Alibaba Group chairman Joe Tsai has warned that there may be an over-provisioning of data center buildout for AI.
During the HSBC Global Investment Summit in Hong Kong today, March 25, Tsai warned that the current pace of construction could outpace the initial demand for AI services.
As reported by Bloomberg, Tsai said: "I start to see the beginning of some kind of bubble.” He noted that there are projects beginning that haven't yet secured uptake agreements. “I start to get worried when people are building data centers on spec. There are a number of people coming up, funds coming out, to raise billions or millions of capital.”
Tsai particularly drew attention to spending on AI data centers in the US.
Investment commitments in the US have, indeed, been significant. This year so far has seen the launch of Stargate, a $500 billion project between OpenAI, Oracle, SoftBank, and investment firm MGX that would see massive data centers developed across the US.
Meanwhile, hyperscalers are planning massive capex investments throughout 2025 into AI data center infrastructure, with Microsoft pledging $80 billion in spending - which it remains committed to despite a TD Cowen report that it had canceled 200MW of AI data center projects, Amazon $100 billion, Google $75 billion, and Meta expecting between $60 to 65 billion.
“I’m still astounded by the type of numbers that are being thrown around in the United States about investing into AI,” Tsai told the audience.
“People are talking, literally talking about $500 billion, several 100 billion dollars. I don’t think that’s entirely necessary. I think in a way, people are investing ahead of the demand that they’re seeing today, but they are projecting much bigger demand.”
Concern that AI might see a "bubble" in a similar way to the dot-com bubble of the late 90s to early 2000s has long been present, but have been exacerbated by the January 2025 claims that AI company DeepSeek had managed to train its R1 LLM using far less advanced chips and for a fraction of the cost, approximately $5.6 million, compared to LLMs developed by the likes of OpenAI and Meta.
Following that news, around $1 trillion was wiped off the US stock market, $600 million of which was experienced by Nvidia.
This month, Alibaba released a 32-billion parameter AI model - QwQ-32B - which it claims can match DeepSeek's R1, with even smaller compute requirements.
Since then, Tencent has said it will slow down its rollout of GPUs, though it still plans a capex in the "low teens" of its revenue. FY 2024's revenue was $91.1 billion, up eight percent year over year.
Despite this, and Tsai's words of warning, earlier this year Alibaba Group revealed that it was expecting to spend around $53 billion in cloud and AI infrastructure in the next three years in what will be the company's "most concentrated and highest level of investments."
