BofA Securities: BABA-W (09988) 2026 Apsara Conference reinforces full-stack AI opportunity

BofA Securities issued a research report maintaining a "Buy" rating on BABA-W (09988, BABA.US), with a US target price of US$175 and an H-share target price of HK$172. Alibaba Cloud held its flagship Apsara Conference this week, where Alibaba CEO Eddie Wu reiterated the company's long-term commitment to AI, emphasizing that as AI advances toward machine intelligence, the company will continue to invest in AI models, chips, and cloud infrastructure. The bank observed strong and rapidly accelerating customer demand, a clear strategic commitment to AI agents, continued development of recursive self-improvement (RSI) architecture, and increasingly autonomous, self-reinforcing full-stack capabilities across infrastructure, models, platforms, and applications. The conference reinforced the bank's positive view on Alibaba, which it believes can capture value across multiple layers of the AI stack given its industry-leading full-stack AI ecosystem, with cloud driving near-term monetization and AI services, and applications and international expansion supporting longer-term growth. The next potential catalyst is the quarterly update for the quarter ended September, to be released before the October quiet period. The bank noted that Alibaba plans to operate over 20 GW of global data center capacity by 2032, highlighting management's confidence that medium- to long-term AI compute demand will significantly exceed current supply, implying average annual capacity additions of approximately 2 GW. The bank estimates construction costs at approximately US$30-35 billion per GW, but expects a substantial portion to be financed through leasing and other partnership arrangements, limiting Alibaba's direct capex burden, and therefore maintains its FY2027-2028 capex forecast at RMB 230-250 billion. The 20 GW target is broadly consistent with management's guidance of US$100 billion in external AI and cloud revenue by FY2031 (approximately calendar year 2030). Alibaba's T-Head released its latest-generation AI accelerator, the Zhenwu V900, which management described as China's highest-performance AI chip, with 3x the performance of the Zhenwu M890, scalable to up to 500,000 chips in a single cluster, and supporting frontier model training and inference. Alibaba management expects T-Head's shipment volume to rise significantly in the coming years, with the proportion of self-developed chips in Alibaba Cloud's deployed compute capacity targeted to increase from less than 10% currently to 40-50% over time. The bank views this as a key driver for improving supply chain control and enhancing the cost efficiency of AI infrastructure. Regarding the Tongyi roadmap, the company plans to further advance the Qwen Max series, with Qwen 4.5/5 targeting 5 trillion to 10 trillion parameters, while addressing broader deployment needs with Flash, Plus, and on-premise models; Qwen 4 is expected to be launched soon, with upgrades to its multimodal portfolio including image, video, audio, and world models.