Two data center billionaires created before anything was even built. A borrower seeking a loan for 150% of the construction cost. And companies that use financial engineering to keep their liabilities off their balance sheet.
For skeptics, these are just a few examples of why the artificial intelligence data center boom is spiraling out of control.
There is a development frenzy to support the AI revolution, and with it an insatiable demand for debt to finance it. Some estimate the overall cost of infrastructure deployment could reach $10 trillion, and with so many lenders lining up to pump money into assets, there are fears a bubble is forming that could eventually leave equity and credit players facing considerable difficulties.
“One of the key risks to consider is the possibility that the data center construction boom could lead to a glut. Some data centers could become unprofitable and some owners could go bankrupt,” Howard Marks, co-founder of Oaktree Capital Management LP, wrote in a note this week. “We will see which lenders maintain their discipline in today’s heady environment.”
Given the influx of money coming in, another danger is that there will be less credit available when facilities currently built with loans need to be refinanced in three to five years.
There are also growing concerns about debt levels, especially as technology may fail to meet high expectations. In such a scenario, lenders might be even more reluctant to refinance, and companies would have to find additional equity capital or pay more to borrow.
“Momentum is strong, but if it’s irrational exuberance, investors will lose when the music stops,” said Sadek Wahba, president and managing partner of infrastructure investor I Squared Capital. He said his company tries to be careful, warning that “every deal has nuances and the fine print matters.”
The broader AI universe has also been caught up in the worries, with circular trading and soaring valuations undermining the bullish sentiment that once dominated.
At Brookfield, CEO Bruce Flatt plans spending of $5-10 trillion to fund the deployment of AI in everything from data centers to power infrastructure. McKinsey & Co. estimates that nearly $7 trillion will be needed by 2030 just for data centers, including those for AI.
“These are amounts that have never been invested before,” Flatt said.
OpenAI, for example, plans to spend $1.4 trillion on AI infrastructure – and would spend more if it could. Sarah Friar, chief financial officer, has repeatedly said the company’s only constraint is finding more IT capacity.
While the scale of transactions is one concern, another is how they are presented and structured.
Lenders slice and dice the debt and resell it to other investors, meaning it becomes increasingly opaque, according to Vinay Nair, CEO of fintech platform TIFIN and a teacher in executive education programs at the Wharton School.
“You spread that risk across the system,” he said. If there is a decline, “I don’t think we fully understand all the ripple effects of that through that credit channel.”
Some borrowers have removed AI data center risk from their balance sheets by using securitization markets, where debt is divided into tranches with varying risks and returns and purchased by insurers and pension funds. A similar story emerges in the graphics processing units that process data.
With such a positive lending environment, some borrowers are even asking for more than 100% of the construction cost of projects, according to two private lenders, who asked not to be identified because the details are confidential. In one case, the demand was 150%, with the real estate developer justifying the request based on how much the value of the facility would rise when rents started rolling in, one of the people said.
At the same time, there is also a risk of hype. Nuclear startup Fermi Inc. has yet to develop any data centers, but its valuation briefly soared to more than $19 billion when it listed this year. That made founders Toby Neugebauer and Griffin Perry, sons of former U.S. Energy Secretary Rick Perry, billionaires.
But there is also growing market nervousness about borrowing and spending.
Fermi has fallen back below the level at which it was made public. Concerns about spending at Facebook parent Meta Platforms Inc. hit its stock in late October and that of Oracle Corp. collapsed this week after the company announced an increase in its investments in data centers and other equipment.
Financing plans
For years, owners financed data centers with a combination of equity and debt and leased the space. Hyperscalers, large cloud computing providers like Microsoft Corp. and Alphabet Inc.’s Google, also developed sites themselves as cloud services took off.
Today, companies want to continue to expand and maintain control over their capabilities, but they are increasingly structuring their transactions to reduce their impact on the financial statements, which helps limit the risk of them being considered overexposed.
Hyperscalers are starting to use so-called synthetic leases, which limit the liabilities that appear on their balance sheets while allowingthey will receive tax relief on the depreciation, according to Jeffrey Shell, vice president of corporate capital markets at CBRE.
Previously, tech giants were content to write their own checks “because they need to move quickly to have first-mover advantage,” Shell said. “At some point, even for the largest companies, funding at these levels has a significant impact on the balance sheet. »
As borrowing soars, credit markets must adapt to meet demand.
“The size has now surpassed what you’re realistically going to put into CMBS, ABS and the private placement project bond market,” said Scott Wilcoxen, global head of digital infrastructure investment banking at JPMorgan Chase & Co.. “He’s going to take them all.”
At least $175 billion in U.S. data center-related credit deals have been made this year so far, according to figures compiled by Bloomberg News. Oaktree’s Marks questions the returns on debt sold by hyperscalers to finance AI investments. Play video
The spread is sometimes only about 100 basis points higher than that of U.S. Treasuries, leading seasoned investors to wonder whether it is “prudent to accept 30 years of technological uncertainty to make a fixed-income investment that yields little more than risk-free debt?”
And not everyone is a fan of the design of certain vehicles that investors are asked to put money into.
“We’ve seen master trust structures where assets can be rotated every few years,” said Michelle Russell-Dowe, co-head of private debt and credit alternatives at Schroders Capital. “It’s hard to subscribe to, so we don’t like it.”
Mentions of bubbles have piqued the interest of regulators. The Bank of England is reviewing data center lending after raising concerns about the level of spending and funding.
According to JPMorgan’s Wilcoxen, a phrase that keeps popping up in the market to describe the vast expanse of financing being tapped is “everything everywhere at once,” a riff on the recent Oscar-winning film.
“The amount of money required for all of this is extraordinary,” he said.
