The AI expansion is borrowing more money, and a growing share of that borrowing carries a below-investment-grade rating. That is making the cost of financing data centers an increasingly important test for the industry.
Data-center borrowers raised $80 billion in bonds so far in 2026, including $37 billion rated below investment grade, according to Barclays figures cited in Forbes’ October 16 report.
That puts the high-yield share at approximately 46%, compared with roughly 24% in 2025, when total issuance reached $36 billion and below-investment-grade borrowing accounted for $8.6 billion.
In dollar terms, the riskier category has more than quadrupled.
Why lenders are demanding more
A below-investment-grade rating signals greater credit risk. Borrowers generally need to offer higher interest rates to compensate investors for the possibility of repayment problems.
Forbes reported that bonds financing several data centers leased by CoreWeave had fallen in price, pushing their average yield to around 11%.
A rising market yield does not immediately increase the interest payment on every existing fixed-rate bond. It does, however, signal that investors are demanding better compensation, potentially making new borrowing and refinancing more expensive.
For projects requiring years of construction and equipment spending, that can change the economics before revenue arrives.
A major customer does not remove every risk
Forbes identified 16 data-center projects financed with high-yield bonds, including facilities leased to or backed by major technology companies.
Those relationships can support financing, but a well-known customer and a financially secure project are different things. Investors still need to examine the borrower, contract terms, construction schedule and protections if a tenant delays or withdraws.
The broader dependence on borrowing is explored in The AI Boom Is Increasingly Being Financed by Debt.
The pressure falls on future cash flow
The central challenge is matching long-term financial commitments with dependable customer revenue.
If computing demand grows strongly and facilities open on schedule, borrowing can help operators expand. If utilization disappoints, power connections arrive late or financing becomes more expensive, the same commitments can squeeze cash flow.
A high-yield rating alone does not establish that a project will fail. Equally, strong demand for AI does not guarantee that every operator will earn enough to repay its lenders.
The financing question is becoming as important as the technology question: can each project generate sufficient cash, soon enough, to cover its obligations?
Disclosure: This article does not represent investment advice. The content is for informational and educational purposes only.
