Alibaba unveils full-stack AI strategy with $380B investment target

Pioneering, dynamic, cutting-edge, and global — this is how many visitors described the 2026 Cloud Summit in Hangzhou. Spanning over 50,000 square meters, the intelligent technology exhibition drew a diverse crowd: foreign trade entrepreneurs from Yiwu Small Commodity Market, university students, and international tech enthusiasts alike.

From September 22 to 24, the annual Cloud Summit served as a key window into Alibaba’s technological core. This year, the company showcased the full chain of AI implementation — from the computing foundation of chips and cloud infrastructure, to the continuous evolution of model capabilities and services, and finally to Agentic applications deployed in real production environments.

AI’s Leap from Large Models to Agent Era

As AI transitions from large models to the AI Agent era, many are focused on AI applications. But Alibaba’s ambitions go further. In the e-commerce era, the company aimed to build the "water, electricity, and gas" of the digital economy. In the AI era, Alibaba is anchoring a full-stack AI strategy, striving to become the global "water, electricity, and gas" of the AI age.

Alibaba CEO Eddie Wu stated that the three cornerstones of the "machine intelligence" era are AI models, AI chips, and AI cloud — the prerequisites for scalable machine thinking. Alibaba is increasing capital expenditure on AI infrastructure.

Organizational Shifts and Leadership Changes

Beneath this clear strategic主线, another thread of frequent change is visible: accelerated organizational restructuring and turnover in AI technical leadership.

• Liu Dayiheng has been appointed as the head of the Qwen LLM project under Alibaba’s ATH Business Group’s Token Foundry. Announced on the eve of the Cloud Summit, this is widely seen as his formal succession to the Qwen large language model project, six months after Lin Junyang’s departure.
• Chen Yusen, born in 1992, took over DingTalk and officially launched Qwen Office, entering the fiercely competitive AI office productivity market.

With inference demand surging, super nodes in short supply, and Agentic applications accelerating across industries, 2026 is being called the first year of AI entering real production environments. For Alibaba, the challenge lies in balancing capital expenditure strength with agile personnel deployment.

Data Center Roadmap: Over 20GW by 2032

Alibaba announced a global data center construction target of over 20GW by 2032, underscoring its commitment to AI infrastructure.

Following the announcement, foreign investment banks offered varied interpretations:

• Citi raised its capital expenditure (Capex) forecasts for Alibaba over the next three years. Analysts including Alicia Yap noted that based on Wu’s remarks about "investment through industrial cooperation," Citi assumes Alibaba will share costs with partners. Citi now expects Capex of 258 billion RMB (2027), 283 billion RMB (2028), and 282 billion RMB (2029) — up 13%, 41%, and 57% respectively from previous estimates.

• Goldman Sachs estimates that expanding from the current ~5–6GW to 20GW by 2032 implies an annual addition of 1–2GW, corresponding to annual Capex of 200–300 billion RMB. Goldman forecasts Alibaba’s Capex for FY2027–2029 at 209 billion, 243 billion, and 260 billion RMB respectively. It emphasized that this scale is manageable, achievable through a mix of self-built heavy assets, light-asset partnerships, and leasing, funded by stable e-commerce EBITA, cloud profit growth, cash reserves, and future financing.

In February 2025, Alibaba announced a 380 billion RMB AI investment plan. By the end of Q2 2026, total investment had reached 190 billion RMB.

"All current AI monetization methods — whether AI software subscriptions, large model API inference, GPU services, or AI training and inference software — are fundamentally built on AI computing centers. Only by building AI computing centers can we achieve rapid market share growth." — Eddie Wu, August 2025 earnings call

At the Cloud Summit, an Alibaba Cloud staff member told National Business Daily: "Super nodes are in short supply. We build them as soon as we have chips, and customer demand follows suit."

In early August, Alibaba Data Center General Manager Wang Chaoyang noted in Ulanqab: "Every single GPU in our data centers is busy — either training models or producing tokens. Faster data center delivery means greater customer value."

Wu also acknowledged the current bottleneck: "Customer AI demand is very strong. Alibaba Cloud is ramping up AI computing capacity as fast as possible, and revenue growth is accelerating. But medium- to long-term demand far exceeds supply. Global shortages in the AI data center supply chain are limiting growth."

Leadership Changes Amid Strategic Focus

When large model parameters exceed 2 trillion, super-node architectures enable better GPU-to-GPU communication. Alibaba’s wholly owned semiconductor subsidiary, T-Head, has launched the Zhenwu AI chip, now online in Alibaba Cloud super nodes.

At this year’s Cloud Summit, T-Head released the Zhenwu V900, a training-inference integrated AI chip. Mass production and sales are expected in Q1 2027, supporting trillion-parameter model training and inference.

Under the AI wave, domestic chip substitution is accelerating. T-Head’s AI chip annual shipments are set to rise significantly. As of April 2026, cumulative shipments had reached 560,000 units.

Founded in 2018, T-Head has become a pillar of Alibaba’s AI strategy.

Over the past six months, Alibaba has rapidly restructured around AI:

• March 2026: The Alibaba Token Hub business group was established, led directly by Eddie Wu.
• June 8, 2026: Token Foundry was formed by merging the former Tongyi Qianwen Large Model Business Unit and the Future Life Lab.

Alongside organizational changes, Alibaba is promoting young technical leaders. Chen Yusen, after taking over as DingTalk CEO, entered the AI office market with Qwen Office. At the Cloud Summit, he emphasized that enterprise-grade Agent deployment hinges on data, with connectivity, understanding, and reuse as three key points.

Large language models and AI office applications are two critical strategic high grounds in Alibaba’s full-stack AI strategy. These new leaders must prove their ability to drive technical teams, achieve product breakthroughs, and demonstrate Alibaba management’s agility and foresight.

After all, Alibaba’s all-in AI push is not only a global race but also subject to capital market scrutiny.

On August 20, Alibaba Group released its Q1 FY2027 earnings. In the quarter, AI commercialization accelerated across the board:

• Alibaba Cloud’s external commercial revenue grew 45% year-over-year, the fastest in 22 quarters.
• AI-related product revenue achieved triple-digit year-over-year growth for the 12th consecutive quarter.

If 2026 marks the first year of AI entering real production environments — shifting from "being able to talk" to "being able to work" — then for Alibaba, which pursues a full-stack self-developed AI path, the key challenges are: enabling AI to deliver tangible results across industries, bridging the last mile of enterprise AI adoption, and matching computing supply with AI demand growth.

(Source: National Business Daily)