Alibaba CEO Warns of Overheating AI Data Center Market

Alibaba Group Chairman Joe Tsai has warned that more AI data centers may be built than necessary. He said at the recent HSBC Global Investment Summit in Hong Kong that the current pace of data center construction could outpace demand for AI services, Bloomberg reports.

According to the publication, the businessman said that he was “starting to see a bubble of some kind.” He noted that some construction projects had already started without preliminary agreements with future clients, without guaranteed demand. Tsai expressed concern that data centers were being built “at random,” at the initiative of people and funds that had appeared in the industry, with millions or even billions of dollars involved.

In particular, he drew attention to the practice of building data centers in the United States. Investment commitments in the relevant sector of the country are already quite significant. This year, the Stargate project was launched with the participation of OpenAI, Oracle and SoftBank, within the framework of which they are going to spend up to $500 billion in the future, including on the construction of data centers throughout the United States.

Meanwhile, other investors are planning large-scale investments in building AI infrastructure in 2025, according to Datacenter Dynamics. Microsoft has promised to spend $80 billion, despite the fact that it intends to abandon 200 MW AI data center projects. Amazon (AWS) is preparing to spend $100 billion, Google – $75 billion, and Meta✴ – $60-65 billion.

Image source: Ben Allan/unsplash.com

Cai said he was still amazed by the level of investment in AI in the U.S. He said outright that people were literally talking about $500 billion, which is a completely excessive amount. He said people may be wasting their money on the idea that demand will grow significantly in the future.

The belief that AI projects are a big “soap bubble” has existed for a long time among some experts, but it worsened by January 2025, against the backdrop of the emergence of economical and efficient AI models from the Chinese startup DeepSeek, which managed to train the R1 model using not the most advanced accelerators, for a relatively small cost compared to the products of OpenAI and Meta✴, according to some sources – about $5.6 million (although the sensational nature of the news is somewhat exaggerated).

Following news of the triumph of Chinese AI technology, the segment of the US stock market associated with artificial intelligence rapidly lost about $1 trillion, $600 billion of which came from NVIDIA shares.

This month, Alibaba unveiled the 32 billion-parameter QwQ-32B AI model, which it claims matches the performance of the DeepSeek-R1 but requires even less computing power.

China’s Tencent then announced it would slow the rollout of accelerators, though its capital expenditures would still account for more than 10% of its revenue, which was $91.1 billion in fiscal 2024, up 8% year-on-year.

Despite Tsai’s words and the more cautious approach of his rivals to investments, Alibaba Group announced earlier this year that it would spend $53 billion on cloud and AI infrastructure over the next three years, the highest level of investment for the company to date.