Alibaba's Joseph Tsai warns: The construction of AI data centers by American manufacturers has shown signs of a "bubble," with many projects lacking clearly defined customers during construction
Joseph Tsai stated that he began to feel concerned when people started building data centers without a clear need. "The numbers regarding investment in AI in the U.S. still shock me. People are literally talking about $500 billion, hundreds of billions of dollars. I think this is completely unnecessary." "People's investments are ahead of the demand we see today, but they predict that the demand will be much larger."
Alibaba Group Chairman Joseph Tsai warned that the construction of AI data centers by American manufacturers may be forming a bubble, and the scale of hundreds of billions of dollars in AI investments by global tech giants is concerning.
On the 25th, Joseph Tsai stated at the HSBC Global Investment Summit held in Hong Kong that large tech companies, investment funds, and other institutions are competing to establish AI training server bases, and this trend is starting to appear somewhat blind. Many projects do not have clear customers during construction.
"I am beginning to see the seeds of some kind of bubble," Joseph Tsai told the attendees. He also added that some planned projects have started raising funds without having obtained "demand" agreements:
“When people start building data centers without clear demand, I begin to feel concerned. Many people are emerging, and funds are being established to raise billions or even millions of dollars.”
Joseph Tsai: The scale of American manufacturers' investments is disconnected from actual demand
Alibaba is set to make a strong recovery in 2025, partly due to the popularity of its Qwen large model, with the company expecting this platform to drive the development of Alibaba's core e-commerce business and cloud services.
From Microsoft to SoftBank Group, tech companies on both sides of the Pacific are spending billions of dollars to purchase Nvidia and SK Hynix chips, which are crucial for AI development. Alibaba itself announced in February that it would fully invest in AI — planning to invest over 380 billion yuan (approximately 52 billion USD) in the next three years.
In the United States, Trump is promoting a project called "Gateway to the Stars," which is expected to invest 500 billion USD.
Joseph Tsai criticized the spending of American competitors:
" The numbers regarding AI investment in the U.S. still shock me. People are literally talking about 500 billion USD, hundreds of billions of dollars. I think this is completely unnecessary.
To some extent, people's investments are ahead of the demand we see today, but they predict that the demand will be much larger."
The "AI Cash Burn Competition" among global tech giants
In just 2025, Amazon, Alphabet (Google's parent company), and Meta Platforms have each committed to investing 100 billion USD, 75 billion USD, and up to 65 billion USD in AI infrastructure, respectively.
However, in February, analysts at TD Cowen pointed out that Microsoft had canceled some leases for data center capacity in the U.S., raising concerns about whether Microsoft had acquired more AI computing power than its actual needs in the long term.
Microsoft executives downplayed these concerns, stating that the company's current spending is higher than at any time in its history, and this spending is primarily on chips and data centers. The company expects to spend 80 billion USD on AI data centers this fiscal year, but the growth rate of this spending should begin to slow down starting in the new fiscal year beginning in JulyAt the same time, the low-cost open-source model released by AIDeepSeek in China at the beginning of the year has raised doubts on Wall Street about the rationality of large-scale investments. Critics also pointed out that the application of AI in the practical, real world remains persistently lacking.
Affected by this news, today, A-share AI computing power concept stocks fell sharply, and most Hong Kong technology stocks declined, with Alibaba dropping over 3%.
Meanwhile, Goldman Sachs' analyst team has downgraded the computing power for training endpoints and the shipment volume of AI training servers, but remains optimistic about the prospects for AI inference computing power. The expected shipment volumes for 2025 and 2026 have been revised down from 31,000 units and 66,000 units to 19,000 units and 57,000 units (based on 144-GPU equivalent calculations). This adjustment is mainly due to the impact of the product transition period and uncertainties in supply and demand