The most urgent bottleneck for AI today: cheap funding is drying up!

Title context: The most urgent bottleneck for AI today: cheap funding is drying up! Text: Over the past few years, AI companies easily raised hundreds of billions of dollars, with Wall Street competing to put up capital and rarely asking questionsthe defining feature was "plenty of money, little scrutiny." Now, lenders and investors are becoming pickier, and terms are increasingly harsh. This week, two of the AI industry's most sought-after leading players hit roadblocks in the capital markets one after another: Blackstone (BX.US) told investors that Anthropic may only receive a junk rating after its IPO; OpenAI cut its new funding target from US$30 billion to about US$20 billion. Both developments point to the same change: for the U.S. AI industry, cheap money is getting harder and harder to obtain. Dan Ivascyn, chief investment officer of Pacific Investment Management Company (Pimco), said: "The bar for proving to lenders that these investments can convert into sufficient revenue is getting higher and higher... The weak links will be hit first." The money is still there, but it is now a buyer's market Since ChatGPT launched, AI companies have tapped almost every source of fundingventure capital, the stock market, banks, bonds and private credit. According to Bank of America Corp (BAC.US), in the U.S. corporate bond market across categories, AI-related issuance totaled US$394 billion in 2025 and had reached US$655 billion by mid-October this year, with analysts expecting it to approach US$1 trillion next year. Demand for capital is rising, but funders are becoming more hesitant. Funders' concerns mainly fall into three areas: resistance to data centers is heating up in various places and may delay large projects; AI profits at hyperscale cloud providers and startups such as Anthropic and OpenAI may not materialize as quickly as promised; and with stock and bond issuance piling up, investors' AI exposure is already too heavy. New deal terms are reflected in every corner of the market. Bonds of large tech companies such as SpaceX (SPCX.US) and Meta (META.US) are trading at steep discounts; AI cloud companies such as CoreWeave (CRWV.US), Nebius (NBIS.US) and Nscale, which rely on debt to build computing power, have effectively been shut out of the bond market in recent weeks; even a strong company like Broadcom Inc. (AVGO.US) has had to offer banks sweeter terms to secure loans. Anthropic is the latest case. It plans to raise up to US$100 billion in a mid-November IPO and stated in its prospectus that it will invest more than US$800 billion in computing infrastructure by 2045. According to the Financial Times, Blackstone recently told investors that, based on guidance from major rating agencies, Anthropic is more likely to receive a rating in the BB range or above after listing, whereas Blackstone had previously expected it to secure an investment-grade rating by the end of the year. Under the systems of S&P (SPGI.US) and Fitch, BBB- and above is investment grade, while BB+ and below is speculative grade, commonly known as "junk." Many insurers and pension funds are restricted from holding large amounts of junk-rated debt. A rating in that range means Anthropic's channels for borrowing will narrow and its costs will be higher. This week's US$63 billion chip leasing deal between Broadcom Inc. and Anthropic was also negotiated amid tense sentiment. According to The Wall Street Journal (NWSA.US), Broadcom Inc. CEO Hock Tan personally intervened in the negotiations at the last minute to stabilize banks' sentiment; more than US$42 billion of senior loans in the deal were backed by Broadcom Inc.'s high single-A rating. Anthropic also listed power suppliers' financing constraints as a risk factor in its prospectus: even if it can raise money itself, it must worry that suppliers cannot. OpenAI's situation is similar. According to the Financial Times, it has cut its funding target to about US$20 billion and lowered its valuation from US$1.4 trillion to US$1.25 trillion; The Wall Street Journal, citing people familiar with the matter, said OpenAI has so far raised less than half of its original US$30 billion target. Some investors reduced their subscription amounts, and others want to wait until after Anthropic lists before deciding. Although top companies with good credit can still obtain money, the market's patience is thinning. This week, as tech companies reported disappointing earnings, AI-related stocks and bonds sold off: Alphabet (GOOGL.US) lost about US$500 billion in market value on Thursday, the largest on record; Intel Corporation (INTC.US) plunged 29% on Friday; the Nasdaq Composite fell 2.9% on Friday, one of the largest selloffs since the AI boom began in U.S. stocks. If volatility persists, financing will only become harder. Neoclouds: the first to have the door shut Pressure is falling first on the weakest credit link. One of the most active issuers in the AI debt market is a group of companies known as "neoclouds." They own or lease data centers, deploy AI servers and then sell computing power to large tech companies, trading firms and AI startups. Many neoclouds evolved from cryptocurrency mining companies, have very short track records operating AI infrastructure, and already faced elevated borrowing costs before the market turmoil, making them most sensitive to rising rates. The turning point came in September. The Federal Reserve raised rates for the first time in three years; Oracle Corporation (ORCL.US) notified the developer of a large data center in New Mexico that it would delay the start of full rent payments due to power supply and approval issues, and lenders in various places immediately reassessed the construction risk of projects in their books. CoreWeave bonds issued earlier this year carry a 9.75% coupon and now yield more than 15% in the secondary market. Falling bond prices push yields higher, indicating that its borrowing costs have risen sharply in a short period. Bankers have warned in recent weeks that the high-yield market for neoclouds is essentially closed to new issuance, and yields above 15% are unaffordable for borrowers. Nebius has sold about US$2.7 billion of stock since 2024 to investors including NVIDIA Corporation (NVDA.US) and issued several billion dollars of convertible bonds, but its funding needs remain unmet. It had considered issuing its first high-yield bond but ultimately abandoned the plan and instead began collecting prepayments from customers. Chief Executive Arkady Volozh said in Berlin last week that some funding sources such as asset-backed securities and the bond market "may be temporarily closed," but instruments such as customer prepayments are "probably enough to let us build at least this 1 gigawatt of capacity." Chief Operating Officer Ophir Nave said the asset-backed financing market remains open. Equity financing is also rough. People familiar with the matter said Lambda sought to raise US$4 billion in its final pre-IPO funding round, but after investors including Coatue decided not to participate, it lowered its target valuation from US$14.5 billion to US$13 billion. British neocloud Nscale had considered listing in November but now leans toward delaying to December. It is also negotiating with investors including NVIDIA Corporation to secure funding support for a US$45 billion data center for Anthropic in West Virginia; after the market shift, the other side demanded stricter terms, slowing its progress in updating its prospectus. Some bankers have told Nscale that its IPO may struggle to reach its previously set valuation of US$30 billion to US$35 billion. Giants: no matter how high the rating, investors can also become "full" Meta, Amazon.com, Inc. (AMZN.US), Alphabet and Microsoft Corporation (MSFT.US) sit at the other end of the credit risk spectrum. They have top ratings and plenty of room to borrow further on their balance sheets; the problem lies in the scale of financing itself. This year they have issued debt intensively in the high-grade bond market and said they will continue to do so. The investment firms, insurers and pension funds lending to them are already feeling constrained by limits and worry about excessive concentration in the same industry. "The real question is, how full can investors get in one name, one theme?" Ivascyn said. On Friday, bonds of several hyperscale cloud providers fell to near record lows, with yields noticeably higher than ordinary investment-grade corporate bonds. Despite top ratings, their borrowing costs are already higher than those of ordinary investment-grade companies. Everyone wants to get ahead of everyone else. Meta plans to raise up to US$97.5 billion through a series of new share offerings, one reason being to move before competitors, and in regulatory filings it cited other giants' financing as a challenge to the capital market's capacity to absorb it. "Everyone wants to raise money for their buildout before liquidity dries up," said Ben Connard, chief investment officer of Carnegie Investment Counsel. "Ultimately, there may not be enough money to support this buildout. Financing may become the ultimate bottleneck that slows everything down." Why the tightness of funding has become so critical The AI buildout is increasingly dependent on external funding. According to a Reuters (TRI.US) analysis of LSEG consensus forecast data, the capital expenditure of five companiesMicrosoft Corporation, Alphabet, Amazon.com, Inc., Meta and Oracle Corporationis expected to exceed their free cash flow by 2027, with about US$1.57 of new capital expenditure for every additional US$1 of operating cash flow. They cannot spend enough from the money they earn themselves, so the gap can only be filled by issuing stock and debt. The giants also cannot stop. A Cornell Capital Group research report described this as a "prisoner's dilemma": if you do not invest while rivals do, you may be eliminated. Investor Citrini noted in an interview with the Monetary Matters podcast that this change alters the nature of the risk. "In the first three years of this AI trade, sentiment-driven selloffs always had a floor, because most spending came from hyperscalers' operating cash flow." Now, incremental spending comes almost entirely from equity and debt financing. "If market sentiment turns, that incremental funding will dry up, because it is no longer just one company spending moneyit is determined by the market." He called this mechanism "reflexivity": confidence supports financing, and financing supports investment; once confidence weakens, funding contraction in turn hits investment and growth expectations. He also noted that AI revenue is currently growing about 25% to 30% a year, while capital expenditure growth is close to 60% to 70%, and the gap between the two widens with each round of guidance. This article is reposted from "Wallstreetcn," author: Gao Zhimou, GMTEight editor: Zheng Yuyang.