AI frenzy: An employee checking on servers in a data centre. Funding from investment-grade and high-yield bond markets, as well as up to US$40bil per year in data-centre securitisations, will still be insufficient to meet demand, say strategists. — Bloomberg
NEW YORK: The furious push by artificial intelligence (AI) hyperscalers to build out data centres will need about US$1.5 trillion of investment-grade bonds over the next five years and extensive funding from every other corner of the market, says an analysis by JPMorgan Chase & Co.
“The question is not which market will finance the AI-boom? Rather, how will financings be structured to access every capital market,” said strategists led by Tarek Hamid.
Leveraged finance is primed to provide around US$150bil over the next half decade, they said.
Even with funding from the investment-grade and high-yield bond markets, as well as up to US$40bil per year in data-centre securitisations, it will still be insufficient to meet demand, the strategists added.
Private credit and governments could help cover a remaining US$1.4 trillion funding gap, the report estimates.
The bank calculates an at least US$5 trillion tab that could climb as high as US$7 trillion, single-handedly driving a reacceleration in growth in the bond and syndicated loan markets, the strategists wrote in a report Monday.
The analysts project US$300bil in high-grade bonds going toward AI data centres next year. That could account for nearly one fifth of total issuance in that market, which a report from Barclays Plc estimates will grow to US$1.6 trillion.
Data centre demand – which the analysts said would be limited only by physical constraints like computing resources, real estate, and energy – has gone parabolic in recent months, defying some market-watchers’ fears of a bubble.
A US$30bil bond sale by Meta Platforms Inc last month set a record for the largest order book in the history of the high-grade bond market, and investors were ready to fork over another US$18bil to Oracle Corp last week to fund a data centre campus.
Despite the eye-popping sums the analysts estimated will be necessary to fund the next half-decade of expansion, they warned the path forward will not simply be “up and to the right.”
Their biggest fear would be a repeat of the bubble that formed around telecom and fibre-optic networks in the late 1990s and early 2000s, when overcapacity led to a wave of defaults and a collapse in many sky-high valuations.
Warning signs that investor exuberance about data centres may be approaching irrational levels have been flashing brighter in recent weeks.
More than half of data industry executives are worried about future industry distress in a recent poll, and others on Wall Street have expressed concern about the complex private debt instruments hyperscalers are using to keep AI funding off their balance sheets.
“Even if everything works, there will be spectacular winners, and probably some equally spectacular losers as well given the amount of capital involved and winner takes all nature of portions of the AI ecosystem,” the analysts said.
According to the bank’s strategists, the largest portion of the funding in the next five years is expected to come from the hyperscalers themselves, which are currently diverting US$500bil of their US$700bil in annual net operating income toward capital expenditures. — Bloomberg
