Bondholders beware: The hidden credit cost of funding the AI revolution

AI’s new industrial revolution is being financed the old-fashioned way, with significant flows of capital spending and a rapidly swelling pile of debt. In a new note, Yarra Capital Management’s head of Australian credit research, Phil Strano, argues that “AI is rewriting the rules of capital markets,” but the frenzy of borrowing by the big US technology platforms raises uncomfortable echoes of a previous debt-fuelled boom that ended badly for bondholders.

Strano focuses on the so-called hyperscalers – the likes of Microsoft, Amazon, Alphabet, Meta and Oracle – which are in the midst of a sizeable build-out of AI infrastructure. Their combined annual capital expenditure is heading towards an almost five-fold increase over the seven years to 2028, as they pour money into data centres, chips and networking gear to stay ahead in AI. These are the companies driving US equity indices and, by extension, a large chunk of global growth, but they are also changing the shape of global credit markets.

The first phase of the AI boom was largely equity-funded; now, Strano notes, the giants are leaning hard into bond and loan markets. A second chart tracks their borrowings surging in 2025 to “many multiples” of previous years as mega-deals roll through in US dollars. He stresses that this is not a one-off: with capital spending plans locked in for years, investors should assume much more issuance is coming, not less.

On the surface, there seems little to worry about. The hyperscalers are renowned for strong balance sheets, dominant market positions and prodigious cashflows. For now, their credit quality is “pristine,” with S&P ratings between AA- and AAA for the main names. But Strano’s warning is that “even the biggest names aren’t immune” to the basic arithmetic of leverage: if capex keeps accelerating, at some point the limits of balance sheets will be tested and ratings pressure will follow.

Credit markets are already starting to signal some discomfort. Strano points to credit default swap (CDS) spreads (the cost of insuring against default) on US tech names, which have been nudging wider in recent months. In his view, this is an early sign of “indigestion” as the market absorbs record supply; if issuance continues at the current clip, spreads are likely to come under further strain.

For Strano, who has seen several credit cycles, the current moment carries an eerie sense of déjà vu. He draws a parallel with the early 2000s, when highly rated European telecommunications companies went on their own spending binge, paying more than US$100 billion for 3G spectrum licences and associated infrastructure. Then, as now, the story was a seductive one: new technology, a promised “Internet of Things” age, and apparently unassailable incumbents racing to secure strategic assets.

The problem, he writes, is that the “roads to 3G riches were more potholed than expected.” Revenues and returns fell well short of the lofty projections that had underpinned the telcos’ bids for spectrum. At the same time, the associated debt issuance materially weakened their balance sheets. For credit investors, the result was a brutal bout of negative credit migration and a sharp repricing of risk as spreads blew out.

Strano highlights Deutsche Telekom as a cautionary tale. At the turn of the millennium, DT carried a solid AA- credit rating; by 2004, after the 3G splurge and weaker-than-hoped-for returns, it had slid to BBB+. Two decades on, he notes, the rating has never fully recovered: DT still sits in the BBB+ band.

The history, he says, is clear: bond investors who funded Europe’s 3G build-out suffered large mark-to-market losses as spreads gapped wider and ratings were cut. There were no outright defaults, but that was cold comfort for institutions forced to crystallise losses or switch out after downgrades breached portfolio mandates. For Strano, this episode is the lens through which to view the AI build-out: pristine today does not mean pristine forever if debt keeps rising faster than cashflows.

He applies that lesson directly to today’s hyperscalers. Meta, Amazon, Alphabet and Microsoft may all sit comfortably in the AA to AAA range now, but Yarra’s base case is that their ratings “in the years ahead are likely to migrate down to single-A and maybe even triple-B categories” if current capex and funding trends persist. In other words, investors should not buy these bonds on the assumption that today’s top-tier ratings are permanent.

That makes current pricing look complacent. Strano singles out a recent Meta deal as an example: the company issued US$13 billion across 10- and 30-year maturities, at spreads of 78 and 98 basis points over Treasuries respectively. You can argue about the relative value of the 10-year tranche, he says, but it is hard to justify the 30-year, where a sub-1 per cent spread offers scant compensation for three decades of downgrade and re-pricing risk.

The mechanics here matter. Long-dated bonds are highly sensitive to changes in both interest rates and credit spreads. If, over the next decade, Meta is downgraded from AA- into the single-A or BBB bucket, the 30-year bonds will likely need to cheapen significantly to clear the market. Even if the company never misses a coupon, those mark-to-market moves could inflict “steep losses” on investors who thought they were buying ultra-safe paper to tuck away.

For Yarra, this is more than an academic point, it goes to the heart of how the firm runs money. Strano says the team’s ethos is to “maximise risk-adjusted returns” by treating credit ratings with a forward-looking bias rather than accepting them at face value on the day of issue. That approach is applied consistently across its Australian Bond, Enhanced Income and Higher Income strategies, where the focus is on how an issuer’s profile is likely to evolve over the life of a security.

In practice, that means scrutinising business models, capex plans and funding strategies, particularly in capital-intensive sectors like AI infrastructure, and asking whether today’s spreads genuinely compensate for the probability of migration down the ratings ladder. It also means preferring parts of the curve where investors are paid adequately for taking risk, and being prepared to pass on fashionable names when the numbers do not stack up.

For investors, the message is not to shun AI altogether; few can afford to ignore companies that dominate equity indices and technological progress, but to be clear-eyed about how they are choosing to gain exposure. Equity holders are explicitly backing growth; bondholders should remember they are underwriting leverage. In Strano’s view, that distinction has been blurred in the rush to own anything with an AI label.

The broader lesson from Europe’s 3G era is that credit markets can take years to fully digest the consequences of an investment boom, but when they do, the re-pricing can be swift and painful. As AI’s debt-fuelled build-out gathers pace, Strano’s central question hangs over the market: will today’s AI leaders manage their balance sheets conservatively enough to avoid the fate of the European telcos, or will history, yet again, rhyme?

James Dunn

James is an experienced senior journalist and editor of The Inside Network's publications.