AI demand is surging, but the financing "credit gate" is tightening! Under the heavy pressure of 5% Treasury yields, the losers in computing power are gradually emerging.

Title context: AI demand is surging, but the financing "credit gate" is tightening! Under the heavy pressure of 5% Treasury yields, the losers in computing power are gradually emerging. Text: Companies in the stock market that benefit from an unprecedented wave of AI computing infrastructure buildout are now facing both expanding demand and sharply tightened financing conditions: AI agent applications are driving continuously expanding demand for AI computing resources, while persistently rising funding costs against the backdrop of the 10-year U.S. Treasury yield surging above 5% have raised the capital threshold for converting excess AI orders into actual cash flow. In the week ended September 25, CoreWeave shares rose nearly 8%, while Oracle fell about 7% and has dropped about 30% year to date, showing that investors are beginning to examine different companies' financing, delivery, and cash flow capabilities in greater detail. JPMorgan estimated in June that, through 2030, debt financing needed for AI infrastructure buildout will reach about $4.1 trillion - this is a multi-year cumulative financing forecast, not a single-year issuance volume, and cannot all be equated with corporate bonds. When such enormous construction demand continues to enter the capital markets, interest rates, credit conditions, and funding availability will directly affect the pace of expansion. The hottest AI agent applications, represented by Muse and Astra, provide a new source of sustainable and strong growth for computing demand. A single user instruction can trigger planning, retrieval, browser operations, code execution, and result validation, while multiple rounds of model calls add demand for context processing and state preservation. Reasoning from engineering mechanisms, GPUs handle model computation, CPUs execute tools and task orchestration, HBM, server DRAM, and enterprise SSDs jointly support data access and cache management, and data center optical interconnects govern high-speed data transmission. A detailed engineering analysis by Nvidia pointed out that CPU execution speed affects effective GPU utilization, while tiered caching can reduce redundant computation through GPU memory, CPU memory, local NVMe, and remote storage. Therefore, the investment value brought by agent penetration ultimately needs to be reflected in how much billable, highly complex task delivery and larger AI computing resource demand each unit of capital can deliver, as well as in stronger corporate operating cash flow. The specific changes on the funding side are undoubtedly global. On September 25, the U.S. 10-year and 30-year Treasury yields intraday touched about 5.23% and 5.53%, respectively, the highest levels since 2007 and 2004. Energy supply shocks, sticky inflation, and resilient economic demand pushed the market to raise its expected future policy rate path; the Federal Reserve already raised rates by 25 basis points in September, and a Charles Schwab analysis dated September 25 showed that since then the futures market has tended to price in expectations of nearly three additional rate hikes through June 2027. As the "anchor of global asset pricing," a higher 10-year U.S. Treasury yield both raises the pricing benchmark for new long-term financing and, through a higher required rate of return, lowers the valuation of companies' future cash flows. The "credit gate" of the AI frenzy: total financing expansion and financing eligibility contraction may occur simultaneously The current AI computing expansion frenzy appears to be undergoing screening by a "credit gate" - demand determines how much companies want to finance, while lending capacity determines which orders and projects are sufficient to obtain funding. Financing professionals at Mitsubishi UFJ Lease's U.S. company believe that the market may truly be willing to support only a portion of the candidate list of new cloud companies; this is a practitioner judgment, yet it reveals a very critical change: higher interest rates can compensate for some risk, but projects lacking reliable power, delivery guarantees, or executable customer contracts, or AI projects overly dependent on a single large customer, may be unable to obtain financing merely by raising pricing. From this it follows that industry-wide borrowing may still grow, while funds gradually concentrate in companies with thicker financial buffers, higher contract quality, and stronger delivery capabilities. The interest rate shock also needs to distinguish among debt structures. Newly issued or refinanced fixed-rate debt will face repricing; existing floating-rate loans generally adjust with short-term benchmarks such as SOFR, and the increase in 10-year U.S. Treasury yields cannot be directly applied to all loans. CoreWeave's SEC filings show that, based on the outstanding floating-rate debt balance as of June 30, 2026, a 100 basis point increase in interest rates corresponds to about $30 million of additional interest expense on a three-month basis and about $61 million on a six-month basis. The company also disclosed that it uses interest rate swaps to mitigate part of the risk. These figures measure the sensitivity of specific existing debt; future new borrowings, financing costs, and credit spread changes will further affect actual funding costs. SoftBank reflects the side where financing demand remains strong. Its bond issuance finalized on September 24 included $10 billion of dollar bonds and EUR 1 billion of euro bonds, totaling about $11.1 billion; the longest dollar bond had a 7.5-year maturity, a 9.75% coupon, was issued at par, and was expected to settle on September 29. Part of the proceeds will be used for subsequent investment payments to OpenAI. This signal shows that when a large company believes the cost of missing the AI positioning window is higher, it may still accept expensive funding; but higher interest also means future investment returns need to clear a higher return threshold. Another variable that is harder to offset by strong demand is the time gap between construction spending and operating cash collection. Equipment procurement, engineering construction, and financing costs occur first, while revenue depends on power connection, project acceptance, and customer usage. The "force majeure" dispute over Oracle's Project Jupiter concerns contractual payment arrangements after a potential delay; media previously reported, citing people familiar with the matter, that the relevant arrangement could extend the payment phase of lower construction-period rent and delay the start of higher operating-period rent, while Blue Owl Capital said its financial commitments were unchanged. It can be seen from this that the financing value of long-term AI infrastructure orders also depends on delivery conditions and the risk-sharing system, and cannot be directly regarded as cash already received. For investors, the "credit gate" ultimately screens for AI cloud computing/computing power companies that can convert computing demand into debt-servicing cash flow on time and retain shareholder returns after paying interest. AI computing demand heat collides with financing pressure: bond yields surge, and AI computing companies hungry for debt financing face greater risk As U.S. Treasury yields rose this week to the highest level since 2007, companies reliant on debt financing will face higher borrowing costs. This means that AI infrastructure buildout, already at a historic scale, is about to become even more expensive. JPMorgan estimated in a June in-depth research report that about $4.1 trillion of AI-related debt issuance is expected through 2030. Data center companies and other companies tied to the AI boom are racing to expand capacity to meet AI service demand that many industry experts consider nearly impossible to satisfy. When borrowers re-enter the financing market, they face a 10-year U.S. Treasury yield of nearly 5.17%, up about 1 percentage point from the start of the year. This means companies issuing debt will have to offer more attractive returns to lure investors. The market has not yet fallen into panic, at least not yet. Shares of debt-heavy new cloud company CoreWeave have remained solid, rising nearly 8% this week; while Oracle, which is highly financially leveraged and relies on the long-term debt market to support AI expansion, has had a harder time, falling 7% this week and about 30% year to date. Meanwhile, Japanese investment giant SoftBank Group, one of the main capital providers for AI projects, raised $11.1 billion this week through a junk bond issuance, with the 7-year bond yielding as high as 9.75%. Mark Malek, chief investment officer at Siebert Financial, said in an interview: "They are basically price-insensitive to this financing, which means they are price takers. In my view, many of these companies have to be less price-sensitive. They need to raise as much capital as possible to compete." At the center of the AI boom are leading model developers OpenAI and Anthropic, both of which have private-market valuations above the $1 trillion super-threshold. To provide the infrastructure needed for these advanced models, as well as the models and services of many other companies, the tech industry's hyperscale cloud providers - Amazon, Google, Meta, and Microsoft - have committed to spending hundreds of billions of dollars in capital expenditure this year, with expectations for further increases to $1 trillion in AI capital expenditure by 2027. Although a considerable portion of these investments is funded through debt financing, these tech giants all have investment-grade credit ratings, enabling them to obtain capital at lower cost. However, some market participants believe other companies will face greater challenges in the future. According to messages cited by media, a veteran private credit investor told media that in the future, financing for new cloud projects will become more difficult because these companies have less buffer to absorb rising costs. The investor requested anonymity in order to discuss the issue candidly. Riley Thompson, vice president at Mitsubishi HC Capital America, said in an interview that even if borrowers are willing to pay higher interest rates, lenders are becoming more selective about which projects they are willing to fund. Thompson said: "If you list 50 new cloud companies, the market may only really be interested in 20." CoreWeave, which went public last year, warned of the risk of rising interest rates in filings with the U.S. Securities and Exchange Commission. In its latest quarterly filing, the company said that based on its outstanding floating-rate debt balance as of June, for every 100 basis point, or 1 percentage point, increase in interest rates, its interest expense could rise by $30 million. An early warning signal may already have appeared this week. On Friday, media reported, citing people familiar with the matter, that Oracle issued a "force majeure" notice for its New Mexico data center project to avoid bearing higher costs, after which the company's shares fell. It is said that if the campus, called Project Jupiter, fails to begin operations as expected in 2028, Oracle wants to delay related payments. Oracle said the project "remains on our established schedule." Rising interest rates are not the only problem currently faced. Before this week's surge in yields, the CEOs of Anthropic and OpenAI had already begun calling for a slower pace of AI development, after industry researchers publicly expressed concerns that advanced models could face risks of escaping human control. At the same time, with the November midterm elections approaching, opposition to AI data centers across the United States has become an important issue. A recent NBC News Decision Desk poll, with technical support from SurveyMonkey, showed that 69% of respondents opposed building such facilities in their areas, mainly due to the unprecedented pace of household utility bill and carbon emission surges brought by the AI infrastructure frenzy. On Monday, Texas Republican Governor Greg Abbott, facing a fiercely contested re-election race, ordered a temporary halt to all data center-related environmental permit approvals; he had already suspended grid connection approvals last month. Nevertheless, demand for AI services is still growing rapidly. The latest example is Meta's personal assistant app Muse, whose popularity has climbed quickly since its launch in early September. In the first two weeks after launch, Muse exceeded 2.5 million global downloads and surpassed ChatGPT at the top of the Apple App Store. Evercore's Mark Mahaney told CNBC this week that Muse could reach 100 million users within 6 to 12 months. Andrew Giudici, global head of corporate, project, and infrastructure finance at credit rating agency KBRA, said that although rising interest rates may affect future deals, he has not yet seen a major impact on borrowing demand. Giudici said: "In a normal environment, people might take a step back and pause a bit. But I don't think that will happen here. I think you will continue to see fairly large-scale issuance." Haim Zaltzman, vice chair of Latham & Watkins' emerging companies and growth practice, said there is no doubt that as costs rise, "someone has to bear it." Zaltzman, who has long worked on core business at the financing level of AI computing infrastructure, said: "But in a structure with such strong demand, it is much easier to absorb these costs." For Bernie Margolis, CEO of American Compute, which provides risk management consulting for GPU financing, the math is even simpler. He said that even with higher costs, borrowers are still eager to obtain financing, especially those with contracts with OpenAI and Anthropic; both companies have been signing agreements to lock in computing supply for years to come. Margolis said: "If you have a large multi-year contract with Anthropic, is an extra 50 basis points really going to stop you?"