“Too big to fail” may no longer be exclusive to banks.

The historic artificial-intelligence build-out of the past few years already touches nearly everything in the U.S., from the stock market’s run to new records to rising utility bills and U.S. Treasury yields.

Critics warn of another potential “Lehman Brothers” moment lurking around the corner for AI, while “Big Short” investor Michael Burry predicted in August a 1987-style fall for the stock market.

Should funding conditions for AI “hyperscalers” take a significant turn for the worse, some Wall Street investors and analysts anticipate a government backstop to emerge.

“The AI revolution is not going away, even though it has challenges,” said James Pruskowski, managing director at Hennion & Walsh Asset Management. “The revolution is similar to building the U.S. highway system,” he said. “It’s too big to fail.”

The U.S. passed legislation in the 1950s to substantially fund and expand the interstate highway system, boosting productivity, incomes and trade, including in regions where travel had been difficult.

The AI build-out so far has been funded mostly by a handful of tech companies, which have been burning through their cash piles and issuing a mountain of new AI-related bonds since last fall.

The sector has faced rising costs of capital this summer as Treasury yields touched two-decade highs, and growing pushback from communities on new data centers ahead of the midterm elections in November.

The borrowing blitz looks poised to continue over the next few years. Companies are expected to spend another $1 trillion in capital expenditures on AI in 2027, and similar amounts each year into 2030.

To justify the capex spending, hyperscalers Oracle, Meta, Alphabet, Amazon and Microsoft will need to keep generating roughly 30% to 50% margins and about $4.31 trillion to $7.18 trillion in annual revenue, according to a new analysis by Peter Berezin, chief economist at BCA Research. Add in the neoclouds and Chinese AI companies, and the annual revenue need grows to about $10 trillion.

“To validate these projections, tech companies will need to bring in unprecedented amounts of new revenue,” Berezin wrote. “That may be a tall order.”

Requests for comment from each of the hyperscalers weren’t immediately returned.

Who owns AI-related debt?

Like with Treasury auctions, companies financing the AI build-out need investors to keep buying their new bond deals.

That’s why the roughly $12 trillion U.S. investment-grade corporate bond market has become such a vital funding source, as well as a base for buyers — mostly foreign owners, mutual funds, exchange-traded funds, life-insurance companies and pension funds, according to BofA Global.

Despite recent pressure from a series of megasize $25 billion AI-related bond deals this year, spreads in U.S. investment-grade corporate remain nearly 80 basis points above the risk-free Treasury rate and around historic lows.

Facilities created by the Federal Reserve during the pandemic, which have since gone dormant, appear to be still helping. By “capping the downside the Fed has permanently reduced the risk premium in IG corporate credit,” wrote a BofA Global team led by Yuri Seliger. “The facilities remain in the Fed’s toolbox and can be reactivated as needed,” the team wrote in an Aug. 14 client note. The Fed declined to comment.

Matt Brill, head of North America investment-grade credit at Invesco, said it’s too early to talk about any backstop for credit. Credit markets have functioned smoothly, heavy supply has been absorbed and the main AI issuers still are “elite corporations with incredible balance sheets.”

“Nothing is too big to fail,” Brill said. “There are still going to be risks out there for companies.”

Bond portfolios have gone from scarcity in tech to a sudden inundation over the past decade.

Big Tech and other major U.S. corporations are expected to issue another $200 billion in highly rated corporate bonds in the typically busy month of September —adding to what BofA Global expects to be a record $2.1 trillion year of new high-grade issuance.

This year’s heavy volume of long-dated corporate supply has been pegged as a key factor in the $31 trillion Treasury market’s selloff this summer. The 30-year Treasury yield last week hit its highest since 2007, while new 30-year AI-related bonds in early August also pricedat higher rates of nearly 6.4%. Corporate bonds price at a premium to Treasurys to help offset things like the risk of borrowers defaulting.

Last week, Treasury Secretary Scott Bessent surprised the market with plans to buy more long-dated Treasurys in September, October and early November, hoping to keep a lid on yields.

Bessent didn’t elaborate on Monday when asked whether the Treasury would do more to help calm the market for U.S. government bonds, after reports emerged that it may use the Treasury’s de facto checking account to fund increased buybacks instead of issuing more short-term Treasury bills.

The benchmark 10-year Treasury yield eased slightly to 4.7% on Monday, but it was still near the 4.743% one-year high set in late July, according to Dow Jones Market Data.

“It tells me that lower rates are important [to this administration],” Brill at Invesco said of Treasury’s Department intervention. The moves aren’t the same as a bailout, he said, but it’s along the lines of providing support. “The clear view is they want to win this AI arms race,” he added.