Financing for AI data centers is tightening across credit markets. Bond investors are asking for deeper discounts and stronger terms, and some major banks have become more selective on project loans. With higher rates and project-specific AI risks in the mix, the cost of funding the build-out is rising fast.
According to The Information, one of the clearest recent examples came from CleanSpark. The bitcoin miner, which is developing a data center for Meta Platforms, sold $2.3 billion of bonds earlier this month at 98.5 cents on the dollar. The deal ranked among the steepest discounts seen over the past year. The bonds carried a 7.875% coupon, and the issuer also agreed to amortize principal over time to reduce refinancing risk.
Data from Morgan Stanley showed that all four high-yield data center bond deals since July were sold with some form of discount. In the previous 12 months, only three of the top 10 transactions included discounts. Signs of contraction are also appearing in the loan market. The report said people familiar with the matter indicated that Societe Generale, Sumitomo Mitsui Banking Corp. and Mitsubishi UFJ Financial Group have become more cautious on data center project lending.
At the same time, a recent force majeure notice tied to Oracle's New Mexico project pushed lenders to review contract language and loan covenants more closely. Together, those developments point to the same issue: the financing chain behind AI infrastructure is under pressure, and some expansion plans are facing a real threat.
Discounted deals are becoming more common
The high-yield market has already sold roughly $55 billion of AI-related bonds this year. Supply has been heavy, but investor risk appetite is shifting.
Connor Minnaar, a fixed-income portfolio manager at Manulife Investment Management, said discounts are "what's new in the market right now." He added that "earlier in the year, investors were much more accepting of structure," but that flexibility has narrowed.
What stood out in the CleanSpark deal was the tenant profile. Its end user, Meta Platforms, is an investment-grade company, the kind of backing that would typically give investors considerable comfort. Even so, buyers still demanded extra compensation, a sign that concerns around construction timelines and execution risk are carrying more weight than tenant credit alone.
Another deal in mid-August reinforced that pattern. Zenith Arc LLC, a developer backed by a Coatue Management venture capital arm and infrastructure startup Fluidstack, sold bonds at 99.5 cents on the dollar to finance a data center that will be leased to trading firm Jane Street. After the bonds began trading, yields moved higher still, meaning investors demanded more compensation than they did on outstanding bonds with the same rating.
Loan syndicates are narrowing after the Oracle notice
The pullback in bank lending is also drawing attention. According to two people familiar with the matter, Societe Generale, Sumitomo Mitsui Banking Corp. and Mitsubishi UFJ Financial Group have each become more cautious about joining data center project loans.
All three banks had previously played major roles in headline AI infrastructure financings. Societe Generale led the $7.1 billion debt financing for the first site of the OpenAI-Oracle Stargate project. Mitsubishi and JPMorgan jointly led another $38 billion of financing across two Oracle projects. Sumitomo Mitsui took part in leading the $18 billion financing for Oracle's New Mexico project.
Last week, Oracle sent a force majeure notice to the project's developer, a Blue Owl Capital portfolio company, after delays in power supply hit the site. Oracle sought to invoke contract terms that would waive or postpone its obligations on the basis of an uncontrollable event.
The report said people familiar with the matter described the loan as "well structured" from the lenders' perspective, meaning Oracle would still be unable to walk away from lease payments even if power did not arrive on time. Still, one banker said Oracle's force majeure claim could push some lenders to seek tighter protections or reassess project risk, which would further reduce potential loan supply.
Projects are still moving, but the funding window is tighter
Market participants said no major deal has been pulled so far, and no active syndication process has been halted because of weak pricing. Developers are still absorbing higher funding costs in the high-yield market to keep projects moving.
Minnaar said, "Right now this is still a story about concessions. For many companies, it's a race to get as much capacity online as quickly as possible, and the absolute cost of financing comes second."
Even so, discounted issuance becoming routine, tighter bank syndication and an unstable IPO window are reshaping the market for AI data center financing. For lower-rated projects, developers with limited track records and projects with longer construction timelines, raising capital has become materially harder. Whether the AI building boom can maintain its earlier pace is increasingly tied to how much stress financing markets can absorb.
Heavy tech borrowing is adding strain
The pressure on AI financing is coming from several directions at once.
Amazon, Google and Microsoft are expected to spend a combined $700 billion in capital expenditures this year, with that level projected to continue for several years. Those companies have already issued nearly $160 billion of investment-grade bonds this year, sending a large amount of supply into the market.
The extra yield investors demand to hold hyperscaler bonds has risen by about 0.25 percentage points this year, while spreads in the broader investment-grade market widened by only 0.04 percentage points over the same period. These companies still generate large operating cash flows, but their relative borrowing costs have clearly increased.
Some large technology companies are also shifting spending to other financing vehicles, which then need to raise money in the high-yield bond market. At the same time, developers building data centers for AI companies such as Anthropic and OpenAI are competing for funds in that same market, deepening the supply-demand imbalance.
Higher Treasury yields are colliding with AI project risk
Rising Treasury yields have introduced another complication for a market that was already under strain. Federal Reserve Chair Warsh said last week that heavy issuance of technology bonds was creating a "crowding out effect" for investors and was one reason Treasury yields had moved higher.
For data center developers, higher Treasury yields mean riskier projects must offer more compensation to attract capital.
One banker involved in the market described the shift as an "upward migration" in credit quality and project quality. Investors that were previously willing to take construction risk on long-duration projects or back less experienced developers in exchange for higher returns may now choose safer bonds offering similar yields.
For some projects, financing costs are getting close to the break-even line. If a project requires a 12% return and borrowing costs approach that level, margins can shrink to the point where the project becomes difficult to sustain financially.
The IPO route is under pressure too
Volatility in the Treasury market is also spilling into IPOs, clouding another funding path for AI companies. According to the report, Anthropic posted a net loss of $42 billion last year, and its draft IPO filing showed more than $500 billion in committed spending on compute and infrastructure.
SB Energy and Nscale, which are developing data center capacity for OpenAI and Anthropic, have recently made their IPO filings public, but the timing of any listing remains uncertain. Anthropic had previously been expected to disclose its IPO filing as early as this month, but that has not happened so far.