Investors favor safer data center projects as AI debt piles up

Photo: panumas nikhomkhai / Pexels

Investors favor safer data center projects as AI debt piles up

Lenders are demanding higher yields and stricter terms, leaving speculative AI builds facing a tougher credit market

The AI data center boom has run on borrowed money. Now the lenders are getting picky about who they lend to.

Investors are pulling support from riskier data center projects and moving toward safer bets backed by major cloud providers. The result is a tighter credit environment. Lenders want higher yields and stricter terms. Developers without blue-chip tenants are finding the welcome mat has been quietly rolled up.

The numbers behind the nerves

AI-related debt issuance reached nearly $500 billion by August 2026. That figure accounts for about 20% of higher-rated US corporate bond issuance. In 2024, the share was 1%. That is a twentyfold jump in AI’s slice of the high-quality bond market in roughly two years.

Morgan Stanley anticipates approximately $3 trillion in AI infrastructure spending through 2028, with half of it financed through debt.

Advertisement

JPMorgan goes further, estimating $4.1 trillion in AI-related debt by 2030.

Yields tell the story

A Meta-backed data center project in El Paso priced with a yield of 7.53% in July 2026. An earlier deal involving Meta and Blue Owl carried a yield of 6.58%. Same type of project, same blue-chip name attached, and investors still wanted almost a full percentage point more.

Further down the risk ladder, the numbers climb fast. BB-minus bonds linked to CoreWeave carried yields around 10%.

When interest rates bite

CoreWeave has spelled out what floating-rate exposure means for its business. As of June 2026, the company indicated that a 100-basis-point rise in rates could add approximately $30 million in quarterly interest expenses on its floating-rate debt.

Oracle’s New Mexico headache

In September 2026, Oracle filed a force majeure notice for its New Mexico data center, known as Project Jupiter. The filing cited power and permitting issues.

Oracle’s stock came under pressure, and the company ran into complications securing syndicated loans.

What this means for the AI build-out

The key things to watch are yields on new data center deals, the width of credit spreads, and whether more projects hit power or permitting walls. If Morgan Stanley’s and JPMorgan’s projections hold, the debt market will be asked to fund a great deal more of this build-out, and the terms it sets will shape which projects actually get built.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.
Investors favor safer data center projects as AI debt piles up
Investors favor safer data center projects as AI debt piles up

Lenders are demanding higher yields and stricter terms, leaving speculative AI builds facing a tougher credit market

Photo: panumas nikhomkhai / Pexels

The AI data center boom has run on borrowed money. Now the lenders are getting picky about who they lend to.

Investors are pulling support from riskier data center projects and moving toward safer bets backed by major cloud providers. The result is a tighter credit environment. Lenders want higher yields and stricter terms. Developers without blue-chip tenants are finding the welcome mat has been quietly rolled up.

The numbers behind the nerves

AI-related debt issuance reached nearly $500 billion by August 2026. That figure accounts for about 20% of higher-rated US corporate bond issuance. In 2024, the share was 1%. That is a twentyfold jump in AI’s slice of the high-quality bond market in roughly two years.

Morgan Stanley anticipates approximately $3 trillion in AI infrastructure spending through 2028, with half of it financed through debt.

Advertisement

JPMorgan goes further, estimating $4.1 trillion in AI-related debt by 2030.

Yields tell the story

A Meta-backed data center project in El Paso priced with a yield of 7.53% in July 2026. An earlier deal involving Meta and Blue Owl carried a yield of 6.58%. Same type of project, same blue-chip name attached, and investors still wanted almost a full percentage point more.

Further down the risk ladder, the numbers climb fast. BB-minus bonds linked to CoreWeave carried yields around 10%.

When interest rates bite

CoreWeave has spelled out what floating-rate exposure means for its business. As of June 2026, the company indicated that a 100-basis-point rise in rates could add approximately $30 million in quarterly interest expenses on its floating-rate debt.

Oracle’s New Mexico headache

In September 2026, Oracle filed a force majeure notice for its New Mexico data center, known as Project Jupiter. The filing cited power and permitting issues.

Oracle’s stock came under pressure, and the company ran into complications securing syndicated loans.

What this means for the AI build-out

The key things to watch are yields on new data center deals, the width of credit spreads, and whether more projects hit power or permitting walls. If Morgan Stanley’s and JPMorgan’s projections hold, the debt market will be asked to fund a great deal more of this build-out, and the terms it sets will shape which projects actually get built.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.