- Company bets AI agents will reshape demand for cloud infrastructure
- New platform spans chips, cloud architecture, models and inference systems
Alibaba Cloud, Alibaba’s cloud service provider, on May 20 unveiled a sweeping overhaul of its AI infrastructure stack, launching a new in-house AI chip, upgraded flagship model and a redesigned cloud platform as the company positions itself for what it sees as an “Agentic AI” era.
The upgrade spans everything from semiconductors and cloud architecture to large language models and inference systems, marking one of Alibaba’s biggest pushes yet to turn AI services into the core growth engine of its cloud business.
Senior vice president Liu Weiguang said AI agents capable of operating around the clock are driving “unlimited” demand for both AI and cloud computing.
“After agents crossed the critical threshold, they can work 24 hours a day nonstop,” Liu said. “Demand for AI and cloud services becomes effectively endless.”
As part of the overhaul, Alibaba Cloud launched “Qwen Cloud” (www.qianwenai.com), a new AI-focused platform that replaces traditional cloud-service menus and management consoles with a simplified interface centered around machine-readable command instructions.
At the hardware layer, the company introduced its Panjiu AL128 supernode server powered by its next-generation Zhenwu M890 AI chip and proprietary ICN Switch 1.0 interconnect chip.
The system allows 128 AI chips to function as a single computing cluster with peer-to-peer latency below 150 nanoseconds, targeting the heavy concurrent inference and training workloads associated with AI agents and large language models.
Alibaba’s chip unit T-Head also disclosed a new roadmap for more advanced AI chips over the next two years. The company said cumulative shipments of its Zhenwu AI chip series have reached 560,000 units serving more than 400 clients.
On the model side, Alibaba released its new flagship Qwen3.7-Max model. Zhou Jingren, head of the Qwen model division, said large models are undergoing “a core paradigm shift — from aligning with human preferences to aligning with task objectives.”
“In the past, we wanted models that could speak well,” Zhou said. “Now we want models that can actually get things done.”
Alibaba Cloud also said it has redesigned its cloud services so AI agents can directly invoke cloud functions as standardized modules, effectively turning cloud infrastructure into machine-readable tools for autonomous AI systems.
Last week, Alibaba reported that annualized revenue from AI models and AI application services had surpassed 8 billion yuan ($1.1 billion). Following the latest overhaul, the company expects AI-driven model-as-a-service (MaaS) revenue to overtake traditional cloud-server rentals as Alibaba Cloud’s largest source of income.
Despite its aggressive in-house AI push, Alibaba Cloud said its platform will remain open to third-party AI developers, including Zhipu AI, MiniMax and Moonshot AI.
