Alibaba Group Holding Ltd (09988) Report Interpretation

Overview

The report examines whether Alibaba's expanding AI infrastructure investment can earn acceptable returns and how it can be funded. Bernstein concludes that datacentre unit economics broadly support management's claimed payback periods, but that sustained cloud expansion requires exceptionally large capital outlays and may leave scope for further financing.

Core views

Alibaba's US$10.2bn equity raise triggered investor concern because the company had US$30.7bn of net cash as of June 30 and because investors questioned whether the financing sequence weakened management's claims about AI-capex returns. Bernstein argues that the market reaction reflects both dilution concerns and uncertainty over the ultimate capital requirement, even though the shares were not viewed as especially expensive. The institution nevertheless maintains an Outperform rating with price targets of US$165 for BABA and HK$161 for 09988.HK. Bernstein tests management's return claims through a bottom-up Alicloud datacentre model covering server purchases, facilities and network spending, power costs, utilisation, bare-metal rental income and MaaS revenue attachment. For RMB60bn of incremental capex deployed with Zhenwu 810E chips, the model assumes 60% server capex, 45% initial paid-capacity occupancy rising to 85%, RMB65,000 monthly rental per server, and a 50% MaaS revenue uplift over bare-metal rental. It produces an implied cash payback of 3.02 years, cumulative unlevered free cash flow turning positive in Year 4, and a Year-3 EBITDA margin of 84.2%. For the new M890 platform, commercial deployment began in August. Bernstein estimates roughly RMB70,000 per GPU plus 40% overhead costs to reach server-level costs, while channel checks suggest an 8xM890 server rental rate around RMB80,000 per month and the model uses RMB660,000 monthly rental for a 64xM890 supernode. The M890 scenario assumes 50% server capex, 55% MaaS uplift, and better interconnectivity than 8xGPU configurations. It generates an implied 2.52-year payback, with Year-3 EBITDA margin of 84.9%. The report notes that Chinese AI labs place the 810E among leading domestic GPUs alongside Huawei and Cambricon, while online rental quotes and customer discussions indicate continuing compute tightness. The report's strategic conclusion is that additional capex should translate fairly directly into higher cloud revenue, with a two-quarter lag between spending and server activation. This also means Alicloud can continue exceeding revenue expectations if Alibaba is willing and able to finance the expanding server base. However, maintaining a 40% cloud-growth rate as the revenue base grows requires ever larger capital commitments. The key economic variable is the relationship between GPU-cost inflation and achievable compute-rental price increases: Bernstein expects rental-price growth to exceed hardware-cost inflation near term, but sees them converging as China's semiconductor supply improves. A temporary compute overbuild could eventually push rental-price growth below hardware-cost inflation. To exceed US$100bn of external cloud revenue by FY3/31, Bernstein estimates close to RMB1tn of cumulative AI capex. It assumes Alibaba's core e-commerce EBITA remains around RMB180bn–200bn annually, Cloud EBITA rises from about RMB32bn to nearly RMB180bn by FY3/31, Quick Commerce losses normalize relatively quickly, and AI Labs plus smaller All Other businesses generate roughly RMB240bn of cumulative EBITA losses over the intervening period. Against this operating-profit outlook, Bernstein estimates a roughly RMB170bn funding gap. The recent equity raise may have addressed close to half of it, but the report notes that RMB100bn–120bn of the post-deal RMB280bn, or US$41bn, net cash may represent merchant cash held in escrow. It therefore considers further market financing a plausible investor concern, although it sees management's FY3/29 free-cash-flow breakeven target as achievable if the gap is funded. Bernstein updated estimates primarily by increasing assumed server hardware costs and rental prices and raising non-AI cloud revenue forecasts, on the view that AI adoption also lifts adjacent infrastructure demand. Revenue forecasts rise 0.4%, 1.7% and 4.6% versus prior estimates for FY3/27E, FY3/28E and FY3/29E, respectively. Non-GAAP EBITA forecasts increase 1.2%, 7.1% and 12.5%, while non-GAAP diluted EPS changes are -1.6%, +3.4% and +8.6%. The US$165/HK$161 valuation uses a sum-of-the-parts approach for FY+1 core e-commerce and Cloud revenue and profits, triangulated against 15x FY+1 group P/E. The base SOTP implies US$168 per share, versus US$223 in the bull case and US$76 in the bear case. The report's longer-term concern is a hyperscaler "tragedy of the commons": if major Chinese hyperscalers all build capacity with external rental as the fallback use, aggregate compute capacity could be overbuilt. This would pressure rental economics and weaken the payback assumptions that underpin the cloud investment case.

Analysis framework

Bernstein begins with the investor reaction to Alibaba's equity raise, then builds datacentre-level unit-economics models for existing 810E and new M890 chips. It combines company disclosures, public power and datacentre benchmarks, rental-price scraping, industry contacts, AI-lab feedback and semiconductor bill-of-material estimates. It then translates capex assumptions into cloud revenue, profit, free-cash-flow and funding projections before updating forecasts and applying SOTP and P/E valuation cross-checks.

Methodology notes

• Industry AnalysisVolume-price decomposition

  Bottom-up datacentre unit economics linking installed servers, utilisation, rental rates, costs and MaaS attachment to cloud revenue and payback.

  The report estimates the revenue and cash generation of incremental compute capacity by modelling physical server capacity and the prices customers pay for it.

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

  FY+1 sum-of-the-parts valuation of core e-commerce and Cloud operations.

  Bernstein separately values Alibaba's main e-commerce and Cloud businesses, then combines them into group equity value and cross-checks the result against group P/E.

• Other

  Capex funding-gap and free-cash-flow breakeven modelling.

  The report compares projected cumulative AI capex with e-commerce and cloud profits, cash resources and expected losses in other businesses to estimate financing needs.

Asset mapping & comparison

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

• Alibaba Group Holding Ltd (BABA / 09988.HK)

  Primary covered company; AI datacentre investment is expected to drive Alicloud revenue growth but creates major funding needs.

  Strengths

  The report finds credible 2.5–3 year AI-capex payback periods, rising cloud profitability and AI-driven adjacent infrastructure demand.

  Weaknesses

  Sustaining rapid cloud growth requires progressively larger capex as the revenue base expands.

  Comparison

  The report compares 810E positioning with Huawei and Cambricon domestic GPUs and cites China H200 rental prices of around RMB110,000–120,000 per month.

  Risks

  Potential repeat equity financing, compute-capacity overbuild, rental-price pressure, macroeconomic weakness, competition, regulation and innovation-business losses.

Key data

• Recent equity raiseUS$10.2bnThe financing prompted investor concerns over dilution and funding needs.
• Net cash before the raise discussionUS$30.7bnReported as of June 30.
• 810E capex payback3.02 yearsBernstein's RMB60bn incremental-capex model.
• M890 capex payback2.52 yearsBased on the 64xM890 supernode scenario.
• Cumulative AI capex requirementClose to RMB1tnEstimated to support external cloud revenue above US$100bn by FY3/31.
• Estimated funding gapApproximately RMB170bnAbout half may have been addressed by the recent equity raise.
• Cloud EBITAApproximately RMB32bn to nearly RMB180bnBernstein's progression from current levels to FY3/31.
• FY3/29E revenue forecastRMB1,485,482mn4.6% above Bernstein's previous estimate.

Impact & implications

Bernstein's modelling supports the view that AI infrastructure spending can accelerate Alicloud revenue and preserve solid incremental margins. The investment case is therefore tied not only to cloud demand and pricing but also to Alibaba's ability to finance escalating capex without damaging investor confidence or encountering an industry-wide capacity oversupply.

Risks

• A significant Chinese hyperscaler compute-capacity overbuild could weaken rental pricing and datacentre returns.
• Alibaba may need to return to equity markets if the estimated funding gap is not covered by internal cash generation.
• Downside risks cited by Bernstein include credit and retail-consumption weakness, user-engagement fluctuations across Taobao and Tmall, internet-platform competition, antimonopoly regulation and losses in innovation initiatives.

What to watch

• Whether GPU rental-price growth continues to exceed server hardware-cost inflation.
• Alicloud revenue acceleration after the assumed two-quarter capex-to-revenue lag.
• Progress toward external cloud revenue above US$100bn and FY3/29 free-cash-flow breakeven.
• Further financing activity and the availability of cash after merchant escrow balances.
• Signs of Chinese hyperscaler compute-capacity oversupply.