China's Alibaba Group is accelerating its AI infrastructure investment. In its quarterly earnings report for the April–June 2026 period, announced on August 20, 2026, the company reported that capital expenditure on AI data centers reached approximately $10 billion, up 75% year-over-year.
Addressing concerns that massive infrastructure spending would squeeze profits, CFO Toby Xu presented a clear financial model. According to the company, its servers have an operational lifespan of five years, and revenue generated by AI servers fully recovers costs within the first three years. The remaining fourth and fifth years become a period during which the hardware generates pure free cash flow. CEO Eddie Wu noted that some servers equipped with Nvidia V100 and A100 chips, deployed back in 2018 and 2020, are still running at nearly full capacity for customers today—demonstrating that even older-generation GPUs can continue generating substantial revenue as long as they are matched to the right use cases.
The company now plans to shorten this initial payback period even further, to 2.5 years. Its primary lever for achieving this is reducing reliance on commercial chips and increasing the proportion of self-developed chips deployed in its data centers. Amid a global shortage of computing power, commercial AI chips currently carry extremely high gross margins, which has been the single biggest factor squeezing cloud vendors' profits. Switching to in-house chips directly boosts both the price competitiveness and profit margins of Alibaba's own services.
"Zhenwu" and "Yitian" Underpin an In-House Ecosystem
Alibaba is rapidly strengthening its proprietary silicon foundation through T-Head (平頭哥), its semiconductor development division. In addition to rolling out its general-purpose server CPU "Yitian" (倚天) 710, the company's current top priority is expanding its latest chip series for AI inference and autonomous AI agents, "Zhenwu" (真武).
The latest Zhenwu M890, announced in May 2026, is equipped with 144GB of high-bandwidth on-chip memory and is specifically designed to handle the large-scale context processing demanded by next-generation AI agents. Its computing performance has roughly tripled compared to the previous generation, and shipments are already said to be underway at a scale of hundreds of thousands of units. Due to U.S. export restrictions on advanced AI hardware, Chinese companies remain limited in their access to Nvidia's latest architectures and continue to rely on downgraded alternatives such as the H20. Against this backdrop, the agent-focused Zhenwu M890 is positioned as a strong rival to expensive commercial chip alternatives.
"Self-developed chips are a long-term and critically important direction for us," Wu emphasized. As production capacity for in-house chips improves, they will increasingly replace commercial chips within data centers. Making high-performance domestic chips the standard equipment in its own data centers has become essential not only for boosting margins but also for avoiding U.S. regulatory risk and stabilizing the supply chain.
E-Commerce's Growth Ceiling and the Bet on AI
Behind this massive infrastructure investment lies a clear growth ceiling in Alibaba's founding business, e-commerce. While the long-running, often irrational price war with PDD Holdings (Pinduoduo/Temu) and Douyin (the domestic Chinese version of TikTok) shows signs of easing, the maturing of the market itself meant e-commerce revenue this quarter reached only $30.34 billion—growth of just 4% year-over-year.
By contrast, revenue from AI services within the cloud division grew 45%. Wu declared that AI has become the company's "most reliable growth engine." Total quarterly revenue for the cloud business reached $7.14 billion, with the company projecting it will cross the $10 billion mark in the coming quarter.
However, it's worth noting that a significant portion of this cloud revenue comes from internal use within the Alibaba Group itself. Massive internal demand—from AI-powered product recommendations on the group's own e-commerce platforms to routing optimization at its logistics arm, Cainiao (菜鳥)—directly supports the cloud division's revenue. Rather than prioritizing profit from external customer sales, a large internal cycle is at work: thoroughly optimizing the company's own massive business operations with AI first, then absorbing the resulting operational costs through in-house chips.
The Barrier to Winning External Customers and What to Watch Next
Alibaba claims it has shortened AI data center construction time to just 100 days, pushing its infrastructure deployment speed to a world-leading level. But whether this enormous infrastructure can be filled with external customer workloads remains an open question.
During its earnings announcement, the company disclosed that more than 650 external customers are now using cloud resources powered by Alibaba's own chips. While this represents progress, the scale gap becomes stark when compared to rival AWS, which has more than 120,000 customers using its general-purpose Graviton processors. AWS built its Graviton ecosystem over many years through software optimization and migration support, but in the AI chip space, Nvidia's CUDA ecosystem stands as a formidable wall—making the hurdle for customers to migrate to proprietary hardware even higher than it is for general-purpose CPUs.
Furthermore, Western government agencies are increasingly restricting or discouraging the use of Chinese cloud services, while Europe continues to raise demands for sovereign cloud solutions that prioritize data sovereignty. The path for Alibaba Cloud to expand its proprietary chip ecosystem outside China remains a difficult one.
The key question going forward is whether this high-margin model built around in-house chips will remain merely an internal cost-cutting tool for Alibaba itself, or whether it can evolve into a platform that attracts external AI developers through overwhelming cost-performance advantages. Whether the bold financial scenario of a 2.5-year payback period becomes reality will ultimately be proven by how quickly the company can expand its external customer base.