In a pointed letter to investors, Alex Davis, the founder of Disruptive, asserted that the notion “data centers will find tenants once built” is a fallacy, as an excessive number are being constructed without any guaranteed occupancy rates.

The institution, which recently spearheaded investment in the AI chipmaker Groq, predicts that speculative data center owners will confront financing crises by 2027 - 2028. It is also worried about the strain these speculators are putting on the entire ecosystem.

The AI boom has catapulted global data center construction into an era of unprecedented rapid expansion. Analysts at Morgan Stanley estimate that by 2035, global data center capacity will need to increase six - fold to meet the demands of cloud computing and AI.

This indicates that infrastructure investments in this area are projected to reach $3 trillion from 2025 to 2028.

Tech behemoths are at the forefront of this construction frenzy. In 2025 alone, Amazon, Alphabet (Google’s parent company), and Meta pledged to invest $100 billion, $75 billion, and $65 billion, respectively, in AI infrastructure.

Alibaba Chairman Joe Tsai commented, “The figures being bandied about for AI investment in the U.S. still leave me stunned.”

Financial institutions are also heavily involved. Traditional banks such as JPMorgan Chase and Deutsche Bank, along with private credit firms like Blackstone Group and Apollo Global Management, are continuously injecting funds into new data center projects.

Moody’s Investors Service reports that major tech companies are rapidly constructing and leasing new data centers while also expanding into newer, smaller markets.

However, beneath this surface - level prosperity, the data center industry’s business model harbors multiple structural risks. As the industry transitions from a scale - based competition to an efficiency - driven one, high leverage and asset quality become crucial for surviving market cycles.

The business models of new cloud providers are highly vulnerable. These companies often rely heavily on debt to lease land and deploy GPUs, with the debt secured by long - term computing power leasing contracts signed with customers.

The problem lies in the mismatch of contract terms. Computing power leasing contracts between new cloud providers and customers typically last four to five years, while their data center leases often extend for fifteen years or more.

This implies that once customers are lost, these providers will inevitably suffer losses.

Meanwhile, larger AI enterprises or cloud service demanders are increasingly choosing to build their own data centers, posing future competition to the data center leasing market from their own customers.

There is a divergence within the industry regarding the sustainability of the current construction boom. On one hand, companies like Microsoft and Meta stress that AI demand far outstrips supply and plan to continue increasing their capital expenditures.

On the other hand, warnings from multiple sources, including Disruptive and Joe Tsai, are growing more strident. Tsai attributes bubble risks to three types of corporate behavior: model - centric companies like OpenAI, infrastructure - focused hardware suppliers, and integrated tech firms.

He believes that some investments by the first two categories lack sufficient synergy.

The Chinese market also shows signs of cooling. Data reveals that as of the first half of 2024, the actual operational status of smart computing centers in China is worrying, with average cabinet utilization rates at a mere 20% - 30%, and some enterprise - level centers as low as 10%.

In the first quarter of 2025, 165 smart computing center projects in mainland China saw new developments, but only 16 were in production or trial operation.

The real risk may not originate from a weakening in AI demand. An analysis report points out that the greatest risk facing data centers is the possibility of lenders raising interest rates amid increased tenant pressure and tightening credit. This would directly impact data center profitability and trigger a chain of negative consequences.

The deep involvement of the private credit market heightens systemic risks. JPMorgan Chase CEO Jamie Dimon cautioned, “Private credit could touch off the next financial crisis.”

With AI infrastructure investment now exceeding 2% of U.S. GDP—surpassing the peak of the 2000 telecom boom and still on the rise—historical lessons call for caution.

The lessons from the 2000 Internet bubble and the 2008 global financial crisis serve as a reminder to stay vigilant about financial and structural risks in the expansion of AI infrastructure.

Regulatory authorities have started to take action. Recently, multiple regions will carry out comprehensive surveys of computing power resources. This government - led coordinated initiative aims to curb the chaotic trend of blind data center expansion.

Against the backdrop of industry - wide pressure, this type of in - depth regulatory oversight underscores a strict scrutiny of project risk resilience.

As the 2027 - 2028 financing crisis warning period draws near, the data center industry stands at a crossroads. Whether this construction frenzy driven by the AI revolution will lead to a rational correction or a crisis outbreak depends on the industry’s ability to balance financial stability with technological foresight.

References:

https://www.cls.cn/detail/2244028

https://36kr.com/p/3221562374343557

https://www.stcn.com/article/detail/1678821.html

https://news.futunn.com/en/post/66769640/major-investors-in-data-centers-have-expressed-deep-concerns-a?level=2&data_ticket=1766116194545042