Credit Conditions Tighten Across AI Infrastructure Financing
TREE NEWS reports: The financing engine behind the AI data center boom is showing clear signs of strain. Bond investors are demanding deeper discounts and stronger covenants, major banks are growing more selective about project loans, and rising Treasury yields are compounding the pressure. The result: the cost of funding AI infrastructure is climbing fast, and the market’s appetite for risk is narrowing.
The latest emblematic case is CleanSpark, a bitcoin miner developing data centers for Meta Platforms. Earlier this month it issued $2.3 billion in bonds at 98.5 cents on the dollar — one of the largest discounts of the past year. The 7.875% coupon was already elevated, and the issuer added amortizing principal payments to reduce refinancing risk. Since July, all four high-yield data center bond deals have priced at some form of discount, compared with only three of the prior ten deals over the previous twelve months.
Bank Lending Pullback and Oracle’s Force Majeure Warning
The bank loan market is contracting in parallel. Société Générale, Sumitomo Mitsui Banking Corp., and Mitsubishi UFJ have become more cautious about data center project loans. These institutions were previously deeply involved in landmark AI infrastructure financings: Société Générale led the $7.1 billion debt package for the first Stargate site, while Mitsubishi and JPMorgan co-led $38 billion for two other Oracle projects, and SMBC helped lead $18 billion for Oracle’s New Mexico project.
Oracle’s New Mexico project has become a cautionary tale. Last week, Oracle issued a force majeure notice to the project developer, a Blue Owl Capital entity, citing delays in power supply. Oracle sought to invoke contract clauses to excuse or defer its obligations due to events beyond its control. While the loan is considered well-structured and Oracle cannot escape lease payments even without power, one banker said the force majeure claim could push some banks to seek tighter protections or reassess risk, further constraining loan supply.
Market Implications: Higher Yields, Wider Spreads, and a Shrinking Window
The implications ripple across asset classes. In credit markets, high-yield AI-related bonds are repricing wider as investors demand compensation for construction and execution risk. The CleanSpark deal showed that even a Meta investment-grade backstop no longer guarantees easy funding. In August, Zenith Arc LLC, backed by Coatue Management and Fluidstack, priced bonds at 99.5 cents to finance a data center leased to Jane Street; yields rose further after trading, with investors demanding more compensation than on similarly rated older bonds.
Equities face a dual squeeze. Hyperscalers — Amazon, Google, and Microsoft — are expected to spend roughly $700 billion in capex this year and have issued nearly $160 billion in investment-grade bonds. The extra yield investors demand on these bonds has risen about 0.25 percentage points this year, versus just 0.04 points for the broader investment-grade market. Meanwhile, developers building for Anthropic and OpenAI are competing for the same high-yield capital, crowding out smaller players.
The macro backdrop is amplifying the stress. Fed Chair Warsh recently noted that massive tech bond issuance is crowding out investors and contributing to higher Treasury yields. For data center developers, higher risk-free rates mean riskier projects must offer even greater compensation. One banker described the shift as an “upward migration” in credit quality — investors who once took construction and developer risk for yield are now opting for safer bonds with similar returns.
Some projects are approaching breakeven. If a project needs a 12% return and borrowing costs approach that level, margins compress to the point of financial non-viability. The IPO window is also unstable, clouding another funding avenue. Anthropic reported a $42 billion net loss last year and its draft IPO filing shows over $500 billion in compute and infrastructure commitments. SB Energy and Nscale have filed for IPOs but timing remains uncertain, and Anthropic has yet to disclose its filing despite earlier expectations.
Key Takeaways for Investors
- Credit risk is repricing: High-yield AI data center bonds now require discounts and stronger covenants. Expect wider spreads and higher coupons for lower-rated developers.
- Bank lending is selective: Major banks are tightening project loan standards. Projects with long build cycles or inexperienced developers face higher funding hurdles.
- Hyperscaler crowding out: Massive investment-grade issuance from Amazon, Google, and Microsoft is raising relative borrowing costs and squeezing smaller issuers.
- Macro rates matter: Rising Treasury yields raise the bar for project returns. Watch Fed communications and Treasury supply for further pressure.
- Equity and IPO risk: AI developers reliant on IPO markets face uncertain windows. Companies with heavy compute commitments and no near-term profitability are most vulnerable.
The AI buildout is not stopping, but the financing window is narrowing. Projects that can secure capital will proceed; those that cannot may see expansion plans delayed or shelved. For investors, the key is distinguishing between well-structured deals with strong tenants and speculative projects that rely on continued market tolerance — a tolerance that is clearly fading.
