How should investors gauge the probability that these off-balance-sheet contingencies will turn into on-balance-sheet problems?
EVEN before Nvidia’s splashy US$500 billion financing partnership on Aug 11, investors were starting to fret over the roughly US$70 billion in phantom liabilities that do not appear on major AI companies’ balance sheets, but could materialise at the worst possible time.
With its latest move, Nvidia is poised to provide potentially tens of billions of dollars of what is called “residual value” support for debt deals tied to the artificial intelligence build out – effectively letting firms rely on its strong credit rating in a bid to contain customer costs.
With demand for computing power soaring, this backstop has become something akin to a free lunch for massive companies like Nvidia and Broadcom, since they can bolster sales to clients and AI firms without taking debt onto their own books.
Meta Platforms, which first used this structure for its own data centres, referred to its residual value guarantees plainly in filings: “RVG payments are not probable, and therefore no liability has been recorded to date.”
“Not probable,” though, is increasingly not enough for some.
These backstops are expected to expand with the looming explosion of AI chip debt. That has led investors to look at the handful of past deals for clues about how Nvidia could structure its agreements.
Under the way Meta put together its deals, the backstop could only come into play if Meta itself chooses to walk away from its data centres early. But Broadcom applied the concept to chip financing and is backstopping Anthropic, tying its own exposure to a customer’s moves.
The arrangements typically involve a multi-step chain: a special-purpose vehicle borrows money to buy the chips, backed by the cashflow from a contract with a company that will use the technology. If the firm stops paying, then the assets are leased out again or sold to pay back the remaining debt. If there is still a shortfall, then the backstopper makes up the difference.
In theory, that sequence is a remote risk. Proponents argue that demand for chips will outstrip supply for years. The debt is structured to be paid down in full over time, meaning the possible cost of any residual value support also drops as the years pass. Arguably, the tech risk lands where it belongs: on the balance sheets of companies with enough cash to absorb a blow if something goes wrong.
SEE ALSO
Yet rating agencies are not ruling out those backstops being used, and investors are urging caution over risks accumulating off balance sheet. Wall Street was already sceptical that the blistering pace of AI infrastructure spending will yield sufficient returns. In an industry downturn, these backstops could force chipmakers to honor billions in pledges precisely when their own earnings are under pressure.
“It’s like you’re really gaming the system here; you’re trying to get preferential treatment from rating agencies so that you get the best rating possible,” said Mariya Entina, a portfolio manager at DoubleLine.
“We’re coming into this era of financial engineering. And that’s one of my concerns: When you have financial engineering, you’re obscuring the financial reality.”
Representatives for Nvidia and Broadcom did not respond to requests for comment. A representative for Meta did not have a comment.
‘Boom-bust’ #
The question consuming investors now is how to gauge the probability that these off-balance-sheet contingencies will turn into on-balance-sheet problems.
With the residual value support, Nvidia is effectively “writing a put,” analysts at CreditSights wrote in a report.
“This is pro-cyclical and exacerbates boom-bust potential,” the CreditSights analysts wrote. “The guarantee is nearly costless in the boom phase, but becomes most relevant in a severe, abrupt downturn, if/when customers are defaulting and market value for hardware is falling.”
Nvidia chief executive officer Jensen Huang wrote in a post on X that his firm may provide a residual-value support mechanism for up to 25 per cent of an opportunity, assessed on a case-by-case basis.
“Our role is to help unlock a very large pool of independent capital while maintaining disciplined risk exposure,” he wrote.
Nvidia offered few other details about the partnership, besides saying it was meant to allay “circular financing” concerns – where AI firms fund one another to buy each other’s products – by bringing in outside capital.
Meta created a template for the AI buildout when it used the residual value backstop structure as part of a roughly US$27 billion debt package to finance the construction of its Hyperion data center in rural Louisiana, in a deal called Beignet. This essentially protects lenders if Meta chooses to exit its 20-year lease early. The company followed a similar format in a roughly US$13 billion debt deal for its Sopaipilla data centre project in Texas.
Broadcom made waves by applying a similar structure to chip financing with project Big Sky, where it backstopped most of a US$35 billion debt deal where investors including Apollo Global Management and Blackstone financed the purchase of custom AI chips to lease to Anthropic. This enabled the senior debt tranches to win investment-grade ratings at lower borrowing costs.
Unlike those decades-spanning data-centre deals, chip financings operate on shorter timelines. They typically amortise in roughly five years to match the rapid depreciation of the technology. This shorter horizon burns down the backstop value quickly, giving lenders the confidence of a visible horizon rather than a multi-decade bet.
“This isn’t your garden-variety investment grade credit underwrite,” said Brian Gelfand, co-head of global credit at TCW, speaking of the arrangements broadly. “It’s far more complex than that. The tail risks are elevated given the off-balance sheet nature.”
Contingent obligations #
The Big Sky deal was the first part of a partnership called AI XPV that Broadcom struck in June – a platform that could amass US$370 billion of senior debt by mid-2029, according to Bank of America strategists.
“The primary risk lies in multiple such transactions occurring over a short period,” analysts at Moody’s Ratings wrote in a note. “We would view a substantial increase in Broadcom’s contingent obligations, even if leverage associated with its outstanding debt remains low, as limiting Broadcom’s financial flexibility and could create an overhang on the company’s credit profile.”
While the medium-term outlook for AI infrastructure remains strong, they wrote, the risk of the backstop being triggered still exists amid limited long-term visibility. S&P Global Ratings said it considers the residual value support offered by Broadcom a “contingent debt-like obligation” that it will add to its adjusted debt calculation.
Under US accounting rules, companies typically record contingent liabilities on their balance sheets when losses are likely and can be reasonably estimated. In some other cases, they may instead disclose potential losses in financial statement footnotes.
Some see a doomsday scenario as an overblown fear.
For the residual value support to be triggered, “you would have to have growth rates of token usage fall of a cliff, which we’re just not seeing,” said John Lloyd, global head of multi-sector and corporate credit at Janus Henderson Investors. “They’re not trying to hide the contingent liability. They’re trying to get it financed.” BLOOMBERG Decoding Asia newsletter: your guide to navigating Asia in a new global order. Sign up here to get Decoding Asia newsletter. Delivered to your inbox. Free.
Share with us your feedback on BT's products and services
TRENDING NOW How BYD disrupted Singapore’s car market – and why the strategy is turning on itself
When every phone becomes a satellite phone, what happens to Asia’s telcos?
Too little, too late? Manila’s billion-dollar bid to ignite its sputtering EV industry
ComfortDelGro H1 profit falls 19.7% as group focuses on long-term transformation