Summary:
- Development: Rule 144A offerings are becoming a key financing solution for the AI-driven data center boom as traditional banks face lending limits, construction-risk concerns, and balance-sheet constraints. These private placements allow developers to access deep pools of institutional capital and help bridge the growing funding gap for large-scale data center projects.
- Why it matters: AI and cloud infrastructure projects require trillions of dollars in investment, exceeding what banks alone can provide. The emergence of 144A financing, private credit, and securitization markets is reshaping how data centers are funded and enabling developers to continue building despite financing, permitting, and construction challenges.
- Action point: Developers, lenders, investors, and counsel should monitor the evolving role of 144A offerings, construction-risk allocation, tenant concentration, power availability, and regulatory developments. Understanding how banks, private credit funds, and capital markets work together within the modern data center capital stack is critical when evaluating financing opportunities and risks.
The Scale of the Problem
The artificial intelligence buildout has created a financing challenge unlike anything the commercial real estate and project finance markets have faced before. A single hyperscale data center campus can cost tens of billions of dollars to construct once land, electrical infrastructure, cooling systems, fiber connectivity, and the GPUs and servers themselves are factored in. Industry-wide estimates for the AI and data center capital expenditure cycle now range from roughly $3 trillion on the more conservative end (Morgan Stanley, Moody’s Ratings) to more than $5 trillion when related power generation is included (JPMorgan)[1]. For context, U.S. data center power demand is projected to reach 35 to 45 gigawatts by 2030 — roughly double 2024 levels[2].
No single capital source, including the traditional bank lending market, can absorb financing needs at that scale. That mismatch is reshaping how developers fund these projects, and it has pushed a once-obscure corner of the debt markets — Rule 144A private placements — into the center of how the AI infrastructure boom gets built.
Why Traditional Lenders Are Hitting Their Limits
Commercial banks have historically been the natural first call for construction financing of large real estate and infrastructure projects. But data center development is straining that model in ways that are structural, not merely cyclical.
Concentration limits.
Banks operate under regulatory and internal limits on how much credit they can extend to any single borrower or counterparty. Oracle’s roughly $300 billion buildout commitment tied to its OpenAI partnership has been widely cited as the transaction that pushed major banks toward those limits[3]. Normally, a bank underwriting a loan of this size would syndicate pieces of it out to other institutional lenders to spread the risk. But because so much of the recent wave of data center debt is tied back to the same handful of hyperscale tenants and AI-linked counterparties — Oracle, Meta, Microsoft, Google, Amazon — the universe of buyers willing to take on more exposure to that same credit risk has narrowed. Banks have, in a number of recent deals, ended up holding more of the debt than they originally intended, which limits their appetite to originate the next deal.
Construction-phase risk is amplified by scale.
Data centers under construction face a unique combination of risks: multi-year grid interconnection queues (often three to four years), volatile equipment and labor costs, supply constraints on transformers and switchgear, and, increasingly, local and state-level political opposition to new projects[4]. Maine has moved toward becoming the first state to temporarily ban large-scale data center construction; similar resistance has surfaced in Virginia, Georgia, and Texas[5]. When a project’s completion timeline or cost base is less certain, the underlying loan becomes correspondingly riskier — and because data center loans are unusually large relative to a typical construction loan, even a handful of delayed or canceled projects can have an outsized effect on a bank’s overall exposure.
A genuine financing gap is emerging.
Analysts now estimate that something on the order of $150 billion to $200 billion of data center capital expenditure originally slated for 2026 is likely to slip into 2027 and 2028 — not necessarily because demand is softening, but because the financing, permitting, and construction pipeline cannot move fast enough to keep pace with announced project volume[6]. Of the roughly 16 gigawatts of U.S. capacity that had been expected to come online in 2026, only about 5 gigawatts were reported to be under active construction as of the spring; the remainder were in earlier, less-financed stages[7].
Enter Rule 144A: A Faster, Deeper Pool of Capital
Rule 144A, adopted by the SEC in 1990, creates a safe harbor allowing issuers to sell debt securities to Qualified Institutional Buyers (QIBs) — large, sophisticated institutional investors — without registering the offering with the SEC[8]. Because the securities are sold only to QIBs rather than to the general public, issuers can avoid the extensive disclosure obligations, review timelines, and ongoing reporting requirements that come with a registered public bond offering, while still tapping a genuinely deep and liquid pool of institutional capital.
For data center sponsors, that combination has proven to be exactly what the moment calls for. As one debt and structured finance executive put it, “what 144A offers is speed, scale and a broad institutional investor base that can move relatively quickly”[9]. A 144A deal can be structured and marketed in a fraction of the time required for a registered offering, and because QIBs are presumed to be sophisticated enough to evaluate risk without the same disclosure protections retail investors receive, issuers have considerably more flexibility in how the deal is documented and marketed.
The market has gone from nearly nonexistent for data centers a year ago to one of the fastest-growing financing channels in the sector[10]. Recent benchmark transactions illustrate the shift:
- Related Digital / Oracle (Saline Township, Michigan): Bank of America led a $14 billion 144A bond sale — anchored by roughly $10 billion from PIMCO — as part of a larger $16 billion financing package for a data center campus where Oracle is the anchor tenant, reportedly to support OpenAI workloads. The financing also included equity from Related Digital and Blackstone-affiliated funds[11].
- Meta / Blue Owl Capital: A 144A bond transaction structured around a special-purpose vehicle (SPV) that holds the underlying data center asset and is repaid from the long-term lease Meta signed with that SPV — insulating Meta’s own corporate balance sheet from the project debt while still giving bondholders investment-grade-quality cash flow backed by Meta’s credit as tenant[12].
- Compass Datacenters: A 144A offering backed by six recently developed or newly built properties, fully leased primarily to hyperscale tenants, that received the first AAA rating a major ratings agency has given to a hyperscale data center securitization[13].
The Mechanics: Why Lenders and Rating Agencies Are Comfortable With This Structure
What distinguishes these 144A data center deals from a more conventional unsecured corporate bond is the underlying collateral structure. In the typical arrangement, the data center real estate and the associated long-term lease or colocation agreement with a hyperscale tenant are placed into a bankruptcy-remote special purpose vehicle. The SPV, not the parent developer or the tenant itself, is the actual issuer or obligor on the bonds. Bondholders are repaid from the contracted lease cash flow, which — particularly where the tenant is an investment-grade hyperscaler such as Microsoft, Amazon, Google, or Meta — gives the debt a credit profile that can look more like infrastructure or commercial mortgage debt than corporate AI risk.
That structure has allowed rating agencies to get comfortable rating these instruments highly, even amid broader market unease about AI-sector credit concentration. As one credit analyst observed regarding the syndication market for these bonds, “these are all, at the end of the day, backed by very high-rated, well-structured leases. There is a difference between the risk of the debt and the risk of the market”[14]. In other words, a 144A bond secured by a fully-leased, already-built data center with a hyperscale tenant is a fundamentally different credit risk than an unsecured loan to a development-stage AI software company — even though both might get grouped together in popular commentary about “AI debt.”
This is also why 144A issuance is increasingly converging with, and sometimes feeding into, the broader securitization market. Single-asset/single-borrower (SASB) CMBS issuance for data centers grew from roughly $3 billion in 2024 to roughly $11 billion in 2025, and asset-backed securities (ABS) issuance — which pools multiple data center assets into a single offering — followed a similar trajectory, growing from a market that one industry estimate placed at around $25 billion in outstanding volume to one with projected refinancing and take-out needs approaching $300 billion[15]. Lenders who originate construction or bridge loans increasingly look to the 144A and broader capital markets as the eventual take-out source, refinancing bank or private credit debt once a project is stabilized, leased, and cash-flowing.
Filling the Gap Banks Can’t Close Alone
The practical effect is a layered capital stack that has emerged organically to match capital sources to the risk profile of each stage of a data center’s life cycle[16]:
- Land acquisition and early development is increasingly financed through land-cost facilities and structured equity, arriving earlier in the process than in prior cycles.
- Vertical construction still relies heavily on bank-led syndicated construction loans, project finance facilities, and private credit — but increasingly paired with mezzanine debt, preferred equity, or GPU-collateralized financing to bridge funding gaps banks alone cannot or will not fill.
- Stabilized, leased assets are where 144A bonds, CMBS, and ABS take over — refinancing the more expensive construction-phase capital with longer-duration, often investment-grade-rated debt sourced from a much larger and more diverse base of institutional buyers than any single bank or bank syndicate could provide.
Private credit funds — Blackstone, Blue Owl Capital, Apollo, PIMCO, and BlackRock prominent among them — have stepped into much of the space banks have had to vacate, both through direct lending and through off-balance-sheet SPV structures of the kind described above. Morgan Stanley has projected that private credit alone could provide an additional $800 billion in data center financing over the next two years[17]. But even private credit has limits, and the 144A market’s ability to tap a broader universe of QIBs — insurers, pension funds, asset managers — gives developers a release valve that neither bank balance sheets nor private credit funds can provide on their own.
A Departure From the Traditional Model: Bond Capital Arriving Before Completion
Historically, the entire reason rating agencies and bond investors could price 144A and securitized data center debt the way they do is that they were buying into stabilized cash flow — a completed building, leased to a creditworthy tenant, already generating rent. Bond investors are not, as a rule, in the business of being construction lenders: they lack the draw-monitoring infrastructure, the contractor oversight, and the risk appetite that banks bring to a project still under construction.
That sequencing is now breaking down at the margins of the market, precisely because no single bank or bank syndicate can underwrite the full cost of a $20–$30 billion campus. As Ropes & Gray observed following a recent industry conference, “debt capital is arriving earlier in development than ever before,” citing GPU financings, land-cost facilities, and forward-sale structures — arrangements in which a developer sells an in-development project to capture value sooner while explicitly retaining completion risk[18]. The Oracle/Related Digital financing for the Saline Township, Michigan campus is a useful illustration of this blurrier middle ground: the $16 billion package closed while construction was still very much underway, not after stabilization, because the project was too large for bank balance sheets to carry through to completion on their own[19].
The more precise way to describe what is happening, though, is not that bond investors have become comfortable underwriting construction risk the way a construction lender does. It is that construction risk is being contractually re-packaged and shifted to other parties — the EPC contractor, a completion guarantor, or the anchor tenant itself — before it ever reaches the bondholder. Three mechanisms illustrate how that risk transfer is being engineered:
- Completion guaranties and sponsor-level credit support. Lenders increasingly require sponsor-level credit support — including completion guaranties, cost-overrun funding obligations, and carry guaranties covering debt service through construction and lease-up — which shifts risk back toward the developer rather than the capital providers[20].
- Sponsor and tenant credit support keyed to completion milestones. Major tenants have in some cases agreed to firm rent-commencement dates regardless of construction delays — “a meaningful shift in risk allocation,” as Ropes & Gray put it — effectively asking the tenant to absorb timing risk that would otherwise fall on bondholders[21].
- Conditional credit backstops that activate only after completion. The clearest illustration is the $3.2 billion bond offering that financed TeraWulf’s data center in Barker, New York, leased to the AI infrastructure company Fluidstack. Google agreed to backstop Fluidstack’s lease obligations — stepping in to pay a termination fee or assume the leases directly if Fluidstack defaults — but that backstop activates only after construction is complete and the lease commences. During the build-out phase itself, bondholders bear full project execution risk with no Google support at all, and if construction is delayed by more than 180 days past the target completion date, Fluidstack can terminate the leases entirely, extinguishing Google’s backstop obligation before it ever attaches[22]. One rating agency analyst acknowledged in connection with the deal that “the construction schedule, in our view, is a bit aggressive or accelerated”[23].
This is also why the rating agency and analyst scrutiny noted above is concentrated so heavily on construction and completion assumptions. When a credit enhancement is structured so that it only attaches once the very risk it is meant to address has already passed, the rating it supports is, in substance, a bet that construction will go smoothly enough that the backstop is never tested. That is a meaningfully different proposition than rating debt against a building that is already finished and already collecting rent — and it is the feature of this cycle that most distinguishes it from the traditional, post-stabilization use of the 144A and securitization markets.
Risks and Open Questions
The speed and scale that make 144A offerings attractive also carry real risks that market participants and counsel are actively grappling with:
Disclosure asymmetry.
Because 144A securities are exempt from full SEC registration, the depth of publicly available information about these deals is more limited than for a registered bond offering[24]. QIBs are presumed sophisticated enough to demand and evaluate the information they need directly from issuers, but the broader market — and, indirectly, the regulators and policymakers monitoring systemic AI-related credit risk — have less visibility into the aggregate scale and terms of this debt than they would if it were flowing through registered channels.
Modeling assumptions under scrutiny.
Rating agencies and analysts are increasingly questioning the assumptions underlying data center financings, particularly around power availability and delivery timelines in newer markets[25]. Grid interconnection queues stretching three to four years, rising industrial power rates, and the small number of investment-grade hyperscale tenants able to anchor these deals all create concentration and timing risk that may not be fully priced into every transaction.
Tenant concentration.
A relatively small number of hyperscale tenants — Microsoft, Amazon, Google, Meta, and a handful of others — sit behind a large and growing share of the leases that collateralize this debt. That concentration is precisely what gives rating agencies comfort in pricing individual deals, but it also means the data center debt market’s overall health is unusually dependent on the continued creditworthiness and capital spending appetite of a small handful of companies.
Community and regulatory backlash.
Permitting fights, local moratoriums, and state-level legislative pushback against new data center construction are no longer fringe risks[26]. They introduce a form of political and regulatory uncertainty that is difficult to underwrite using traditional credit metrics, and that uncertainty is already showing up in how cautiously banks are willing to lend into new, less-established markets.
A genuinely new asset class for capital markets.
The scale of recent single transactions — a $14 billion bond tranche, a $16 billion total financing package, AAA ratings on instruments that did not exist as an asset class five years ago — reflects a market still developing its own standards, benchmarks, and stress-tested assumptions in real time, even as deal volume accelerates.
What to Watch
For developers, the lesson of the last twelve months is that no single capital source — not banks, not private credit, not the 144A and securitization markets alone — can fund a data center campus from groundbreaking through stabilization. The capital stack that has emerged blends all three, with each source filling the gap the others cannot: banks anchor construction-phase lending up to the limits their concentration rules allow; private credit funds and infrastructure debt provide the flexibility and speed banks cannot match for early-stage and bridge financing; and the 144A market — backed by SPV structures and long-term hyperscale leases — provides the scale and lower cost of capital needed to refinance stabilized assets and free up balance sheet capacity for the next project.
Whether that structure proves durable likely depends less on the financing technology itself than on the underlying fundamentals it is built on: continued hyperscaler demand for AI and cloud capacity, the pace at which the power grid can actually deliver the electricity these projects require, and the willingness of communities and state legislatures to permit the next wave of construction. The 144A market has proven remarkably effective at closing the funding gap traditional lenders cannot close on their own. Whether it can keep doing so at the pace the AI buildout now demands is the question the rest of this cycle will answer.
As data center developers, investors, and lenders continue to navigate an evolving financing landscape, understanding the legal and structural considerations behind 144A offerings, private credit arrangements, and project finance transactions will be increasingly important. Leech Tishman regularly advises clients on sophisticated financing, real estate, corporate, and capital markets matters that arise in connection with large-scale infrastructure and development projects. For assistance or additional information, please contact Forrest T. Passerin, a member of our Real Estate and Corporate Groups.
[1]Matt Wasielewski, The $3 Trillion AI Data Center Build-Out Becomes All-Consuming for Debt Markets, Insurance Journal / Bloomberg (Feb. 4, 2026) (citing estimates from Morgan Stanley, Moody’s Ratings, and JPMorgan).
[2]Ropes & Gray LLP, Data Center Investment in 2026: AI Demand, Power Constraints, and Private Equity Trends (May 2026).
[3]AI Data Centers Push Banks to Their Limits, eMarketer (May 5, 2026) (discussing Oracle/OpenAI-related bank concentration exposure).
[4]Ropes & Gray LLP, Data Center Investment in 2026, supra note 2 (grid interconnection timelines and Bring-Your-Own-Power trends).
[5]eMarketer, AI Data Centers Push Banks to Their Limits, supra note 3 (Maine moratorium and similar state-level resistance); on April 24, Maine Governor Janet Mills vetoed “An Act to Establish the Maine Data Center Coordination Council and Place a Temporary Limitation on Certain Data Centers” (LD 307), which would have made Maine the first state to impose a statewide moratorium on data center development. Although the Governor vetoed the bill, it did pass both chambers of the Maine Legislature with bipartisan support and the Governor indicated the only reason that she vetoed the bill was because it lacked an exemption for one project.
[6]Half of 2026 US Data Centers Delayed: CRE Impact Guide, The AI Consulting Network (May 13, 2026) (citing Sightline Climate research as reported by Bloomberg and Tom’s Hardware).
[7]The AI Consulting Network, Half of 2026 US Data Centers Delayed, supra note 6.
[8]What is SEC Rule 144A?, DFIN (Apr. 1, 2026).
[9]Matt Wasielewski, Wall Street Is Funding The AI Build-Out With Bonds The SEC Never Sees, Bisnow (June 11, 2026) (quoting Chris Lozinak, Managing Director, Newmark Global Debt and Structured Finance).
[10]Bisnow, Wall Street Is Funding The AI Build-Out With Bonds The SEC Never Sees, supra note 9.
[11]Oracle Data Center $16 Billion Financing Gets Over The Line, Bloomberg, republished on Yahoo Finance (Apr. 25, 2026).
[12]Bisnow, Wall Street Is Funding The AI Build-Out With Bonds The SEC Never Sees, supra note 9 (Meta/Blue Owl Capital 144A transaction, October 2025).
[13]Bisnow, Wall Street Is Funding The AI Build-Out With Bonds The SEC Never Sees, supra note 9 (Compass Datacenters 144A offering and Moody’s AAA rating, October 2025).
[14]Bisnow, supra note 9 (quoting John Medina, Moody’s Ratings).
[15]2026 Outlook: Capacity Growth Remains Robust as Tenants Prioritize Speed to Market, Data Center Dynamics (Mar. 31, 2026) (SASB CMBS issuance figures); Ropes & Gray, supra note 2 (ABS market size and projected take-out needs).
[16]Ropes & Gray LLP, Data Center Investment in 2026, supra note 2.
[17]Quinn Emanuel Urquhart & Sullivan LLP, Client Alert: Emerging Litigation Risks in Financing AI Data Centers Boom (Mar. 13, 2026) (citing Morgan Stanley’s $800 billion two-year private credit projection).
[18]Ropes & Gray LLP, Data Center Investment in 2026, supra note 2 (forward sale structures and debt capital arriving earlier in development).
[19]Bloomberg, Oracle Data Center $16 Billion Financing Gets Over The Line, supra note 11.
[20]Foley & Lardner LLP, Financing the Data Center Boom: Trends, Structures, and Considerations for Market Participants (Apr. 2026), https://www.foley.com/insights/publications/2026/04/financing-the-data-center-boom-trends-structures-and-considerations-for-market-participants/ (describing completion guarantees, cost-overrun funding obligations, and carry guarantees covering debt service during construction and lease-up as sponsor-level credit support mechanisms in data center project finance).
[21]Ropes & Gray LLP, Data Center Investment in 2026, supra note 2 (“in some cases, hyperscalers agree to firm rent commencement dates regardless of construction delays—a meaningful shift in risk allocation”).
[22]Aaron Weinman & Gowri Gurumurthy, ‘Google Backstop’ Adds New Twist to Data Center Financing Frenzy, Bloomberg (Oct. 16, 2025); see also Quinn Emanuel Urquhart & Sullivan LLP, Financing the AI Infrastructure Boom: Emerging Litigation Risks in AI Data Centers, at 5, 13-14 (describing the structure of the $3.2 billion TeraWulf/Fluidstack/Google transaction).
[23]‘Google Backstop’ Adds New Twist to Data Center Financing Frenzy, supra note 22 (quoting Anubhav Arora, Senior Director, Fitch Ratings).
[24]DFIN, What is SEC Rule 144A?, supra note 8.
[25]Ropes & Gray LLP, Data Center Investment in 2026, supra note 2 and accompanying text.
[26]The AI Consulting Network, Half of 2026 US Data Centers Delayed, supra note 6 (discussing Maine, Virginia, Georgia, and Texas community opposition).