Banks retreat, bond discounts, and it's hard to borrow money for the AI construction boom

The AI data center financing market is undergoing a significant credit tightening. Bond investors are demanding deeper discounts and more favorable terms, while some major banks have started to become more selective with project loans. Coupled with macro interest rate pressures and the AI industry’s inherent risks, the cost of financing for this construction boom is rising rapidly.

According to technology media outlet The Information, the latest landmark case comes from CleanSpark. This bitcoin mining company, currently developing data centers for Meta Platforms, earlier this month issued $2.3 billion in bonds at 98.5 cents on the dollar—one of the steepest discounts in the past year. The 7.875% coupon rate is already high, and the issuer further committed to amortizing principal repayments to reduce refinancing risk. According to Morgan Stanley data, all four high-yield data center bond deals since July have been sold at some form of discount; whereas, in the previous twelve months, only three out of the top ten deals included discounts.

The bank loan market is also showing signs of contraction. According to reports, sources say institutions like Société Générale, Sumitomo Mitsui Banking Corporation, and Mitsubishi UFJ Financial Group have become more cautious about project loans to data centers. Meanwhile, Oracle’s New Mexico project recently issued a force majeure notice, prompting lenders to further scrutinize contract terms and loan covenants. These developments all point to one risk: the financing chain for AI infrastructure construction is under pressure, with some projects’ expansion plans facing substantial threats.

Discounts Become the Norm, Investor Appetite Not What It Used to Be

The high-yield bond market has sold about $55 billion in AI-related bonds this year, a massive supply, while investors’ risk appetite is quietly shifting.

Connor Minnaar, fixed income portfolio manager at Manulife Investment Management, said discounts are “the new trend in the current market.” He noted, “At the beginning of the year, investors were much more accepting of deal structures,” but now this tolerance has narrowed substantially.

The peculiarity of the CleanSpark case is that its end-user, Meta Platforms, is an investment-grade company—a backing that, in the past, usually offered investors enough confidence. Yet investors still demanded additional compensation this time, reflecting that market concerns over construction cycle and execution risks have surpassed reliance on tenant credit.

Another transaction in mid-August also confirmed this trend. Developer Zenith Arc LLC, backed by venture capital from Coatue Management and infrastructure startup Fluidstack, issued discounted bonds at 99.5 cents on the dollar to finance a data center that will be leased to trading giant Jane Street. After listing, bond yields rose further, with the required compensation from investors exceeding that of similarly rated older bonds.

Bank Loan Market: Tighter Syndication, Oracle Incident Rings Alarm Bells

The tightening of the bank loan market is equally noteworthy. According to two informed sources, Société Générale, Sumitomo Mitsui, and Mitsubishi UFJ Financial Group have all become increasingly cautious when participating in data center project loans. Previously, these three banks were deeply involved in milestone AI infrastructure financings: Société Générale led the $7.1 billion debt financing for OpenAI and Oracle’s “Stargate” project’s first phase; Mitsubishi co-led $38 billion in financing for two other Oracle projects with JPMorgan; Sumitomo Mitsui was a lead arranger for $18 billion financing for Oracle’s New Mexico project.

The recent setbacks with Oracle’s New Mexico project became a warning to lenders. Last week, Oracle issued a force majeure notice to the project developer (a company under Blue Owl Capital) due to electricity supply delays. Oracle attempted to invoke contract terms to exempt or postpone its contractual obligations citing uncontrollable events.

Reports say that, from the lenders’ perspective, the loan structure was “well-designed”—even if the project failed to secure a power supply, Oracle could not use this as a basis to avoid its lease payment obligations. However, a banker noted that Oracle’s force majeure claim may prompt some banks to seek stricter protective clauses or reassess loan risk, further narrowing the potential loan supply.

Construction Boom Continues, but Financing Window Is Narrowing

Market participants emphasize that there have been no significant deals forced to withdraw, and no active syndication process has stalled due to weak pricing. Developers are still bearing higher financing costs from the high-yield bond market to keep their projects moving forward.

Connor Minnaar said, “Right now, it’s still a story about concessions. For many companies, this is a race to bring as much capacity online as quickly as possible—the level of financing costs is secondary.”

However, discounts becoming the norm, bank syndicates tightening, and an unstable IPO window—the triple pressure is reshaping the landscape of AI data center financing. For projects with lower ratings, less development experience, or longer construction cycles, financing has already become substantially more difficult. Whether the AI construction boom can sustain its previous expansion pace will increasingly depend on the financing market’s capacity to absorb risk.

Megacap Tech Company "Crowding Out" Effect Intensifies Market Pressure

The pressure in the AI financing market does not stem from a single source but rather the result of overlapping factors.

Amazon, Google, and Microsoft are expected to spend a combined $700 billion in capital expenditures this year and maintain that level in the coming years. These companies have already issued nearly $160 billion in investment-grade bonds this year, resulting in a massive supply entering the market.

This year, the additional yield investors require to hold megacap tech company bonds has increased by about 0.25 percentage points, while the broader investment-grade market spread has widened only 0.04 percentage points. Although these tech companies have large operating cash flows, the relative cost of their bonds is clearly rising.

Meanwhile, some large tech companies are shifting expenditures to other financing entities, which have to raise funds themselves in the high-yield bond market. Developers building data centers for AI companies such as Anthropic and OpenAI are also competing in the same market, further worsening the supply-demand imbalance.

Resonance of Macro Interest Rate Shocks and AI Risks

The rise in Treasury yields has introduced new uncertainties to an already stressed market. Last week, Federal Reserve Chair Jerome Powell said that large-scale tech bond issuance creates a “crowding-out effect” for investors, leading to higher Treasury yields.

For data center developers, rising Treasury yields mean riskier projects must offer higher compensation to attract capital.

Reports say a banker involved in such deals described the shift as an “upgrade migration” in credit rating and project quality—investors who formerly took on the risk of longer construction periods or less experienced developers for higher returns may now choose to earn similar yields from safer bonds.

The cost of capital for some projects is already approaching break-even levels. If the required return for a project is 12% and the borrowing cost nears that, the profit margin is squeezed to the point where the project becomes financially unsustainable.

IPO Market Volatility: Another Financing Channel Also Facing Pressure

Volatility in the Treasury market has also affected the IPO market, leaving AI companies’ alternative financing prospects uncertain. According to reports, Anthropic lost $4.2 billion last year, while its IPO prospectus shows over $50 billion in committed spending for computing and infrastructure, implying enormous financing needs.

SB Energy and Nscale, which are developing data center capacity for OpenAI and Anthropic, have recently filed for IPOs, but the timing remains uncertain. Anthropic earlier anticipated it could disclose its IPO filing as soon as this month, but has yet to take any action.