Charting the Path to Superintelligent AI: Insights from Alibaba’s Cloud Conference

At the 2025 Cloud Conference, Alibaba Group Chairman and CEO Wu Yongming delivered a keynote titled “The Road to Super Artificial Intelligence,” sharing his views on super AI, large models, the “Tongyi Qianwen” project, capital investment, intelligent agents, and robotics.

Three stages of super AI : achieving AGI is a deterministic event, but the ultimate goal is a self‑iterating, fully human‑surpassing super AI (ASI). The path consists of three phases – intelligent emergence, autonomous action, and self‑iteration.

Large models as the next OS : In the transformation from AGI to ASI, large models will become the operating system of both the physical and digital worlds, inheriting the role of current OSes.

Tongyi Qianwen’s goal : The project follows an open‑source route, aiming to become the “Android system of the AI era.”

Capital investment : Alibaba’s capital expenditure will increase beyond the existing 380 billion RMB, with AI‑related power consumption projected to grow tenfold by 2032, driving exponential growth in compute resources.

Future agents and robots : The number of intelligent agents and robots could eventually exceed the global human population, working alongside people and profoundly impacting the real world.

The current AI‑driven intelligent revolution surpasses past industrial and information revolutions. AGI will amplify human intellect and free human potential, paving the way for ASI.

Recent years have shown AI’s rapid progress: within a few years AI advanced from high‑school‑level to PhD‑level performance, winning International Math Olympiad gold medals, and achieving unprecedented user penetration with chatbots.

Global AI investment has already exceeded $400 billion, and the next five years are expected to see over $4 trillion in cumulative AI spending, accelerating model development and application penetration.

AGI is now a certain event, but it is only the starting point; AI will continue beyond AGI toward self‑iterating superintelligence.

ASI will create “super scientists” and “full‑stack super engineers,” solving scientific and engineering challenges at unprecedented speed, from medicine to sustainable energy and interstellar travel.

Stage 1 – Intelligent emergence : The internet digitized almost all human knowledge, providing the foundation for large models to acquire generalized intelligence, conversational ability, and multi‑step reasoning.

Stage 2 – Autonomous action : AI will move beyond language to act in the real world, decomposing complex tasks, using and creating tools, and delivering massive impact across logistics, manufacturing, software, commerce, biomedicine, finance, and research.

The key enablers are large‑model tool‑use capabilities and improved coding abilities, allowing agents to autonomously develop and execute software solutions.

Natural language will become the programming language of the AI era, enabling anyone to create agents that perform virtually any digital or physical task.

Stage 3 – Self‑iteration : AI must connect to the full spectrum of raw data from the physical world and engage in continuous self‑learning. By building its own training infrastructure, optimizing data pipelines, and upgrading model architectures, AI will achieve self‑learning and self‑iteration.

With persistent interaction and real‑time feedback, AI will iteratively fine‑tune itself, eventually forming an early form of superintelligence.

Large models are envisioned as the next operating system, replacing traditional OSes and serving as the middle layer that schedules user, software, and AI compute resources.

Analogies: natural language is the AI‑era programming language, agents are new software, context is memory, and model‑tool interfaces act like a PC‑era bus.

Large models will “consume” software, allowing anyone to create unlimited applications via natural language, dramatically expanding the developer base.

Model deployment will become ubiquitous, running on all devices with persistent memory, cloud‑edge synergy, and self‑updating capabilities, similar to modern operating systems.

Alibaba Cloud’s strategy: open‑source the Tongyi Qianwen models, provide a one‑stop model service platform (BaiLian), support model customization and rapid agent development, and operate a vertically integrated AI super‑cloud with proprietary storage, networking, and compute chips.

The AI super‑cloud will serve as the next‑generation computer, delivering massive compute power for billions of agents and robots, with AI becoming the primary commodity—tokens acting as the future “electricity.”

Alibaba Cloud plans a three‑year, 380 billion RMB AI infrastructure investment, scaling data‑center energy consumption tenfold by 2032 to meet the demands of the ASI era.

Looking ahead, humans and AI will collaborate in a new partnership where AI augments human capabilities, enabling unprecedented productivity and creating new societal demands.

“We are optimistic about the future. The stronger AI becomes, the stronger humanity becomes,” Wu Yongming said.