Juicy yields pull junk bond 'tourists' towards high-grade AI debt

By Davide Barbuscia

 

Companies looking to finance data center projects are increasingly turning to junk bond investors to help them raise billions of dollars, even for debt that is investment-grade. 

 

QTS Realty Trust Inc. this week sold $3.9 billion of bonds to fund a facility in Georgia tied to Microsoft Corp. The bonds carried high-grade ratings, but yielded about 7.23 per cent, a level higher than even middle-tier junk bonds would often pay. BlackRock Inc. is paying a 7.53 per cent yield on blue-chip securities sold in July for a Texas data center project.

 

In both cases, underwriters sold the debt to high-grade as well as high-yield investors, according to people with knowledge of the transactions.

 

The fact that investment-grade rated projects are tapping junk debt investors to get at least some bond deals done underscores the feverish competition for capital as firms pour hundreds of billions of dollars into artificial intelligence infrastructure. Companies have already borrowed more than $410 billion for data centers and other AI investments this year, according to data compiled by Bloomberg News.

 

“We’re seeing high-yield investors become tourists in investment-grade technology debt,” said Steven Schweitzer, a high-yield portfolio manager at Advent Capital Management. “When you can buy a fortress balance sheet at a yield and a spread that looks like a double B, it’s hard not to visit.”

 

Junk bond investors opportunistically buying investment-grade credit is not totally new. During the Covid-19 pandemic, for example, many money managers snatched up relatively highly rated debt that looked like a bargain.

 

However, big tech companies are borrowing at unprecedented levels — a sharp contrast to how they used to operate. For decades, they could fund their necessary investments first with equity and later, with cash flow from their profitable businesses. Now, with the vast upfront costs that AI requires, they’re selling mountains of debt for projects where a payoff could be years away, which means financing them is riskier.

 

And in secondary markets, high-grade notes from companies like Oracle Corp. and SpaceX are trading at junk-like yields, a sign that they’d have to offer speculative-grade compensation if they tapped debt investors again.

 

“The tech guys are borrowing the money, but they’re also borrowing money at high costs. That is going to change the weighted average cost of capital. And that’s where you end up with a challenge,” said Mark Malek, chief investment officer at Siebert Financial. 

 

Still, a lot more debt is coming. Vanguard Group Inc. wrote in a note this week that tech companies known as hyperscalers could spend almost $800 billion this year on AI, and more than $1 trillion every year from 2027 to 2030. Most of that expenditure will be funded in debt markets.  

 

“There’s a big question every time you’re underwriting a new deal of how much more is behind that, so I think investors are simply just asking for more to feel comfortable underwriting these bonds today,” said Andrew Keches, co-head of US high grade research at Barclays Plc.

 

“We’re seeing high-yield and even distressed investors with interest in those bonds because they’ve reached a yield level where it could be a return opportunity that meets their thresholds,” he said.

 

While higher compensation potentially broadens the buyer base for AI debt, there are limits to how much support junk investors can ultimately provide. The US junk bond market is about $1.5 trillion in size, or less than a fifth of its high-grade counterpart, and is generally less liquid. 

 

“Even if junk bond holders would like to diversify by buying into the new AI hype, they would need to find buyers for the junk bonds they hold, which normally is a challenge, not to mention in the summer time when liquidity dies out,” said Slawomir Soroczynski, global chief investment officer for fixed income at Crown Agents Investment Management.

 

Eventually, higher borrowing costs for tech companies could curb debt issuance by lower-rated businesses in the sector, according to strategists at Morgan Stanley. While companies with large balance sheets can afford to spend heavily on AI infrastructure upfront, knowing it can take years to generate returns, it’s more difficult for firms with smaller financial cushions.  

 

“For these borrowers, wider spreads represent a more meaningful constraint, making funding costs a natural stabilizer of future supply,” Morgan Stanley’s Vishwanath Tirupattur and Vishwas Patkar wrote in a recent note.