Alibaba Unveils Pragmatic AI Roadmap Focusing on Returns
Chinese technology organisation Alibaba Group Holding presented its artificial intelligence roadmap at the Apsara Conference. This signals a pragmatic shift, aiming for clearer monetisation despite projected capital expenditure escalation. Industry analysts observe investor sentiment prioritises tangible returns and efficient infrastructure. According to a report by SCMP, this reflects market expectations.
The conference, themed "Intelligence goes beyond", ran for three days. Cathy Chan, an analyst at CCB International, described it as "Alibaba's disciplined execution across its full-stack AI ecosystem". She noted a shift in investor outlook, from previous year's AI enthusiasm, to a demand for actual profits and infrastructure efficiency.
Alibaba's strategy directly addressed this pragmatic investor mood. The company proposed treating AI tokens as a utility. It intends to use proprietary chips and cloud architecture to reduce inference costs. Releasing open-source model weights is also part of the plan. So, the organisation aims to retain devs within its ecosystem by offering these provisions.
Alibaba Chief Executive Eddie Wu Yongming outlined three core elements for the firm's "machine intelligence" era. These included AI models, chips, and cloud infrastructure. The Zhenwu V900, billed as "China's most powerful AI processor", was introduced. Forthcoming Yitian 720 and 730 server central processing units are expected in 2027, alongside a Panjiu supernode server able to scale to a single cluster containing 500,000 chips.
The company mentioned future AI systems with 5 trillion to 10 trillion parameters. Multimodal model upgrades, capable of processing and generating more than text, are planned. Alibaba Cloud aims to operate over 20 gigawatts of global data centre capacity by 2032, powering internal and external AI models.
Citic Securities analyst Liao Yuan noted Alibaba's expansion into enterprise and consumer markets, suggesting opportunities to monetise AI applications at the application layer. Alibaba presented Qwen Intelligence, a solution for smartphone makers building agents across apps, Model-as-a-Service platforms Bailian and Qwencloud, productivity tool QwenWork, and Accio Work for cross-border commerce. And hardware products also debuted.
The hardware included the "agentic computer" QwenBook, AI glasses, earbuds, and a note-taking device. Following the 20GW announcement, Citi's Alicia Yap raised capital expenditure forecasts for Alibaba. Fiscal 2027, 2028, and 2029 estimates rose to 258 billion yuan (US$38.5 billion), 283 billion yuan, and 282 billion yuan respectively. These represent increases of 13, 41, and 57 per cent over prior figures.
Citic Securities projects the 20GW capacity could yield approximately US$170 billion in annual external cloud revenue. This aligns with Alibaba's objective to generate over US$100 billion from combined external cloud and AI revenue by 2030. Citic's Liao stated tighter integration of Alibaba's models, chips, and cloud improved growth visibility.
The expansion reflects a broader industry trend. Morgan Stanley forecast China's AI infrastructure spending to reach 8.5 trillion yuan between 2026 and 2030, averaging 17 per cent of US peer spending. China's information technology power capacity is expected to increase from 26GW to 81GW by 2030. But, Alibaba's semiconductor roadmap indicates constraints shifted beyond chip design, according to Parv Sharma, a senior analyst at Counterpoint Research.
Mr Sharma identified leading-edge nodes, high-bandwidth memory supply, and advanced packaging as the primary current limitations. He stated that Alibaba intends to mitigate Nvidia's single-chip lead. This will be achieved through cluster scale, improved interconnects, and tighter integration with its Qwen model and software stack.
Mr Sharma concluded that the situation presents two distinct AI operational structures, rather than China surpassing the US. He noted Nvidia benefits from global supply chains. Alibaba, meanwhile, holds increasing domestic dominance.
Alibaba presented a pragmatic AI roadmap at its Apsara Conference, focusing on monetisation and infrastructure efficiency.
Chief Executive Eddie Wu Yongming detailed a "machine intelligence" era built on AI models, chips, and cloud infrastructure.
New hardware, including the Zhenwu V900 processor and QwenBook computer, was unveiled.
Alibaba Cloud aims for 20 gigawatts of global data centre capacity by 2032.
Analysts noted a shift towards tangible returns for AI investments and a growing divergence in US and Chinese AI development.
Source: SCMP
