The AI data center financing market is experiencing a credit crunch. CleanSpark, a Bitcoin mining company developing a data center for Meta, issued $2.3 billion in bonds at a discount of 98.5 cents on the dollar, with a coupon rate of 7.875% and a commitment to repay principal in installments—the largest discount on any such transaction in the past year. According to Morgan Stanley data, all four high-yield data center bonds issued since July have sold at a discount, compared to only three such cases in the previous 12 months. The bank loan market has also contracted, with Société Générale, Sumitomo Mitsui Banking Corporation, and Mitsubishi Bank adopting a more cautious approach to project lending. Oracle’s notice of force majeure on its New Mexico project has further heightened lenders’ concerns. Super-large tech companies are projected to spend $700 billion on capital expenditures this year, and their massive bond supply is crowding out financing opportunities for other AI projects. Rising macro interest rates and AI-related risks are reinforcing each other, pushing financing costs for some projects close to the 12% breakeven threshold.Article author and source: Wall Street Journal
The financing market for AI data centers is undergoing a significant credit crunch. Bond investors are demanding higher discounts and more favorable terms, while several major banks have begun to be more selective about project loans. Combined with macroeconomic interest rate pressures and inherent risks in the AI industry, financing costs for this construction boom are rising rapidly.
According to technology media The Information, the latest high-profile case comes from CleanSpark. This bitcoin mining company, which is developing data centers for Meta Platforms, issued $2.3 billion in bonds earlier this month at a price of 98.5 cents—among the deepest discounts of any offering over the past year. Although a coupon rate of 7.875% is already high, the issuer additionally committed to scheduled principal repayments to reduce refinancing risk. According to Morgan Stanley data, all four high-yield data center bond offerings since July have been sold at some form of discount; in contrast, only three of the top ten offerings in the previous 12 months included a discount.
The bank lending market is also showing signs of contraction. According to reports, informed sources indicate that institutions such as Société Générale, Sumitomo Mitsui Banking Corporation, and Mitsubishi Bank have become more cautious in extending loans for data center projects. Meanwhile, Oracle’s recent notice of force majeure regarding its New Mexico project has prompted lenders to conduct even stricter reviews of contract terms and loan covenants. These developments collectively point to a single risk: the financing chain for AI infrastructure is under pressure, with expansion plans for some projects facing substantial threats.
Discounts have become the norm; investor appetite has changed significantly. This year, approximately $55 billion in AI-related bonds have been sold, reflecting a large supply, while investors’ risk preferences are quietly shifting.
Connor Minnaar, Fixed Income Portfolio Manager at Manulife Investment Management, said that discounts are "a new trend in today’s market." He noted that "investors were much more accepting of trading structures at the beginning of the year," but this tolerance has now clearly narrowed.
The uniqueness of the CleanSpark case lies in the fact that its end user, Meta Platforms, is an investment-grade company—such endorsements typically have been sufficient to instill strong confidence in investors. However, investors still demanded additional compensation, reflecting that market concerns over construction cycle risks and project execution risks now outweigh reliance on tenant creditworthiness.
In mid-August, another transaction confirmed this trend: Zenith Arc LLC, a developer backed by Coatue Management’s venture fund and infrastructure startup Fluidstack, issued bonds at a discount of 99.5 cents to finance a data center that will be leased to trading giant Jane Street. After the bonds began trading, yields rose further, with investors demanding compensation that exceeded that of older bonds with the same rating.
The bank loan market: Syndicates tighten as the Oracle incident serves as a wake-up call. The tightening in the bank loan market is equally significant. According to two informed sources, Société Générale, Sumitomo Mitsui Banking Corporation, and Mitsubishi Bank have all become more cautious about participating in data center project financing. These three banks had previously been deeply involved in landmark financings for AI infrastructure: Société Générale led the $7.1 billion debt financing for the first site of OpenAI’s “Stargate” project with Oracle; Mitsubishi and JPMorgan Chase jointly led $38 billion in financing for two other Oracle projects; and Sumitomo Mitsui participated in leading the $18 billion financing for Oracle’s New Mexico project.
Recent setbacks in Oracle’s New Mexico project have served as a warning to lenders. Last week, Oracle issued a force majeure notice to the project’s developer (a company under Blue Owl Capital), citing delays in power supply. Oracle sought to invoke contractual provisions to excuse or delay its obligations due to uncontrollable events.
According to reports, insiders revealed that, from the lender’s perspective, the loan is “well-structured,” meaning Oracle cannot use the failure to secure power supply as grounds to exempt itself from lease payment obligations. However, one banker noted that Oracle’s force majeure claim may prompt some banks to seek stricter protective clauses or reassess loan risks, further constraining potential loan supply.
The construction boom continues, but the financing window is narrowing. Market participants emphasize that no major deals have been forcibly withdrawn, and no active syndication processes have been halted due to weak pricing. Project developers are still facing higher financing costs in the high-yield bond market to keep their projects moving forward.
Connor Minnaar said: "This is still a story about concession terms. For many companies, it’s a race to get as much capacity online as quickly as possible, with financing costs being secondary."
However, the normalization of discounts, the narrowing of bank syndicates, and an unstable IPO window—three pressures叠加—are reshaping the financing landscape for AI data centers. For projects with lower credit ratings, limited development experience, or longer construction cycles, the difficulty of securing financing has risen substantially. Whether the AI construction boom can maintain its previous pace of expansion will increasingly depend on the financing market’s capacity to absorb these pressures.
The "crowding out" effect from mega-tech companies is intensifying market pressure. The pressure in the AI funding market is not due to a single source, but rather the result of multiple factors overlapping.
Major tech companies such as Amazon, Google, and Microsoft are expected to collectively spend approximately $700 billion on capital expenditures this year, with plans to maintain this level over the coming years. These companies have already issued nearly $160 billion in investment-grade bonds this year, flooding the market with large-scale supply.
The additional yield demanded by investors for holding bonds of mega-cap tech companies has risen by approximately 0.25 percentage points this year, while the spread in the broader investment-grade market has widened by only 0.04 percentage points over the same period. Despite these tech companies’ substantial operating cash flows, the relative cost of their bonds has risen noticeably.
Meanwhile, some large tech companies have shifted their spending to other financing entities that must raise capital independently in the high-yield bond market; developers building data centers for AI companies such as Anthropic and OpenAI are also competing for funding in the same market, further exacerbating the supply-demand imbalance.
Macro interest rate shocks and AI risks are reinforcing each other; the rise in Treasury yields adds further uncertainty to an already pressured market. Last week, Federal Reserve Chair Walsh stated that large-scale tech bond issuances are creating a "crowding-out effect" on investors, contributing to the rise in Treasury yields.
For data center developers, rising government bond yields mean that risky projects must offer higher returns to attract funding.
According to reports, a banker involved in the transaction described this shift as an "upward migration" in credit ratings and project quality—investors who previously accepted risk in projects with long construction cycles and inexperienced developers in exchange for returns may now opt for similar returns from safer bonds.
The financing costs for some projects are approaching the break-even point. If the required return on a project is 12% and borrowing costs are nearing this level, profit margins will be compressed to the point where the project becomes financially unsustainable.
The IPO market is volatile, and another funding channel is also under pressure—fluctuations in the government bond market are affecting the IPO market, casting uncertainty over the financing prospects for AI companies. According to reports, Anthropic posted a net loss of $42 billion last year, and its IPO draft filings reveal commitments to computing and infrastructure spending exceeding $500 billion, indicating an enormous funding requirement.
SB Energy and Nscale, which are developing data center capacity for OpenAI and Anthropic, have recently filed public IPO documents, but the timing of the listing remains uncertain. Anthropic had previously anticipated filing its IPO application as early as this month, but has not yet done so.