J.P. Morgan rejects the “AI credit bubble” argument: record buying can hold up the AI computing power arms race

The Zhitong Finance App learned that Kelsey Bello, portfolio manager of the asset management business division of Wall Street financial giant J.P. Morgan Chase, said that the global investment-grade bond market will enter a busy issuance period in September, but the actual demand for AI-related technology corporate bonds indicates that the market's concerns and anxiety about the huge supply wave may be serious and excessive. Concerns about the “bursting of the AI credit bubble” are rapidly heating up, and J.P. Morgan's asset management department is trying to drastically cool down market anxiety. The agency's core judgment is that AI capital expenditure is indeed driving a surge in corporate financing, but the high interest rate environment is also attracting retail capital and OTC institutional capital, causing “record supply” and “record demand” to expand at the same time.

In an interview with the media on Monday, Bello said that the market's expectations for the distribution scale span a large range, ranging from 175 billion US dollars to 250 billion US dollars. She added that credit portfolio managers generally believe that even if the supply scale reaches 250 billion US dollars, “it's not a real problem,” but the uncertainty of the final numbers is indeed making it more difficult for global financial markets to prepare for supply and demand in advance.

The spread of the forecast range itself is the source of market tension. There is a $75 billion gap between the highest and lowest expectations, making it difficult for portfolio managers to deploy positions before September, which has historically been one of the busiest issuance months of the year. As AI infrastructure spending boosted corporate borrowing, the number of blue-chip corporate bonds issued in the US set a monthly record for three consecutive months in August, continuing the fastest issuance pace in the history of the market.

The $410 billion funding torrent is likely to hit record buying

Therefore, according to the asset management department under J.P. Morgan Chase (Xiaomo), even if the issuance of investment-grade bonds reached 250 billion US dollars in September, the market may still digest smoothly, and even trigger wait-and-see capital to catch up. However, Xiaomo's opinion refutes the risk of short-term “supply crunch” and does not prove that the return on AI projects is sufficient to cover long-term financing costs. Xiaomo's core view is that high yields continue to attract retail and OTC capital, and record demand is sufficient to handle record supply. Therefore, the AI financing wave has yet to evolve into a liquidity crisis in the investment-grade bond market, but project profitability and forward credit risk have yet to be verified.

As the borrowing frenzy begins to have an impact, investors are becoming more picky about which bonds to buy and at what price. Some companies seeking to finance data center projects are increasingly turning to junk bond investors to help them raise billions of dollars — even if the debt itself is investment grade.

Two weeks until the end of August, the monthly bond issuance volume had reached 157 billion US dollars, including the 25 billion US dollar issuance transaction of Google's parent company Alphabet, as well as other major transactions between AbbVie and Chaowei Semiconductor (AMD).

Despite market concerns about supply pressure, demand has been able to keep up with record distribution volumes since this year. According to capital flow data from high-rated bonds quoted by Bello, investors' retail demand for investment-grade bonds in 2024 has exceeded the total volume for the whole year of any year since 2010.

“While supply hit a record, so did demand.” She said and pointed out that this dynamic is a key reason why the intensive release schedule in September may be easier to digest than market concerns.

Bello warned that if September's bond supply can be digested without market disturbances, investors who are still on the market may act quickly. “There will be a round of capital inflows into the market,” she said. Any sign of market stability could attract buyers back in droves.

When the Token Reasoning Factory swallows up the bond market: Record buying holds up the AI arms race, the next stress test is return on capital

According to the latest statistics compiled by the agency, companies have borrowed more than 410 billion US dollars for data centers and other artificial intelligence investments since this year.

As of August 24, 2026, “the AI credit bubble is about to burst systemically” is not a benchmark scenario. The more accurate judgment is that the financing frenzy has passed through the stage of undifferentiated easing and has entered the phase of weighted credit spread pricing and issuer stratification.

Goldman Sachs estimates that AI-related debt has been issued close to 500 billion US dollars, of which hyperscale cloud vendors have issued about $1940 billion, accounting for only 40% of the total — the two sets of data have different calibers, but they all point to the same conclusion: AI has become the largest single capital theme in the global credit market. Kelsey Bello of J.P. Morgan Asset Management believes that even if the supply of investment-grade bonds reached 250 billion US dollars in September, the record demand and a comprehensive yield of about 5% to 6% would still be sufficient; this only proves that the market has temporary liquidity; it does not prove that all AI projects can generate cash flow covering the cost of debt in the future.

What we really need to be wary of is “poor digestion in the bond market” rather than the immediate repayment crisis of large technology companies. Goldman Sachs's AI credit basket spread has nearly tripled from about 74 basis points in the past year; AI-related companies accounted for about 40% of investment-grade bond issuance over 15 years this year, and the number of insurance institutions' participation in large 30-year orders decreased by about half compared to the first quarter.

According to institutional statistics, as of August 10, AI hyperscale cloud vendors issued bonds of about 220 billion US dollars during the year, compared to only 12.5 billion US dollars in the same period last year; interest spreads on technology bonds reached 89 basis points, about 9 basis points wider than the overall investment-grade market, indicating that investors are making concessions, shortening the term, and limiting the exposure of a single issuer to recover “blank checks.” Therefore, the risk is primarily focused on high-leverage new cloud vendors, lack of customer-locked data center projects, and special purpose carriers that rely on forward token revenue to repay debts, rather than Alphabet, Microsoft, or Amazon, which have huge cash flows and low leverage.

According to Xiaomo and other institutions that are more optimistic about the future of the credit bond market, the AI computing power infrastructure frenzy can continue, but it will shift from “expansion as long as financing is possible” to “structural construction bound by contracts, cash flow, and return on capital.” Hyperscale cloud computing vendors still have strong operating cash flows and strong AI revenue prospects led by cloud computing, and multi-tier capital sources such as investment-grade bonds, private credit, infrastructure funds, and project financing. Therefore, orders for the entire AI computing power industry chain, such as AI GPU/TPU, HBM/DRAM/NAND storage components, optical interconnection, power equipment chains, data center CPUs and high-performance network infrastructure, and liquid cooling equipment are unlikely to suddenly shrink at least this year and next two years.

However, rising long-term interest rates and widening credit spreads will force marginal projects to be delayed, curtailed, or require advance payments from customers. Choosing an infrastructure platform with investment-grade customers, long-term contracts, prepayment mechanisms, implemented electricity, and verifiable utilization rates will be critical. Important warning signs that actually indicate the bursting of the “AI credit bubble” will be the continued contraction of the order book, the fall of GPU rental prices below the cost of capital, the failure of project financing, and simultaneous cuts in capital expenses by large manufacturers. These conditions have not yet arisen at the same time.