AI Affairs, home

Sunday 27 September 2026

Corporate

Alibaba targets 20GW of data centre capacity as Citi raises spending forecasts

Citi lifted its fiscal 2029 capital-expenditure estimate by 57 per cent to 282 billion yuan after Alibaba set out its cloud expansion target.

Phase 4 of Alibaba Xixi Park modern orange‑facade building
Photo: Windmemories, CC BY-SA 4.0, via Wikimedia Commons (cropped)

Alibaba set a target on 22 September to operate more than 20GW of global data centre capacity by 2032 in a company announcement. Citi subsequently raised its fiscal 2029 capital-expenditure estimate for Alibaba by 57 per cent to 282 billion yuan, the South China Morning Post reported.

Key points

  • Citi also raised its fiscal 2027 and 2028 spending estimates to 258 billion yuan and 283 billion yuan.
  • Citic Securities estimates that 20GW of capacity could support roughly US$170 billion in annual external cloud revenue.
  • Alibaba presented its Zhenwu V900 processor and a server designed to scale to a 500,000-chip cluster.
  • New Qwen devices include an agentic computer and a note-taking device priced at RMB 1,199.

Citi raises three Alibaba spending estimates

Citi’s Alicia Yap, head of Pan-Asia internet research, increased her estimates for Alibaba’s capital expenditure in fiscal 2027, 2028 and 2029 to 258 billion yuan, 283 billion yuan and 282 billion yuan. Those figures are 13 per cent, 41 per cent and 57 per cent above her previous estimates, respectively, according to the South China Morning Post.

The revisions followed Alibaba Cloud’s capacity target. Alibaba said the planned data centres would support demand from its own AI models and those run by external customers. At the Apsara Conference in Hangzhou, chief executive Eddie Wu Yongming set out models, chips and cloud infrastructure as the three components of the company’s approach.

Citic Securities estimated that Alibaba’s 20GW target could underpin roughly US$170 billion a year in revenue from external cloud customers. Alibaba’s stated ambition is for external cloud and AI revenue together to exceed US$100 billion by 2030. The capacity target has a later deadline, in 2032.

CCB International analyst Cathy Chan said investors had moved from last year’s enthusiasm for AI catalysts towards tangible returns and infrastructure efficiency. She described Alibaba’s use of its own chips and cloud architecture to reduce inference costs, alongside open-source model weights intended to retain developers, as suited to that demand.

Citi’s capital-expenditure forecasts for Alibaba sit within a larger forecast for China. Morgan Stanley projected 8.5 trillion yuan of AI infrastructure spending in the country between 2026 and 2030, averaging 17 per cent of total spending by US peers. It forecast China’s IT electrical capacity at 81GW by 2030, against 26GW in 2025.

Zhenwu V900 joins Alibaba’s cloud buildout

Alibaba introduced the Zhenwu V900 for AI training and inference. The company says it has more than three times the performance of its predecessor, the M890, and plans mass production and commercial release in the first quarter of 2027, The Elec reported.

Alibaba also presented a supernode server that it says can scale to clusters of as many as 500,000 chips. Its Yitian 720 and 730 server CPUs are targeted for release in 2027. AI Affairs previously reported T-Head’s plans for those CPUs and their integration with the Zhenwu V900.

Alibaba says more than 650 customers use T-Head’s Zhenwu chips across industries including automotive, finance and manufacturing, The Elec reported. The company is also training Qwen 4 and projects that future Qwen 4.5 and Qwen 5 models will reach between 5 trillion and 10 trillion parameters. AI Affairs earlier covered the model roadmap and Zhenwu V900 launch.

Counterpoint Research senior analyst Parv Sharma identified leading-edge manufacturing nodes, high-bandwidth memory supply and advanced packaging as constraints on Alibaba’s chip plans. He said Alibaba was seeking to counter Nvidia’s advantage in individual chips through larger clusters, interconnects and closer integration with Qwen software.

QwenBook and QwenNote A2 reach devices

Alibaba extended its Qwen range into devices at Apsara. Qwen Book is an agentic computer powered by Qwen Desktop OS, which Alibaba says brings its models, applications and cloud services into one environment, TechNode Global reported. In a demonstration, the company showed the computer retrieving a presentation from the cloud and preparing slide edits for a user’s approval.

QwenNote A2, a device intended to turn recorded conversations into material for workplace agents, went on sale in China on 22 September at RMB 1,199. Alibaba says it can work with QwenWork to create tasks, send messages and schedule meetings from transcribed conversations. The company also introduced Qwen Glasses N1 and N1 Pro and Qwen Clip earbuds, which it says were co-engineered with Bose.

Alibaba opened pre-orders for the new wearables on 22 September. Alibaba has scheduled their sales to begin on 13 October.

Topics: Chips, Data centres, Enterprise adoption, Inference