High-grade AI debt attracts junk bond investors as data centre funding costs rise

Investment-grade debt issued to finance the artificial intelligence infrastructure boom is increasingly drawing buyers traditionally associated with the high-yield market, as unusually attractive returns broaden the investor base for large technology and data centre financings, according to a report by Bloomberg.

The trend was highlighted last week when Blackstone-backed QTS Realty Trust raised $3.9bn through bonds to finance a Georgia data centre linked to Microsoft. Although the securities received investment-grade ratings, they offered a yield of about 7.23% — above levels typically associated with much of the BB-rated market.

A similar dynamic emerged in July when BlackRock raised $12.5bn for a Texas data centre project, with some of the debt carrying a yield of around 7.53%.

People familiar with both transactions said the bonds were marketed to investors across both investment-grade and high-yield markets.

For private capital managers, the development illustrates how the enormous financing requirements of AI infrastructure are blurring the traditional boundaries between credit strategies. Companies have borrowed more than $410bn this year to fund data centres and other AI-related investments, according to Bloomberg data.

High-yield investors have previously moved into higher-quality corporate debt when pricing becomes attractive. Similar behaviour was seen during the Covid-19 pandemic, when credit managers bought investment-grade securities offering unusually high returns.

The scale of the current AI investment cycle, however, is creating a different backdrop. Major technology companies have historically relied heavily on equity and operating cash flow to fund capital expenditure. The enormous upfront costs associated with AI infrastructure are now prompting them to turn increasingly to debt markets, despite projects potentially taking years to generate returns.

Some technology debt is already trading in the secondary market at yields more commonly associated with speculative-grade securities. Bonds issued by companies including Oracle and SpaceX are examples of the broader repricing of technology credit.

Higher financing costs could also alter the economics of AI investment, with rising borrowing costs increasing companies’ weighted average cost of capital, which could eventually put pressure on projects where returns remain uncertain or distant.

The financing requirement is unlikely to disappear soon. Vanguard estimates that the major technology companies known as hyperscalers could spend almost $800bn on AI this year, followed by more than $1tn annually from 2027 through 2030. Debt markets are expected to provide a substantial portion of that funding.

Andrew Keches, co-head of US high-grade research at Barclays, said investors are demanding greater compensation because of uncertainty around the amount of additional borrowing that could follow an individual transaction.

He added that high-yield and distressed investors are increasingly interested in investment-grade AI-related debt where yields have risen sufficiently to meet their return targets.

There are limits, however, to how much additional demand high-yield investors can provide. The US junk bond market is roughly $1.5tn, less than one-fifth the size of the investment-grade market, and generally has lower liquidity.

That creates a potential trade-off for investors seeking to rotate into higher-quality AI debt. They may first need to sell existing high-yield holdings, potentially proving difficult when liquidity deteriorates.

The surge in AI-related capital expenditure is contributing to record levels of corporate bond issuance.

TD Securities has raised its forecast for 2026 US investment-grade issuance by about $100bn to $2.2tn. European issuance has also accelerated sharply following the summer slowdown.

Alphabet recently returned to the bond market with its highest-yielding debt on record, raising AUD5.5bn ($3.9bn) in Australia. Meanwhile, ByteDance attracted more than $30bn of orders for a jumbo syndicated loan.

Private credit is also becoming an increasingly important source of AI financing. Eagle Point Credit Management is providing about $1.3bn to finance an AI data centre in Texas linked to Anthropic, while Anthropic’s revolving credit facility is expected to exceed its original target of roughly $10 billion as the company prepares for a potential public listing.

Broadcom is separately in discussions with lenders over more than $60bn of debt for an AI chip financing transaction expected to benefit Anthropic and other companies.

Elsewhere in credit markets, Moody’s has upgraded hundreds of US and European collateralised loan obligation tranches following changes to its assessment methodology, while placing almost 1,000 additional tranches under review. The changes affect a significant proportion of the CLO market and have prompted some concerns that rating standards could be becoming less conservative.

Jefferies Credit Partners is also seeking around €1bn ($1.16bn) for a continuation vehicle focused on its private credit portfolio. The vehicle is expected to acquire existing loans while providing additional capital for new lending, with the structure split approximately evenly between equity and leverage.