Alibaba Cloud's AI compute investment could establish a path to 13%-20% ROIC

Overview

This report assesses the commercialization of AI compute in China using server-level unit economics, with a focus on Alibaba's return-on-capital potential in IaaS and MaaS. It concludes that, despite server procurement costs in China being higher than in the United States and overall ROIC being lower than that of U.S. peers, Alibaba can achieve 13%-20% ROIC through scaled AI infrastructure, cloud-service capabilities, and Qwen model capabilities, while recovering the related cash investment in approximately 2-3 years.

Core views

The report divides Alibaba Cloud's AI commercialization into three models: self-built GPU IaaS, leased/neocloud IaaS, and self-built MaaS. Self-built IaaS benefits from higher pricing and utilization but bears upfront capital expenditures; the leasing model has lower margins but immediate cash returns; MaaS has the highest ceiling for margins and ROIC, but depends heavily on token throughput per GPU, the inference mix, and token pricing. The report believes Alibaba's current cloud margins remain constrained by legacy cloud burdens, early-stage infrastructure, utilization ramp-up, and R&D and personnel costs; as the revenue mix shifts toward MaaS, margins have further upside potential.

Analysis framework

The analysis uses a three-part bottom-up, server-level unit economics framework, constructing a base case around revenue pricing, server capital expenditures, depreciation, IDC and energy costs, utilization, token throughput, training/inference mix, and token prices, and compares the results with U.S. AI cloud economics. Valuation uses a discounted cash flow model, with key assumptions of a 10% WACC and a 3% terminal growth rate.

Methodology notes

• Unit Economics ModelBottom-up Server-Level Analysis

  Uses an 8-GPU AI server as the basic accounting unit to estimate revenue, costs, operating margin, ROIC, and payback period across different business models.

  This framework identifies the sensitivity of AI compute economics to pricing, utilization, hardware costs, and operating costs.

• Return on Capital AnalysisROIC and Cash Payback Period

  Assesses the economics of upfront capital expenditures through return on invested capital and cash payback time.

  The report believes both self-built IaaS and MaaS can recover cash investment in approximately 2-3 years, although returns are significantly affected by hardware costs and commercialization efficiency.

• Valuation methodsDiscounted Cash Flow Model

  Values the business by discounting future cash flows.

  The report uses a 10% WACC and a 3% terminal growth rate, consistent with its China internet coverage.

Asset mapping & comparison

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

• ALIBABA GROUP HOLDING LTD（US.BABA）

  Direct coverage target

  Strengths

  Possesses large-scale AI infrastructure, high-quality cloud-business capabilities, and Qwen model capabilities; MaaS revenue expansion can improve cloud-business margins and returns on capital.

  Weaknesses

  Current cloud margins remain weighed down by legacy low-margin workloads, early-stage infrastructure, utilization ramp-up, and R&D and personnel costs.

  Comparison

  China's base-case ROIC for self-built IaaS is approximately 13%, versus approximately 31% in the United States; China's MaaS ROIC is approximately 19.5%, versus approximately 46.2% in the United States, although high throughput can partly narrow the gap.

  Risks

  Intensifying competition, reinvestment costs exceeding expectations, weaker-than-expected AI demand, a slow increase in the inference mix, token pricing pressure, and regulatory uncertainty.

Key data

• Equity RatingOverweight; Top PickMorgan Stanley's relative rating system.
• Target Price and UpsideUS$180.00; 45%Calculated based on the US$123.81 closing price on 2026-08-14.
• Self-Built GPU IaaS Base CaseApproximately 44% operating margin, 13% ROIC, and a 3.1-year cash payback periodAssumes server capital expenditures of more than RMB8mn per server and long-term lease pricing of approximately RMB250k/month.
• Leased/neocloud IaaS Base CaseApproximately 20% operating marginCustomer monthly lease payments of approximately RMB250k, corresponding to compute leasing costs of approximately RMB200k, with no upfront server capital expenditure.
• Self-Built MaaS Base CaseGross margin above 76%, 53% operating margin, approximately 19% ROIC, and a 2.5-year cash payback periodAssumes 4,000 tokens/second/GPU, a 50% inference mix, and blended token pricing of RMB9.5/million tokens.
• Comparison of Self-Built IaaS ROIC in China and the United StatesChina 13%; United States 31%China's server costs are nearly three times those in the United States, partly offset by lower IDC and energy costs.
• Alibaba MaaS TargetsMid-year MaaS ARR target above RMB10bn and year-end target above RMB30bnThe report believes achieving these targets and improving throughput and the inference mix would help approach base-case MaaS returns.

Impact & implications

For Alibaba, the report's core investment implication is that AI capital expenditures are not merely an expense-related pressure, but could become return-generating investments that enhance Alibaba Cloud's external revenue and long-term profitability. In the short term, earnings and free cash flow will remain pressured by high capital expenditures; in the medium term, if high-end compute maintains high utilization, MaaS revenue share rises, and inference workloads increase, the cloud business could improve from its current approximately 11%-12% margin to a higher level. The report also emphasizes that higher hardware costs in China make return realization more dependent on cost reductions and improved commercialization efficiency.

Risks

• Intensifying competition in cloud computing and AI services, leading to pricing or utilization below expectations.
• Capital expenditures and depreciation costs for servers, GPUs, and related infrastructure exceeding expectations.
• A slowdown in enterprise digitalization, preventing cloud revenue growth from reaccelerating.
• AI demand and MaaS commercialization falling short of expectations, with training workloads remaining a high proportion over the long term.
• Weak consumption potentially affecting monetization of core e-commerce and overall earnings growth.
• Increased regulatory scrutiny of internet platforms.

What to watch

• Alibaba Cloud external revenue growth, EBITA margin, and changes in high-end compute utilization.
• Progress toward MaaS ARR targets above RMB10bn and RMB30bn.
• The proportion of AI revenue in cloud revenue and the increase in MaaS's share of the cloud revenue mix.
• Changes in Qwen model capabilities, token throughput efficiency, inference mix, and token pricing.
• GPU/server procurement costs, IDC and energy costs, and capital expenditure intensity.
• Changes in China's AI cloud competitive landscape and enterprise digitalization demand.