In short
Mezzanine debt and preferred equity fill the gap between senior secured debt and equity in the capital stack of an AI data center project. They lift the combined loan to cost ratio from roughly 65 percent to 85 percent or more. PeerSense, Data Center Construction & Mezzanine Financing 2026 Mezzanine debt is subordinated financing secured by a UCC pledge of the borrower’s ownership interests in the property owning LLC. It typically carries 11 to 15 percent current interest with a current pay, accrual, or hybrid structure, runs three to seven years, and is provided by institutional debt funds, family offices, and private capital sources. George Smith Partners Preferred equity is an equity investment with priority distributions over common equity, a preferred return of 12 to 18 percent, and no intercreditor agreement with the senior lender. George Smith Partners, Mezzanine Debt & Preferred Equity Private credit broadly is expected to supply around $800 billion of the estimated $1.5 trillion financing gap for AI data centers through 2028. Morgan Stanley, Bridging a $1.5tr Data Center Financing Gap The market has moved quickly. U.S. AI data center secured debt issuance was projected to reach about $25.4 billion in 2025, more than double the prior year. The Kobeissi Letter Hyperscalers issued roughly $121 billion in AI-related debt in 2025, more than four times the five year average. Mellon analysis, Fortune Here are the forms these subordinated financings take, the deals that shaped the market, and the terms that matter now.
Why are mezzanine and subordinated debt so important for AI data center financing?
The scale of AI infrastructure building is enormous and still growing. McKinsey estimates the cost of AI infrastructure to meet global compute demand by the end of the decade at $5.2 trillion. McKinsey, The Cost of Compute (April 2025) The big four hyperscalers (Google, Microsoft, Meta, Amazon) alone are projected to spend around $700 billion on AI capital expenditures in 2026. CNBC No single source of capital can cover these sums. Senior construction lenders, mostly banks, will finance 65 to 75 percent of the cost of a build to suit AI data center. That leaves a large equity gap.
Mezzanine debt and preferred equity fill that gap. They allow a developer or sponsor to reduce its own cash equity contribution, push the total leverage higher, and still get a project built. For an AI data center campus that costs $340 million to build, a typical senior loan might cover $220 million. Mezzanine debt can add another $70 million, lifting the combined loan to cost ratio to 85 percent, as seen in a recent hyperscale campus deal. PeerSense
The market has shifted from funding these projects with corporate cash flows to funding them with debt and private credit. In 2025, the five largest AI hyperscalers issued about $121 billion in corporate debt, up from an annual average of roughly $28 billion between 2020 and 2024. Mellon Investments analysis, Reuters Morgan Stanley projects a $1.5 trillion financing gap through 2028 for global AI data center investment, out of a total projected $2.9 trillion in capex. Morgan Stanley Private credit, including mezzanine funds and asset based finance, is expected to supply about $800 billion of that gap. Morgan Stanley
Subordinated capital is no longer a niche. It is a main source of funding for the buildout.
What exactly is mezzanine debt in an AI data center deal?
Mezzanine debt is a loan that sits between the senior mortgage and the sponsor’s equity in the capital stack. It is subordinated. The senior lender gets paid first if the borrower defaults. The mezzanine lender’s security is not a mortgage on the real estate. Instead, the mezzanine lender takes a UCC (Uniform Commercial Code) pledge of the borrower’s ownership interests in the entity that owns the AI data center property. That gives the mezzanine lender the right to take over the ownership entity if the borrower fails to pay, rather than foreclosing on the real estate directly.
This structure matters because it usually avoids triggering a due on sale clause in the senior loan documents. The senior lender’s mortgage stays in place. The mezzanine lender steps into the borrower’s shoes as owner. This also means the mezzanine lender and the senior lender sign an intercreditor agreement. The agreement sets out their relative rights, including standstill periods and cure rights before a foreclosure.
For AI data center projects, mezzanine debt typically carries a current interest rate of 11 to 15 percent. It also has a payment in kind (PIK) feature. PIK means the borrower can pay part of the interest by adding it to the principal balance instead of paying cash today. Loan terms run from three to seven years, often coterminous with the senior construction loan. The loans are interest only, meaning no principal amortization during the term. Mezzanine deal sizes range from $5 million to $100 million. PeerSense
Lenders in this space include large private credit funds, infrastructure debt funds, and alternative asset managers like Blackstone, Blue Owl, and Magnetar. They are comfortable with the construction and lease up risk because the borrower typically has a pre lease from an investment grade hyperscaler tenant. Pre leasing to investment grade hyperscalers and secured power are the top two underwriting drivers. PeerSense
How does preferred equity work and how is it different from mezzanine debt?
Preferred equity is not debt. It is an equity investment in the ownership entity that sits above the common equity in the distribution waterfall. The preferred equity investor pays cash for a class of ownership interest that gets paid its preferred return, usually 12 to 18 percent, before the common equity holders receive any distributions. George Smith Partners The investor may also get a share of the upside, called a promote, after the preferred return is paid.
Preferred equity offers several practical advantages over mezzanine debt in an AI data center deal. First, it does not require an intercreditor agreement with the senior lender because it is equity, not debt. Second, it can bypass a due on sale clause altogether because the property owning entity does not change hands. Third, if the project runs into trouble, the preferred equity investor’s remedies are governed by the operating agreement and corporate law, not by UCC foreclosure rules. The process can be faster and more private.
The tradeoff is cost. Preferred equity is deeper in the capital stack and gets paid after debt but before common equity, so its required return is higher. For an AI data center, a preferred equity investment might be structured with a 15 percent preferred return plus a promote, while mezzanine debt might cost 12 percent. Deal sizes can range from $10 million to over $500 million. PeerSense
The table below compares the two structures.
| Feature | Mezzanine debt | Preferred equity |
|---|---|---|
| Legal form | Loan secured by UCC pledge of LLC interests | Preferred membership interest in the LLC |
| Current return | 11-15% | 12-18% |
| Upside participation | Usually none | Promote (share of residual profit) |
| Intercreditor agreement | Required with senior lender | Not required |
| Foreclosure remedy | UCC foreclosure and sale | LLC operating agreement remedies |
| Typical size | $5M-$100M | $10M-$500M+ |
George Smith Partners and PeerSense
How are special purpose vehicles and residual value guarantees used to finance AI data centers?
The largest AI data center financings in 2025 and 2026 used a structure that goes beyond traditional mezzanine debt. A special purpose vehicle, or SPV, is created to own the AI data center assets. The sponsor, often a hyperscaler like Meta or a large AI provider like CoreWeave, enters a long term lease with the SPV. The SPV then issues bonds or borrows money secured by the lease payments and the value of the real estate. This structure keeps the debt off the sponsor’s balance sheet. It keeps the underlying debt off the technology company’s balance sheet, because the SPV is a distinct legal entity whose liabilities are not consolidated onto the parent’s books. Tax alert
A key credit enhancement in these SPV deals is the residual value guarantee, or RVG. The sponsor tenant promises to pay the SPV a capped amount if the facility’s market value at the end of the lease is below a certain floor. This shifts the risk of technology obsolescence back onto the sponsor. The sponsor is often better able to bear that risk because it controls the technology inside the building. In the Hyperion data center financing, the SPV debt was rated A+ by Standard & Poor’s, one notch below Meta’s own corporate credit rating, and the residual value guarantee protects lenders by ensuring repayment even if the tenant exits. Columbia Business School, Financing the AI Buildout
The largest example to date is the Meta Hyperion campus in Louisiana. Meta formed a joint venture with Blue Owl Capital. The SPV, called Beignet Investor LLC, issued $27 billion in A+ rated fixed rate bonds. It was the largest investment grade corporate debt deal in U.S. history. PIMCO bought about $18 billion of the bonds. Meta entered a 20 year triple net lease with the SPV and provided an RVG for the first 16 years. Meta retained only 20 percent of the venture, while Blue Owl funds owned 80 percent. This structure effectively levered the project at roughly 90 percent debt to asset value. Columbia Business School
Similar techniques appeared in the Google TeraWulf deal. TeraWulf, a former bitcoin miner, issued $3.2 billion in high yield bonds to build an AI data center in New York. Google agreed to backstop the lease payments of the AI tenant, Fluidstack. If Fluidstack defaulted, Google would pay a termination fee or take over the lease. Google received equity warrants in TeraWulf as compensation. Legal analysis, Quinn Emanuel
These structured finance techniques blur the line between senior debt, subordinated debt, and equity. The SPV bonds are senior at the project level, but the sponsor’s equity stake is small. The private credit fund’s equity or mezzanine investment (like Blue Owl’s $7 billion cash contribution) acts as a subordinated cushion for the bondholders.
What are the landmark deals for mezzanine and subordinated financing in AI data centers?
The table below lists the largest and most instructive subordinated and private credit deals since 2024. Each used mezzanine debt, preferred equity, or a structured SPV to fund AI infrastructure at scale. We then walk through the three that best illustrate the current market.
| Deal | Amount | Key structures | Year |
|---|---|---|---|
| Meta / Blue Owl Hyperion SPV | $27 billion | SPV, RVG, preferred equity in JV, A+ bonds | 2025 |
| CoreWeave debt facility | $7.5 billion | Private credit facility (possible mezz component), plus $650M revolver | 2024 |
| Oracle Stargate multi-tranche | $18B bonds, $38B term loan, $18B project loan | Senior secured, project finance, SPV | 2025 |
| Google / TeraWulf / Fluidstack | $3.2 billion | High-yield bonds, Google lease backstop, warrants | 2025 |
| Aligned Data Centers capital raise | $12 billion+ | $5B equity, $7B debt, plus $600M credit facility | 2025 |
| Switch dual securitization | $3.5 billion | SASB CMBS and ABS | 2024-2025 |
| QTS SASB CMBS | $2.05 billion | SASB CMBS on four data centers | 2024 |
PeerSense, CoreWeave press release, Meta IR, Bloomberg, Nov 7, 2025, Legal analysis, Quinn Emanuel
CoreWeave $7.5 billion debt facility
In May 2024, CoreWeave, a GPU cloud provider for AI, closed a $7.5 billion debt financing. It was one of the largest private credit deals in history. Blackstone and Magnetar co led the facility. Coatue, Carlyle, CDPQ, DigitalBridge, BlackRock, and others also participated. The facility funded CoreWeave’s GPU fleet expansion. It likely included a mix of senior and subordinated tranches, though full details were not disclosed. Blackstone’s Global Head of Private Credit Strategies called it one of the largest private credit financings in history. CoreWeave press release CoreWeave later added a $650 million revolving credit facility led by JPMorgan, Goldman Sachs, and Morgan Stanley. CoreWeave $650M credit facility This deal showed that private credit funds were willing to underwrite AI exposure at enormous scale, using the collateral value of GPU fleets as well as contracted revenue.
Meta Hyperion SPV with Blue Owl
The October 2025 joint venture between Meta and Blue Owl Capital to finance the Hyperion campus in Louisiana is the most consequential structured deal in AI-related companies. It demonstrated that a large hyperscaler could finance a $27 billion project almost entirely with external debt, keeping the liability off its balance sheet. Blue Owl’s $7 billion equity contribution acts as a subordinated cushion, similar to mezzanine or preferred equity, while $20 billion of investment grade bonds formed the senior layer. The 20 year lease and RVG shifted long term obsolescence risk to Meta. The deal also showed that institutional investors like PIMCO and BlackRock would buy billions in AI data center bonds with a spread at least 100 basis points higher than Meta’s own corporate debt. Meta IR, Columbia Business School, Wall Street Journal
Oracle Stargate multi-tranche financing
Oracle’s financing for its role in the $500 billion Stargate initiative showed that even an investment grade company with BBB ratings needed layered debt structures to fund its AI buildout. Oracle raised $18 billion in investment grade bonds in September 2025. It then arranged a $38 billion senior secured term loan for facilities in Texas and Wisconsin, and a separate $18 billion project finance loan for a New Mexico campus. The term loan, led by JPMorgan and MUFG, priced at roughly SOFR plus 2.5 percentage points. The New Mexico loan was syndicated to about 20 banks. Oracle also used an SPV structure through which Blue Owl and JPMorgan invested roughly $13 billion. Law firm analysis The size of these loans shows how even the largest corporate borrowers rely on project level and secured structures, blending senior and private credit.
What are the current market terms for subordinated debt in AI data centers?
The table below summarizes the range of terms observed in 2025 and early 2026 for each layer of the capital stack. The data comes from PeerSense surveys of lenders and participants.
| Financing type | All-in rate / return | Term | Typical size | LTC |
|---|---|---|---|---|
| Senior construction debt | SOFR + 350-550 bps (~9-11%) | 24-48 months + extensions | $25M-$1B+ | 65-75% |
| Mezzanine debt | 11-15% current + PIK | 3-7 years | $5M-$100M | lifts stack to 85%+ |
| Preferred equity | 12-18% + promote | 3-7 years | $10M-$500M+ | sits below mezz |
| CMBS (SASB) | 6.25-9.00% fixed | 5/7/10 years | $25M-$2B | by property value |
| Data Center ABS | 5.75-7.50% fixed | 7-25 years | $250M-$5B+ tranche | by asset pool |
A note on pricing. The senior all in rate of 9 to 11 percent reflects the current SOFR floor (around 4.5 percent in mid 2026) plus spreads that have compressed slightly as more banks enter the market. Over 50 commercial banks are now active in AI data center financing. Data center financing analysis Mezzanine rates remain in the low to mid teens because mezzanine lenders take the risk of the equity cushion and construction completion. Securitized products like ABS and CMBS carry a yield premium of about 100 to 150 basis points over hyperscaler corporate bonds. Tax alert
What risks and regulatory concerns surround subordinated AI data center debt?
The speed and scale of this debt buildup have drawn attention from regulators and some market participants.
First, leverage levels are high at the project and corporate level. CoreWeave’s 2025 financials showed $14 billion in debt and $34 billion in lease obligations. Its interest coverage ratio was just 0.17. news analysis In other words, its earnings before interest covered only 17 percent of its interest expense. Any downturn in AI demand could stress highly leveraged borrowers.
Second, the interconnectedness of lenders is a concern. More than 50 banks are active, and private credit funds often participate alongside traditional banks in syndicated deals. In January 2026, four U.S. Senators sent a letter urging financial regulators to investigate the AI sector’s growing debt loads. The letter warned of destabilizing losses for interconnected institutions. Legal analysis, Quinn Emanuel The Bank for International Settlements also noted the shift from cash flow funding to debt funding. Private credit in AI-related sectors now exceeds $200 billion. BIS Bulletin
Third, there is political risk around who bears the cost if a hyperscaler walks away. In Louisiana, the Public Service Commission declined to investigate Meta’s financing with Blue Owl. Advocacy groups had argued that if Meta exercises its walk away option after four years, the remaining gas plant costs could shift to ratepayers. Union of Concerned Scientists, Feb 25, 2026 This shows that the off balance sheet structuring, while efficient, can create misalignments between private contract rights and public utility costs.
Fourth, construction cost inflation from tariffs pushed costs up 8 to 12 percent in 2025. That can tighten margins and strain cash flows. SFA Research Corner
Despite these concerns, the structured market has been resilient. AI data center ABS and CMBS deals are heavily oversubscribed. The Meta bonds reportedly attracted far more demand than supply. The market treats the credit of pre leased, investment grade tenanted AI data centers as high quality. But the rapid growth and complex layering of debt warrant ongoing attention.
Key takeaways
- Mezzanine debt and preferred equity are the primary tools to increase leverage beyond senior debt capacity. They can push loan to cost from 65 percent to over 85 percent.
- Mezzanine debt uses a UCC pledge of ownership interests. Preferred equity is an equity investment with a priority return. Preferred equity avoids intercreditor agreements and due on sale triggers.
- The private credit market for AI data centers is massive. Morgan Stanley estimates $800 billion will be needed from private credit through 2028, part of a $1.5 trillion gap.
- Special purpose vehicles with long term leases and residual value guarantees enable hyperscalers to finance projects off their balance sheets. Meta’s $27 billion Hyperion deal is the blueprint.
- Landmark deals like CoreWeave’s $7.5 billion facility show that private credit funds are willing to underwrite large AI infrastructure loans based on GPU collateral and contracted cash flows.
- Market terms for mezzanine are 11 to 15 percent current plus PIK. Preferred equity demands 12 to 18 percent plus promote. Senior construction loans price around SOFR plus 350 to 550 basis points.
- Regulatory scrutiny is rising, but no formal action has stopped the flow of capital. Participants must understand the intercreditor and structural risks.
Frequently asked questions
Q:What is the difference between mezzanine debt and preferred equity in an AI data center deal?
A:Mezzanine debt is a subordinated loan secured by the borrower’s ownership interests. Preferred equity is an equity investment that gets paid a preferred return before common equity but does not create a loan relationship. Preferred equity avoids intercreditor agreements and may avoid sale triggers. George Smith Partners
Q:Why is mezzanine financing needed for AI data centers?
A:Senior lenders typically cap their loan to cost at 65 to 75 percent. Mezzanine debt fills the gap to 85 percent or higher, reducing the amount of cash equity a developer must put in. PeerSense
Q:How big is the private credit market for AI data centers?
A:Morgan Stanley projects $800 billion of private credit will be deployed into data centers from 2025 through 2028. This includes mezzanine debt and asset based finance. Morgan Stanley
Q:What is a residual value guarantee in an SPV AI data center financing?
A:A residual value guarantee is a promise by the tenant sponsor (for example, Meta) to pay the SPV a capped amount if the data center’s market value falls below a threshold at lease end. It transfers obsolescence risk back to the sponsor and helps the SPV bonds achieve investment grade ratings. Columbia Business School
Q:What was the largest mezzanine or subordinated deal for an AI data center?
A:The Meta Hyperion SPV transaction at $27 billion is the largest private credit deal, combining Blue Owl’s equity with investment grade bonds. Fortune The CoreWeave $7.5 billion private credit facility is one of the largest private credit financings in history. CoreWeave press release
Q:What are typical mezzanine interest rates for AI data center projects?
A:Mezzanine debt carries a current interest rate of 11 to 15 percent, often with a payment in kind feature. Preferred equity requires a 12 to 18 percent preferred return plus a promote. PeerSense
Q:Are regulators concerned about AI data center debt levels?
A:Yes. Four U.S. Senators urged an investigation in January 2026. The Bank for International Settlements has noted the shift to debt funding. Capital kept flowing into AI data center debt at record levels, reaching at least $200 billion in 2025, despite warnings from the Federal Reserve, the Bank for International Settlements, and four U.S. Senators who urged investigation. Legal analysis, Quinn Emanuel and Bloomberg
Q:How do construction cost increases affect subordinated debt terms?
A:The SFA Research Corner reports tariff driven construction cost increases of 8 to 12 percent in 2025, and separately notes that projects with high loan to value ratios face funding gaps between senior debt and sponsor equity that are increasingly filled with higher-yielding mezzanine debt. SFA Research Corner
Q:Can a mezzanine lender foreclose without the senior lender’s consent?
A:No. The intercreditor agreement between the mezzanine and senior lenders typically includes standstill periods and cure rights. The mezzanine lender must give the senior lender time to cure defaults before taking over the ownership entity. The exact terms vary by deal.
Q:What is the difference between an SPV bond and a mezzanine loan?
A:An SPV bond is often senior at the project level, secured by real estate and lease payments. A mezzanine loan is subordinated and secured by ownership interests. However, in a structure like Meta Hyperion, the Blue Owl equity (and the related sponsor equity) acts like a subordinated layer to the bonds, achieving a similar effect.
Subscribe to The Compute Law Brief
The Compute Law Brief is a free weekday newsletter on the law of AI infrastructure across tax, real estate, construction, power, and deals. The big US build markets and federal law. Three minutes a morning. No paywall, and no email gate to read the blog. Subscribe if you want it in your inbox.
Junde Liu, JD, LL.M. (Taxation) candidate at UF Law. Originally published on Compute Law Blog. This article is general information and does not constitute legal advice. Reading it does not create an attorney client relationship. The reader should not act on the basis of any content here without first consulting a licensed attorney in the relevant state. Last reviewed for accuracy May 23, 2026.