If you’re evaluating a colocation contract or a GPU-cloud commitment right now, the questions your finance team should be asking have quietly changed. Power availability and site location still matter, but they are no longer the biggest risk in the deal. The bigger risk is whether the company on the other side of your signature can actually make good on a 10- or 15-year obligation that, in a growing number of cases, is itself backed by financing nobody outside a 10-K footnote can fully see.
That is not a hypothetical. Moody’s Ratings put a number on it this year: five major hyperscalers — Amazon, Alphabet, Microsoft, Meta, and Oracle — are carrying roughly $969 billion in undiscounted future data center lease commitments, and $662 billion of that is for leases that haven’t even started yet, sitting entirely off their balance sheets under current GAAP treatment. Moody’s analysts David Gonzales and Alastair Drake noted that figure equals 113% of the five companies’ combined adjusted debt. In other words, the capacity buildout you’re competing for is now larger, on paper, than the companies funding it.
The $662 Billion Hiding Off the Balance Sheet
The mechanics matter for anyone signing a contract downstream of a hyperscaler or neocloud. Alphabet’s disclosed uncommenced lease payments jumped from $23.9 billion in Q2 2025 to $42.6 billion by Q3, with commencement dates stretching out to 2031 and terms running as long as 25 years. Meta has taken a different route on some deals, committing to leases that don’t start until 2029 while backing them with a $28 billion residual value guarantee — a promise to cover the facility’s value if the lease ends early. Both are legal, both are disclosed, and both are structured specifically to keep the obligation off the primary balance sheet while still locking in the capacity.
None of this makes the hyperscalers themselves a credit risk — they remain investment-grade, cash-rich companies. The risk transfers downstream, to the layer underneath them: the neoclouds, GPU-collateral lenders, and specialty REITs who are taking on debt predicated on those same hyperscaler commitments continuing on schedule. If you’re buying capacity from that second layer, you’ve inherited exposure to a structure you didn’t negotiate and can’t see in full.
The Financing Loop Feeding the Buildout You’re Competing For
Layered on top of the lease structures is a financing pattern the Bank for International Settlements flagged in its 2026 Annual Report as one of the three biggest risks to global financial stability right now: circular financing. The pattern is straightforward once you see it. Nvidia invests in or extends credit to a neocloud — CoreWeave, Nebius, Nscale — which then commits to buy Nvidia chips with that same capital, while a hyperscaler signs a multi-year compute purchase agreement with the neocloud that makes the whole loop bankable. Microsoft has roughly $60 billion in commitments to CoreWeave, Nebius, and Nscale combined; Meta has committed $35.2 billion to CoreWeave and up to $27 billion to Nebius. Nvidia itself backed CoreWeave with a $6.3 billion backstop agreement running through April 2032, on top of roughly $2 billion in direct equity.
CoreWeave’s own numbers show what that financing is covering: $24.9 billion in total debt, $7.7 billion in quarterly capex, and negative $4.71 billion in free cash flow as of its most recent quarter, even with revenue up 112% year over year. Mizuho chip analyst Jordan Klein put the concern bluntly: “It smells like you are pre-funding the purchase of your own GPUs.” Supporters of the structure have a real counterargument — in a capacity-constrained market, pairing long-term purchase commitments with financing is how you lock in supply at all, and most of these deals are disclosed, rated, and backed by real contracted backlog (CoreWeave reports $99.4 billion of it). The honest summary sits between the two: the arrangement isn’t inherently unsound, but it concentrates risk in places a traditional balance-sheet review won’t find.
For a buyer, the practical takeaway isn’t to avoid neocloud providers — it’s to know which loop your vendor sits inside, and what happens to your contract if one link in it breaks.
Need Expert Guidance?
Talk to a Data Center Expert
21+ years of hands-on experience in data center design, operations & infrastructure. Book a quick discovery call to discuss your project.
📞 Book a Discovery CallWhy a Neocloud Lease Doesn’t Behave Like a Hyperscaler Lease
Credit markets have already started pricing the difference, and the gap is the clearest buyer signal in this whole story. Lending spreads on neocloud debt are running 275 to 325 basis points above investment-grade benchmarks, according to Dave Ferdman of Primary Digital Infrastructure — a premium that exists precisely because the underlying tenant credit isn’t hyperscaler-grade. Synergy Research Group clocked neocloud segment revenue up 223% year over year in the most recent quarter it tracked, and JLL now counts more than 190 neocloud operators in the market. Growth is real. So is the structural mismatch underneath it: hyperscaler leases typically run 15 years or longer against investment-grade balance sheets, while the customer contracts many neoclouds depend on to service that same debt often run just four to five years, and in some wholesale GPU-rental cases, far shorter. One infrastructure analyst described the resulting vehicle as one that “finances GPUs like power plants, collateralizes them like aircraft” — long-duration debt against an asset whose economic life, and whose tenant’s staying power, is far less certain.
That duration mismatch is exactly what sank WeWork, and data center analysts are now using that comparison explicitly. CoreWeave’s own $8.5 billion financing facility, closed in March 2026, used Nvidia GPUs as direct collateral and still needed an investment-grade wrapper to clear the market — a sign that lenders already treat the hardware-backed structure as the exception requiring extra protection, not the norm.
What This Means for a Build-Buy-or-Colocate Decision
This changes the calculus in the build-versus-buy-versus-colocate math that most capacity buyers are already running. A colocation contract with an investment-grade hyperscaler-anchored facility carries a different risk profile than a wholesale capacity agreement with a neocloud whose own revenue is concentrated in one or two anchor tenants and whose debt was priced at a 300-basis-point premium. Backfill timelines for specialized, power-dense space also run 12 to 18 months if a tenant fails — long enough to matter if your own business plan assumed uninterrupted access to that capacity.
None of this is a reason to avoid neocloud or specialty providers; for many AI workloads they remain the fastest, sometimes only, path to near-term GPU capacity. It is a reason to underwrite the counterparty the way a lender would, not the way a traditional colocation customer historically could afford to.
The Due-Diligence Checklist Before You Sign
Before committing capacity spend to a neocloud or specialty data center provider in 2026, buyers should be able to answer five questions with specifics, not reassurances: What share of the provider’s contracted backlog comes from a single hyperscaler customer, and what happens to your contract if that customer renegotiates? What is the provider’s actual credit rating on recent debt facilities, and from which agency — Moody’s A3 and DBRS A-low ratings mean something different than a Fitch BB+ on the same company’s later facility. Is any of the provider’s financing collateralized directly by the hardware serving your workload, and what triggers a backstop or margin call? How long is the provider’s own debt maturity schedule relative to your contract term, and is there a renewal gap in between? And finally, what residual value guarantees or off-balance-sheet structures sit behind the facility you’re contracting into — ask for the disclosure, not the summary.
These aren’t questions procurement teams were trained to ask three years ago, when the main diligence items were power availability, latency, and physical security. They are now standard underwriting for anyone treating data center capacity as critical infrastructure rather than a commodity lease, as this site has argued is also true for risk coverage in data center financing more broadly.
The capacity shortage that has defined this market for two years hasn’t gone away, and it won’t resolve the underlying tension: everyone building this capacity needs the financing to keep flowing, and the financing increasingly depends on everyone else’s commitments holding. That’s not a reason to sit out. It’s a reason to read the footnotes, rate the counterparty, and build a contract that survives a link in the chain breaking. Start with TechInfraHub’s data center buyer’s toolkit for the underwriting checklists and vendor-comparison frameworks to put this into practice before your next contract goes to signature.
Written by
Raajeev Ratra
Data Center Infrastructure Expert | 15+ Years in DC Design, Operations & Project Management
Raajeev is a seasoned data center professional with hands-on experience in hyperscale facilities, colocation design, power & cooling infrastructure, and global DC operations. He shares practical insights to help engineers and IT leaders build better infrastructure.