AI hyperscalers are transforming debt

As Silicon Valley’s tech giants fan out across the world in search of cash, often changing the economic facts of life in the process, Europe’s debt markets are set to welcome a new borrower.

Meta will tap the European bond market for the first time this autumn to raise funds for its AI expansion, according to investors familiar with its goals.

It is due to be the latest stop in the sector’s record-breaking hunt for funds from new and old sources alike — a headlong rush with big implications for other borrowers during the current AI investment supercycle.

“The overarching narrative is — anywhere and everywhere,” says Greg Peters, co-chief investment officer of asset manager PGIM’s credit business. “The quantum of debt that is hitting the marketplace is historic. The numbers are just absolutely enormous. It’s transformational.”

The fundraising binge is changing the way the world borrows money, affecting where, when and at what cost companies and governments issue debt. Some have said it could be the biggest shift since the creation of the Eurobond market in the 1960s.

Many companies are now being forced to tiptoe around hyperscalers, timing their bond issues to avoid clashing with those of the tech giants and, investors say, often borrowing for shorter periods to avoid the market’s glut of long-term debt.

Even the very biggest financial players say they are affected.

Federal Reserve chair Kevin Warsh and US Treasury secretary Scott Bessent argue that the hyperscalers are competing for capital with the $31tn US Treasury market, where last week 10-year borrowing costs hit their highest since 2007.

Staff at the European Central Bank have expressed concern that “the surge in big tech borrowing could make it harder for other companies and other economic sectors to access finance.” In a blog post last month, they asked: “Can euro area financial markets smoothly handle such large and concentrated debt inflows?”

The scale of the borrowing has shattered precedents. Last week Japan’s SoftBank raised more than $11bn of debt in the biggest junk bond issue in history to finance an investment in ChatGPT creator OpenAI.

Overall, according to Goldman Sachs, investors have provided about $500bn of financing to AI-linked groups so far this year.

The hyperscalers — Amazon, Alphabet, Meta, Microsoft and Oracle — account for about $200bn of the total, and are expected to issue more than $1tn of new debt over the next few years to finance their vast investment in the infrastructure needed to power AI.

Meta did not respond to a request for comment about its plans to raise funds in Europe.

Such outsize demand has created credit markets that can support larger and longer-term debt issues than ever before. But it has distorted pricing, with highly rated bonds sometimes trading at discounts to lesser-quality issues because of the sudden glut in supply.

It also brings new risks. Chipmakers, data centre operators and their power providers are all borrowing from public and private credit markets as well as financing each other, making it hard for the pension funds and insurers who buy their debt to assess their true level of exposure to the sector.

The interconnections between borrowers, in a new sector whose long-term trajectory is still unclear, have sparked concerns that if one falters, others could be dragged down with it.

“Inevitably, this is a one-way bet,” says Scott Schulte, global co-head of investment-grade debt syndicate at Barclays. “As long as everything goes right, it’s good. But the minute there’s a problem anywhere . . . the ecosystem is so intertwined.”

‘How much can we issue?’

In 2025, when hyperscaler debt issuance was just beginning, the tech giants set their bankers a challenge.

The question, according to Schulte, was: “How much debt do you think that we can issue over the course of the year?” They did not just mean in their home market, the world’s largest, he adds. They wanted to know how much they could possibly borrow “across currencies, globally”.

Their bankers spent that year “putting pencil to paper, and working out what [they] think the actual capacity is in any given market”, he says.

Then they acted, beginning with the market they know best, the $12tn US corporate bond market.

Fraser Lundie, global head of fixed income at Aviva Investors, says the ensuing “supply shock” quickly overwhelmed investors. With so much debt on offer, they were able to demand lower prices, meaning higher yields, or borrowing costs, for hyperscalers.

Lotfi Karoui, multi-asset credit strategist at Pimco, adds that the “repricing” in hyperscaler bonds — in some cases, close to levels of junk or lower-quality bonds — is occurring “because the pipeline is too unpredictable right now”.

During the second-quarter earnings season, Google, Amazon, Microsoft and Meta increased their projections for this year’s capital spending to a combined $745bn. “We haven’t found a landing zone for capex and supply . . . the pace and magnitude of revisions to capex are quite significant,” Karoui says.

So far, this repricing has mostly been confined to the tech names themselves. “At the moment you’ve almost got two credit markets, the AI-related issuers and everyone else,” says Andy Chorlton, chief investment officer for fixed income at M&G. But he is concerned that, over time, borrowing costs for all high-grade corporate issuers will be dragged higher. “That is something that we are certainly focused on.”

There have been few signs that higher financing costs are deterring the tech giants from issuing more debt, however. “Elon Musk is endeavouring to colonise Mars,” says PGIM’s Peters, referring to SpaceX’s ambitions to build data centres in space. “It doesn’t matter to him if a bunch of bond investors charge an extra hundred basis points.”

To avoid overwhelming the market, some hyperscalers have given investors assurances, for instance, that they will not issue debt in two consecutive quarters.

But companies in other sectors are still growing concerned that they will be crowded out by hyperscalers’ massive debt offerings. Coming to market at the same time as Oracle or Amazon could mean higher costs as bond buyers hold back cash to deploy in the AI-related issue.

David Brown, co-head of global investment-grade fixed income at asset manager Neuberger Berman, says that when a large AI issuance is coming to market, “we will definitely prepare . . . We will hold off or make sure we have capacity to add that issuance.”

Hyperscalers on tour

The magnitude of the borrowing, and its impact on pricing even in the highly developed US bond market, has led the tech giants and their bankers to travel the world.

Records have tumbled as a result. In June, Amazon sold C$14bn ($10bn) of debt, smashing the Canadian currency’s previous record of C$8.5bn set by Alphabet just a few weeks earlier.

Canadian borrowers now have to make sure that they “are not going head to head with some of the bigger deals that are out there”, says Abeed Ramji, head of Canadian debt capital markets at TD Securities. “The windows of access matter so much more than they used to in the past.”

Ramji adds, however, that Canada’s ability to absorb large deals has helped promote the country as a strong credit market and encouraged other companies to increase the sizes of their own debt issues.

“I don’t think this takes away from any demand,” he says. “The pie is just getting bigger.”

Last month Alphabet issued a record A$5.5bn ($3.9bn) bond in Australia, more than twice as large as the previous biggest. The group also holds the record for the largest Swiss franc bond issue, having raised SFr3.1bn ($3.7bn) there in February. Amazon’s SFr2.8bn offering in May is the second-largest.

The deluge of debt in hitherto less sought-out markets is pushing up borrowing costs for other companies and prompting them to shift the timing or duration of their deals, investors said.

Barclays analysis of the European credit market found that non-hyperscaler bonds were underperforming in parts of the market where hyperscaler issuance has been heaviest, such as longer-dated, higher-rated bonds.

The same appears true in Canada. Since Amazon and Alphabet’s record deals, the country’s highest-quality AA corporate credit index — where both hyperscalers sit — has traded at a discount to the equivalent index of A-rated bonds.

Even though hyperscaler deals can represent a meaningful share of current issuance in individual foreign-currency markets, their overall outstanding debt is a relatively small component of bond indices.

“When we look at the percentage of outstanding debt across the different markets, the euro capacity for hyperscaler debt is still untapped,” says Rehan Latif, global head of credit trading at Morgan Stanley.

The new deals have revitalised some of the smaller markets, say bankers. Adam Bothamley, global head of debt capital markets at HSBC, says there had been a slower pipeline of companies seeking to borrow in pounds this year, until the hyperscalers stepped in. “Having these sizeable transactions has been really helpful and has increased interest in and focus on the sterling market,” he adds.

Alphabet sold £5.5bn worth of sterling bonds in February, part of a $31.5bn package that included a rare 100-year £1bn bond that some say rekindled interest in long-maturity sterling debt. Amazon also raised £4.25bn earlier this month, its first-ever sterling issue.

The unknowns of gobbling up new debt at such a long maturity are huge, some investors warn. “If you just take it at its face value, investors are buying up to 100 years of unsecured debt in a transformational technology,” says Peters of PGIM. “Not to overly simplify it, but 100 years is a long time. I don’t think it’s good risk reward . . . it’s unknowable.”

Christian Hantel, head of global corporate bonds at Swiss bank Vontobel, highlights the practicalities of piling into smaller debt markets. “Those local markets will be filled quicker than the dollar market,” he says. “In some smaller markets like Swiss, sterling, Canadian, the market is getting close to its limit soon.”

The Treasury effect

The AI debt spree comes at a fragile moment for bond investors globally, as borrowing costs for governments in mature economies such as the US, UK, Germany and France rise to levels last seen in the late-2000s.

Whether investors have cashed out of government debt to buy into bonds issued by tech giants, as some credit-watchers have asserted, is a hot debate on Wall Street.

“We can see clearly . . . [hyperscaler debt] competes with government bonds,” says Marion Le Morhedec, chief investment officer for fixed income at Fidelity International.

“As a portfolio manager . . . that question about whether to invest in a 30-year bond from a hyperscaler versus a government is relevant, where it wasn’t before,” she adds. The hyperscalers offer an attractive “pure credit story”, as investors grow less tolerant of governments’ inability to rein in deficits and spending.

Others are less sure, arguing that the sell-off in government bonds has principally been due to macroeconomic factors such as inflation and fiscal deficits.

“The numbers we are talking about on the sovereign side are so much higher,” said HSBC’s Bothamley. “It’s a stretch to say that . . . hyperscaler issuance is causing any sustained increase in [sovereign] yields.”

It is difficult to prove that investors are ditching government bonds for corporate debt. But markets have nevertheless been jolted by the arrival of such large and relatively price-insensitive sellers of debt, at a time when some traditional bond buyers — notably central banks and defined-contribution pension funds — have reined in their purchases.

“As heavy government borrowing collides with growing AI financing needs . . . persistent demand for capital is one reason we think borrowing costs can stay elevated,” BlackRock analysts wrote last week.

Matthew Hornbach, global head of macro strategy at Morgan Stanley, adds that part of the sell-off of US Treasuries that has pushed up yields could be down to banks or brokers hedging their hyperscaler credit exposure. When such companies provide liquidity for investors to buy and sell previously issued bonds, they are known as primary dealers.

“We can find a small amount of evidence, through US primary dealer balance sheet holdings of corporate bonds”, that hyperscaler issuance is having an effect on US Treasury yields, Hornbach says, though he adds it is a far smaller factor than the upward trend in US interest rates and the deteriorating US fiscal outlook.

Even so, some investors think the authorities in charge of governments’ debt-issuing plans are having to think harder about when they conduct debt auctions, to avoid competing with a popular hyperscaler deal.

“We are starting to see countries thinking about AI issuance, going shorter or avoiding timing when they think there will be AI issuance,” says Henrietta Pacquement, chief operating officer for fixed income at Allspring Global Investments.

What if something breaks?

As the AI-related share of credit markets grows around the world, so do the risks associated with a downturn in the sector that now influences all other capital markets to some degree.

Although the majority of AI’s debt needs are being met in the safer investment-grade market, the sector’s more speculative names such as cloud-computing company CoreWeave have been venturing into the market for junk debt.

Such riskier issuers are deeply intertwined with the rest of the AI ecosystem. “If something did break in the below-investment-grade market, either a credit or a specific data centre, that would not play well in investment grade,” says Brown, at Neuberger.

With tech giants tapping all corners of the financial market for capital, assessing the overall exposure to individual borrowers has also become increasingly difficult.

Meta, for instance, has issued various bonds through its corporate balance sheet and via special-purpose vehicles tied to individual data centre projects.

Loren Moran, fixed-income portfolio manager at Wellington Management, says that “people have crept higher on single-issuer limits than they’ve really accounted for” and that close monitoring is needed across asset classes to avoid being unintentionally overweight on a single name.

With so much money committed to the bet that AI will transform the global economy to the benefit of the hyperscalers and their investors, a sector slowdown could come with huge financial consequences.

“The credit story in hyperscalers rests on a single consensus assumption, that operating cash flow triples from $600bn to $2tn,” Apollo’s chief economist, Torsten Sløk, wrote this month.

“If this doesn’t happen, then the risk is that the AI trade weakens, with credit spreads widening, capex plans getting cut and ultimately US GDP growth slowing,” Sløk warned.

Leading AI labs Anthropic and OpenAI have already talked of slowing development after rogue AI agents hacked into websites. Popular opposition to power-hungry data centres is growing, especially in the US. It is far from clear what the long-term financial returns from AI will be, or which companies will reap them. Analogies with the bust that followed the railway boom of the 19th century, which was also capital-intensive and largely debt-financed, are commonplace.

In the wake of mounting worries by equity investors that the world’s stock markets are dangerously tied to the fortunes of AI, the same risk is creeping into the bond market — risking an even bigger fallout if cracks emerge.

“New bonds . . . are increasingly a claim on the same AI investment cycle that has been driving equity return,” says Lucas Baynes, a senior investment strategist at Vanguard. In other words, the places investors can hide from the financial consequences of AI are getting fewer and further between.