Alibaba Group Holding Limited

AI + Cloud is the clear growth engine; accelerating demand and improving profitability - Alibaba Group reported total revenue +9% YoY to RMB 269b for the quarter. - Alibaba Cloud external revenue +45% YoY (management highlighted this as an accelerated, “broad-based” figure; 22-quarter high). - Adjusted EBITDA margin expanded to ~11.6% for Cloud; management also cited sequential improvement and expected continued margin expansion. - AI-related products revenue is triple-digit for the 12th consecutive quarter, RMB 12.4b in the quarter, implying ~RMB 49.5b annualized run-rate (~$7.3b); these products were 35% of external cloud revenue. - Management expects high-growth to sustain into coming quarters, driven by both compute/storage MaaS and AI applications/agents.

MaaS/agents create a “flywheel” across compute, cloud infrastructure and tokens; supply constraints remain - Eddie described strong agent-driven monetization: agents increase token demand and GPU compute, while also boosting demand for traditional cloud (CPU, storage, DBs, networking). - MaaS ARR surpassed RMB 16b (as of August) (management referenced “current market feedback and contract pipelines”), with implied continued strong momentum. - Management said compute demand continues to outstrip supply, and as Alibaba ramps supply further (especially 2H, per chip/data center build-out), growth should accelerate alongside improved profitability.

Full-stack AI positioning is strengthening: proprietary silicon scaling + faster model iteration + open ecosystem - T-Head full-stack silicon (GPU/CPU/networking) is central to the cost/margin thesis; management said Zhenwu chips served 650+ customers (on Alibaba Cloud). - Zhenwu M890 “supernode” launched at commercial scale, positioned to run inference for models >2 trillion parameters; customers (e.g., Kimi K3, Qwen 3.8 Max) already using it for MaaS services. - Data center delivery pace: management claimed hyperscale AI data centers can be delivered in ~100 days. - Open ecosystem: Qwen model series downloaded 3b+ times globally and has 300k+ derivative models, which management argues increases cloud consumption demand (a “virtual cycle”).