Signs of a slowdown are emerging in the boom of artificial intelligence lending, with the global issuance of AI related debt plummeting by nearly 80% from $113 billion in June to just $23 billion in September.
At the time of this decline, banks are striving to distribute loans totaling approximately $18 billion related to a large data center project of Oracle located in New Mexico. This also adds another warning sign for the AI industry, which plans to raise a record amount of funds in 2026.
Investors are now beginning to question how quickly the billions of dollars invested in AI infrastructure can be converted into revenue, especially in light of rising borrowing costs and the potential for construction delays that could threaten the economic viability of the projects.
AI Debt issuance nearly halved from its June high
According to data cited by The Financial Times from Morgan Stanley, global AI related debt issuance fell to $23 billion in September, less than half of the amount in August, and far below the peak of $113 billion in June.
The slowdown is particularly evident in the U.S. investment-grade bond market. In September, AI related issuances came to a complete halt, while large technology companies borrowed a total of about $306 billion between January and August.
However, this decline does not mean that financing has disappeared. In the first nine months of 2026, the total amount of loans related to AI was still approximately 466 billion US dollars, reflecting the company's highly aggressive efforts to raise funds for expansion at the beginning of the year.
Morgan Stanley believes that the weakening in September was largely due to the previous surge in borrowing, but higher interest rates and a deeper market scrutiny of the economic viability of data centers have also made new financing more complex.
$18 billion data center loan of Oracle traded at a discount
Oracle provides a concrete example of the pressure on the credit market faced by AI.
The loan of approximately $18 billion supporting the “Jupiter project” was quoted at only 89 to 91 cents of its face value in September. This project is related to Oracle’s data center campus in New Mexico and its commitments regarding the infrastructure for OpenAI.
It is reported that banks, including Santander Bank and Jefferies, have encountered difficulties in distributing these loans. The reasons are that investors have re-evaluated construction delays, the continuously expanding debt burden of Oracle, as well as the uncertainties regarding the project's power infrastructure.
Financing difficulties arose after Standard & Poor's downgraded Oracle to BBB in July, which is its lowest investment-grade rating. Previously, the financing for the Jupiter project had shown that even with a large amount of contracted demand, the debt at the project level could still be under pressure.
A loan transaction price lower than the face value does not mean that Oracle has defaulted; rather, it indicates that investors are requesting a discount to assume the associated risk exposure.
This situation also reflects a broader issue faced by AI infrastructure stocks: high revenue expectations must ultimately cover financing costs and capital expenditures.
