Key Points

• In 2026, the borrowing scale related to AI has reached $466 billion.
• Banks, bondholders, and private lending institutions jointly bear the financing risks.
• If the revenue growth of AI slows down, it may expose hidden financial vulnerabilities.

In a certain location in the United States, a new data center is under construction, which will deploy thousands of high-priced AI chips. Its future customers may include some of the world's largest technology companies, but these customers are not necessarily the ones who will pay for its construction.

On the contrary, banks, bondholders, and private investment funds tend to provide most of the funding in advance, expecting to recoup their investments over the coming years through payments from companies that purchase computing power.

This financial structure has expanded significantly. According to data from Morgan Stanley, as reported by the Financial Times, in the first nine months of 2026, companies raised approximately $466 billion in AI-related debt, compared to $101 billion during the same period last year.

This figure raises an uneasy question: if the AI craze begins to cool down before these infrastructures become profitable, then who will bear the losses?

What kind of structure lies behind the $466 billion AI debt boom?

To build a AI data center, it is necessary to invest in land, power infrastructure, cooling systems, and high-performance processors first, yet at this point, the first customer may not have received any bills yet.

Large companies such as Microsoft, Amazon, and Alphabet can fund large-scale investments with their operating cash flows. Smaller operators often rely on loans secured by equipment, customer contracts, or the facilities themselves.

Coinpaper As previously explained in the investigation into the infrastructure financing of AI, the companies that develop the AI model, the companies that own the servers, and the institutions that provide financing for these servers may not be the same parties.

The chip infrastructure of Anthropic is one example. According to UBS research, a private credit transaction worth approximately $35 billion provided financing for specialized equipment through an independent vehicle, while the customer payments were used to support this debt.

Such arrangements can disperse financing across multiple institutions, rather than concentrating it on the balance sheet of a single technology company.

This is also why the identity of the ultimate borrower is so important. Coinpaper's previous research on the OpenAI and Anthropic data centers found that in the same project, developers, cloud service providers, and financiers may be in different positions.

What will happen if the income from AI falls short of expectations?

Imagine a hypothetical data center that borrows $1 billion by using equipment and long-term customer contracts as collateral.

If demand develops as expected, customer payments will help cover operating expenses and interest, and lenders will also receive the returns stipulated in the contract.

However, if customers cut back on spending, projects are delayed, or financing costs become higher, the company may find it difficult to fulfill its obligations.

Losses usually first affect shareholders through a decline in equity value. If the borrower defaults, unsecured creditors may face losses, while secured lenders can attempt to recover their funds by taking control of the pledged equipment and other assets.

Coinpaper explained in the research on guarantees for loans to GPU that the actual outcome depends on the contract guarantee terms, the value of the collateral, and the creditor's priority in repayment.

CoreWeave explains why these differences are important. Even though the company reported having a large backlog of contracted revenues, its bond yield still exceeded 16% in October.

The expected gap between future sales and financing pressures has made the borrowing cost of CoreWeave a closely watched signal in the credit risks of AI.

Will ordinary investors end up bearing the costs in the end?

It is possible, although not every AI financing issue will evolve into a broader financial crisis.

Corporate bond funds, pension funds, and insurance institutions may directly hold debt related to AI, or indirectly hold it through investment vehicles. If a company issues additional shares, cuts expenditures, or its investments do not generate sufficient returns, shareholders will also face losses.

Recent issues surrounding the financing of Oracle's data centers indicate that even before borrowers default, loan prices may fall first.

However, the recent slowdown in the AI lending activities also indicates that lenders are becoming more selective when deciding which projects are worthy of financing.