution supports the application layer. With compute demand tight, the firm sees a direct link between incremental capex and lease arrangements with T-head operators, on one hand, and incremental Alicloud revenue and profit on the other; it expects those results to continue exceeding estimates. In contrast, anaemic consumption growth in China continues to pressure Taobao and Tmall. Bernstein nevertheless argues that Alicloud is valued modestly even if e-commerce is assigned only limited value. Bernstein values Alibaba at US$165 per BABA share and HK$161 per 9988.HK share using a sum-of-the-parts valuation of FY+1 revenue and profit across core ecommerce and Cloud. Its table shows adjusted EPS of CNY26.82, CNY44.78 and CNY64.65 for F26A, F27E and F28E, respectively, alongside BABA adjusted P/E of 28.9x, 17.3x and 12.0x. Revenue is forecast at 1,024, 1,162 and 1,320, with a 13.6% CAGR, while operating earnings rise from 50,150 to 98,824 and 159,019, implying a 78.1% CAGR.

Analysis framework

Bernstein starts with Alibaba's capacity guidance and translates usable AI-compute capacity into a long-term revenue opportunity. It then tests the implied capital requirement against operating cash flow, assesses the chip sale and lease-back model as the financing mechanism, and evaluates the AI product stack and cloud demand drivers. The valuation uses FY+1 revenue and profit for core ecommerce and Cloud in a sum-of-the-parts framework.

Methodology notes

• Valuation methodsSOTP (Sum-of-the-Parts) Valuation

  Sum-of-the-parts valuation of Alibaba's FY+1 core ecommerce and Cloud revenue and profits.

  The report values Alibaba by separately considering its core ecommerce and Cloud businesses, then combines those values to derive the US$165 and HK$161 price targets.

• Industry AnalysisSupply-demand framework

  AI-compute supply tightness and capacity-driven cloud revenue analysis.

  Bernstein links incremental AI-compute capacity and constrained compute supply to Alicloud revenue and profit potential, while considering the capex needed to add capacity.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Alibaba Group Holding Ltd (BABA / 9988.HK)

  Primary covered company; beneficiary of AI-compute growth and Alicloud's full-stack AI strategy.

  Strengths

  20GW capacity ambition, Alicloud distribution, Qwen models, enterprise-context capture, T-head partner model and prepaid compute demand.

  Weaknesses

  Weak China consumption pressures Taobao and Tmall; Qwen4 remains unlaunched.

  Comparison

  Workbuddy is described as having an early lead in harness adoption.

  Risks

  Capital intensity, execution of in-house silicon and lease-back partnerships, macro and platform risks.

Key data

• AI data-centre capacity target20GW by 2032Alibaba's first GW-denominated capacity guidance.
• Implied revenue targetUS$170-180bn by 2032Bernstein estimate assuming 25-30% of capacity is reserved for internal use.
• Greenfield capex benchmarkc. US$30bn per GWUsed to illustrate the scale of direct-capex requirements.
• Implied annual capex at 2.5-3GW additionsOver RMB500bnCompared with RMB200-300bn and rising operating cash flow.
• BABA price target and upsideUS$165; 43% upsideAgainst a US$115.75 close on 21 Sep 2026.
• Adjusted EPSCNY26.82 / CNY44.78 / CNY64.65F26A / F27E / F28E.

Impact & implications

The report argues that AI cloud capacity, chip partnerships and prepaid compute demand can extend Alibaba's growth runway and support Alicloud earnings above estimates. It also highlights that this opportunity must be financed through a less asset-intensive model because direct greenfield expansion would be inconsistent with current operating cash flow.

Risks

• Macroeconomic risks, including credit conditions and retail consumption.
• Fluctuations in user engagement across Taobao, Tmall and other platforms.
• Competition from rival internet platforms.
• Regulatory risk, including China's anti-monopoly regulations.
• Losses in Alibaba's innovation initiatives and other segment.
• Insufficient fab or memory capacity could hinder the 50% in-house-silicon objective.

What to watch

• Progress toward 50% in-house silicon and the T-head chip sale/compute lease-back model.
• The arrival and performance of the V900 platform in 2027, including its planned 500k-PPU, 1GW cluster support.
• Qwen4 as a critical proof point.
• Deferred revenue, as customers reportedly prepay compute contracts of less than one year.
• Alicloud revenue and profit relative to estimates, and the impact of China consumption on Taobao and Tmall.