US banks insulated from AI data centre downturn
Tougher underwriting standards adopted since the global financial crisis will enable banks to weather a sharp reversal in the AI boom, according to analysts, following a challenging week for stocks connected to the technology.
Global share indices fell in the second half of last week over concerns about the sustainability of this year’s AI-driven rallywith some traders unwinding leveraged bets that have been magnified by using substantial amounts of debt.
Shares posted a slight recovery on Monday. Google owner Alphabet Inc, which will on Wednesday become the first of the AI “hyperscalers” to report second-quarter results, saw its shares rise on a report it is developing a server chip designed to optimise its Gemini AI model.
Analysts note that any slowdown in the data centre construction boom would not lead to bank losses comparable to the subprime mortgage crash.
This is because banks have limited direct construction loans made to data centres themselves, and have instead focused on supporting the subcontractors that build them, such as trucking, plumbing and electric companies.
“The banks learnt some very hard lessons in 2008/09,” Gerard Cassidy, head of US bank equity strategy at RBC Capital Markets, told The Banker. “They have established conservative underwriting standards and they won’t take the excess risk of underwriting these types of large [data centre] buildings.”
He added that bank leaders’ answers about lending to data centres during second-quarter earnings calls last week reveal cautious policies.
JPMorgan chief financial officer Jeremy Barnum told analysts last week that the bank had avoided some data centre deals as they were just too risky.
“We saw some deals come through where we were just like, ‘yeah, we’re not doing that’,” he said.
A key risk banks face with data centres is whether the tenant (such as Alphabet or Meta) guarantees the loan to build the data centre, according to Cassidy.
“If they say ‘no’, the banks will say ‘we don’t want to make the loan because the tenant could walk in five years and there’s nobody to pay the loan back’,” he said.
Brendan Browne, managing director, S&P Global Ratings, said that most data centre finance has thus far come from the capital markets, rather than banks.
“Public disclosures are limited, but we believe on-balance-sheet exposures to data centres that rated banks have account for a very small portion of their loans, and tend to be investment grade in nature,” he said.
Banks’ underwriting practices have limited direct exposure to data centres, according to Browne.
Some banks with asset management arms can fund data centre construction loans off balance sheet, while some rated banks have also set internal limits on how much data centre exposure they want, he said.
Wells Fargo’s chief executive Charlie Scharf last week said there are “others” willing to take “more risks” than his bank when lending to data centres or being involved in “some of the strategic transactions” within the AI sector.
He highlighted key factors in data centre construction ranging from the choice of chip used and how energy is generated, noting that banks have to understand the different types of risks when they underwrite such projects.
RBC’s Cassidy says that one longer-term question is the indirect exposure banks have to a downturn in the AI boom from their loans to cement, plumbing and electricity companies.
If these suppliers to data centres suffer a recession due to a slowdown in AI, banks will face problems.
“The real question is what the exposure here is, and the banks do not know themselves,” Cassidy said.
Building data centres, computing power and energy are at the heart of the current AI capital expenditure boom that Goldman Sachs estimates will be worth roughly $7.6tn between 2026 and 2031.
Yet the speed and scale of data centre construction is causing a political backlash in the US. New York state governor Kathy Hochul last week announced the first statewide ban on the construction of new data centres in the US.
DOCUMENTED REFERENCES
Exploring Documented Records
Public interest in the Epstein case continues not only because of court proceedings and testimonies, but also due to the growing body of documented records that help researchers and readers understand the broader context. Beyond legal files and media reports, some independent projects have organized publicly available data connected to Epstein’s activities.
One example is a structured archive of documented Amazon order records, where purchases are cataloged with dates and product details. While individual items do not prove wrongdoing on their own, examining documented information alongside established facts helps paint a clearer picture of the environment and circumstances surrounding the case.
For readers looking to review primary-source style data rather than interpretations, exploring compiled records can provide additional context to the broader discussion.
