Two data center billionaires were born before anything was built. The borrower seeks a loan for 150% of the construction cost. and companies that use financial engineering to take liabilities off their balance sheets.
For skeptics, these are some examples of the AI data center craze getting out of hand.
The development frenzy to support the AI revolution is underway, and with it comes an insatiable demand for the debt to fund it. Some estimate the overall cost of infrastructure construction could reach $10 trillion, and with so many lenders lining up to pour cash into the assets, there are concerns a bubble is forming that could ultimately cause significant pain for equity and credit players.
“A key risk to consider is that the boom in data center construction could lead to oversupply. Some data centers could become uneconomic and some owners could go bankrupt,” Howard Marks, co-founder of Oaktree Capital Management, wrote in a note. notes This week. “We will see which lenders remain disciplined in today’s exciting environment.”
Given the influx of money, another danger is that available credit will shrink when facilities currently being built with loans need to be refinanced in three to five years.
People are becoming more and more worried Leverageespecially considering that the technology may perform below its high expectations. In this case, lenders may be less willing to refinance and the company will have to find additional equity or pay more to borrow.
“There’s momentum, but if it’s irrational exuberance, then when the music stops, investors will suffer,” said Sadek Wahba, chairman and managing partner of infrastructure investor I Squared Capital. He said his firm was trying to be cautious, warning that “there are nuances to every deal and the fine print matters.”
The broader field of artificial intelligence is also mired in concerns, recurring transaction and soaring valuations bullish sentiment once dominant.
exist BrookfieldChief Executive Bruce Flatt expects to spend $5 to $10 trillion to fund the rollout of artificial intelligence in everything from data centers to power infrastructure. McKinsey & Co. Estimate Nearly $7 trillion By 2030, this technology will be needed in data centers alone, including data centers for artificial intelligence.
“These are amounts that have never been invested before,” Flatt said.
Take open artificial intelligence as an example, Have a plan $1.4 trillion spent on AI infrastructure — we’d be willing to spend more if we could. Chief Financial Officer Sarah Friar has repeatedly said the company’s only limitation is finding more computing power.
If deal size is one concern, another is the packaging and structure of the deal.
Vinay Nair, CEO of fintech platform TIFIN and a faculty member in Wharton’s executive education program, said lenders are slicing and dicing debt and selling it to other investors, meaning debt is becoming increasingly opaque.
“You’re spreading that risk through the system,” he said. If there is a decline, “I don’t think we fully understand all the knock-on effects of the credit channel yet.”
Some borrowers have been taking the risk off their balance sheets by using the securitization market, in which debt is split into tranches with varying risks and returns and purchased by companies such as insurance companies and pension funds. A similar situation occurs in graphics processing units that process data.
Two private lenders said that because the lending environment was so positive, some borrowers were even asking to cover more than 100% of the project’s construction costs. The agency spoke on condition of anonymity because the details are not public. In one case, a 150% payment was requested, which the property developer justified on the basis that facility valuations rose when rents started flowing, one of the people said.
At the same time, there is also the risk of hype. Nuclear power startup Fermi Inc. has yet to develop any data centers, but its valuation briefly jumped to more than $19 billion when it went public this year. Founder Toby Neugebauer and Griffin Perry, son of former U.S. Energy Secretary Rick Perry, both became billionaires.
But concerns about borrowing and spending are also growing.
Fermi’s share price has fallen back below the levels it was at when it went public. Concerns about Facebook parents meta platform company expenses hit its stock late October and Oracle corps dejectedly This week, the company reported a significant increase in investment in data centers and other equipment.
financing plan
Over the years, owners have financed data centers and leased space through a combination of equity and debt. Hyperscale enterprises, large cloud computing providers, e.g. Microsoft Corp. and Alphabet Inc. GoogleWith the rise of cloud services, they have also developed their own websites.
Now, companies want to continue adding capacity and maintaining control over it, but are increasingly structuring transactions to reduce the impact on financial statements, which can help limit the risk that they are seen as over-exposed.
Hyperscalers are beginning to use so-called synthetic leases, which limit the liabilities that appear on the balance sheet but still allowThey could reportedly benefit from depreciation tax breaks Jeffrey SchellVice Chairman, Corporate Capital Markets, CBRE.
Tech giants would previously just write their own checks “because they needed to move quickly to gain first-mover advantage,” he said. shell. “At some point, even for the largest companies, these levels of financing will have a meaningful impact on the balance sheet.”
As borrowing surges, credit markets must adjust to cope with demand.
“It’s beyond what you can actually put into the CMBS, ABS and private placement bond markets,” he said. Scott Wilcoxon, JPMorgan Chase & Co. Global Head of Digital Infrastructure Investment Banking. “This will take them all away.”
So far this year, at least $175 billion in data center-related credit deals have been struck in the United States, according to data compiled by Bloomberg. Oaktree’s Marks questions the yields on debt sold by hyperscale companies to finance investments in artificial intelligence. play video
At times, spreads were only about 100 basis points higher than Treasuries, leaving investment veterans wondering, “Is it wise to accept 30 years of technical uncertainty and make fixed-income investments that yield only a little more than risk-free debt?”
Not everyone likes the design of some of the vehicles investors are being asked to invest in.
“We’ve seen major 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 difficult to underwrite, so we don’t like those.”
References to bubbles piqued the interest of regulators. The Bank of England is reviewing Loans to data centers This follows growing concerns about spending and funding levels.
J.P. Morgan’s Wilcoxon said a word that keeps popping up in the market to describe the huge financing being leveraged is “everything happening at once,” a riff on the recent Oscar-winning movie.
“The amount of money that goes into pursuing this is staggering,” he said.