Highlights

• AI/AI cloud remained the bright spot in 2Q26, with Alibaba and Baidu delivering 45%/50% yoy revenue growths in AI cloud respectively and Tencent reporting healthy cloud growth on AI demand. However, the surge in capex and Alibaba’s HK$80b (US$10.3b) placement shifts the AI debate from project returns to funding duration and capital structure. The investment cycle is pressuring near-term FCF, but stronger utilisation, proprietary chips and higher-margin AI services should improve compute economics over time.
   
• Maintain MARKET WEIGHT. Top BUY: Alibaba.
  
   

Analysis

• Funding risk is increasingly differentiated across CSPs. Due to the step-up in capex, we expect Alibaba to have the largest structural funding gap among Chinese hyperscale peers, and it is the most likely to rely on debt, selective equity and potentially cloud-level financing. We do not expect Alibaba to have any solvency risk, as it is: a) sustaining >50% external cloud growth, b) keeping mature infrastructure ROIC in the mid-teens or above, c) containing non-infrastructure AI losses, and d) funding the multi-year build without repeated parent-level equity issuance. Alibaba also has sizeable investment portfolios in AI/LLM companies including CXMT, Z.ai and MiniMax totalling around Rmb330b in listed assets. Underlying compute economics remain attractive, with an estimated 20% IRR and payback period of 2-3 years, but the recent capital raising suggests the capex cycle will stay elevated for longer. 
   
• Tencent largely self-funding despite front-loaded AI spending. Tencent remains largely self-funded, with limited parent-equity risk. Its annualised 1H26 operating cash flow of roughly Rmb310b should comfortably cover our Rmb200b capex assumption, dividends and most buybacks. The decline in net cash to around Rmb58b mainly reflects the timing of the Rmb41.6b dividend and approximately Rmb50b of compute prepayments, rather than a structural funding gap. Tencent has also diversified funding through long-dated USD/RMB notes, short-term borrowings, and sizeable investment portfolios of Rmb480b listed and Rmb350b unlisted assets. We therefore see limited parent-equity issuance risk; the main economic costs are instead interest expense, potential tax leakage and the opportunity cost of monetising investment assets.
   
• Baidu has weak internal cash generation, but parent dilution remains unattractive. Baidu’s weaker cash generation is offset by its reluctance to dilute at a low parent valuation, making Kunlunxin the more likely external funding vehicle. Baidu’s 2Q26 operating cash flow of only Rmb3.4b fell short of Rmb11.4b of capex, with the gap increasingly funded through bank borrowing and leasing. Nevertheless, roughly Rmb100b of adjusted liquidity can cover the base-case deficit for several years. The key constraint is not near-term liquidity, but Baidu’s low parent valuation and management’s limited economic incentive to issue parent equity. We therefore see Kunlunxin, rather than Baidu parent equity, as the more logical external equity-financing vehicle. This avoids direct parent share-count dilution, but transfers part of the AI asset’s economics to outside investors.