When a neocloud fails, the sequence runs from covenant breach to forbearance, SPV enforcement, lease default and asset sale, and the prize is the powered hall rather than the GPUs. With North American vacancy at 1% (JLL, North America Data Center Report Midyear 2026, Aug 2026) and wholesale rates at $140 to 155 per kW per month (Data Center Knowledge, Apr 2026), the landlord usually recovers first and most.
This supporting page to the stranded GPU assets pillar describes the sequence qualitatively.
What triggers the default?
Almost never a missed payment. GPU-backed loans carry utilization, coverage and collateral-value tests, and coverage usually breaks first. In StackedAI’s model, a GPU-only operator at 64% loan-to-cost sees DSCR fall below 1.0x by year five when hourly pricing decays 20% per year (StackedAI analysis, AI Compute Partnership Returns Model, Jul 2026).
The market signals it first. AI-linked credit default swaps surged in July 2026 (Bloomberg, Jul 2026), and a wholesale colocation deal with a neocloud tenant failed at $155 to 160 per kW on a 15-year term with six months of prepayment (Data Center Knowledge, Apr 2026). Vendors see it six to eighteen months earlier, through deferred invoices (StackedAI analysis, Stranded Asset Report, Aug 2026).
What happens between breach and enforcement?
The first response is forbearance: the lender agrees not to exercise remedies for a defined period in exchange for fees, tighter reporting or more collateral. That buys the operator time to sign take-or-pay contracts or raise equity.
If forbearance fails, enforcement follows the structure. Most deployments hold the GPUs in a special purpose vehicle, with the loan secured on the SPV’s assets and, ideally, on customer contracts assigned to it. The lender takes control of the SPV, replaces its manager, and either operates the fleet or markets it. Whether the contracts survive that change of control is the decisive variable, which is why StackedAI’s screening framework scores it separately.
The SPV cuts both ways: it shields the GPU lender from the parent’s other creditors, and it keeps the colocation lease out of the lender’s estate, so the landlord cannot reach the GPUs.
What does the colocation landlord do?
The landlord holds the strongest hand. The lease is secured by a deposit or prepayment; the stalled April 2026 deal carried six months of prepayment (Data Center Knowledge, Apr 2026). On default:
- Declare default and apply security. Draw the deposit or prepayment and issue notice.
- Exercise step-in rights where the lease has them. Some AI-era leases let the landlord assume the tenant’s customer contracts; most limit step-in to space and power.
- Negotiate with the GPU lender. The collateral sits in the landlord’s hall, drawing the landlord’s power; the landlord can charge holdover or demand removal, and the lender needs time to sell.
- Re-let the hall. With vacancy at 1% for three consecutive years and rents up about 70% since 2020 (JLL, North America Data Center Report Midyear 2026, Aug 2026), and wholesale pricing at $140 to 155 per kW per month (Data Center Knowledge, Apr 2026), a fitted, liquid-cooled hall re-lets quickly.
The landlord’s exposure is the gap between defaulted rent and new rent, less security, and in this market that gap is often negative. A landlord that let a powered shell rather than a turnkey hall may also inherit the tenant’s liquid-cooling fit-out. This is why StackedAI holds that the NNN lease carries the deal: it survives the tenant.
How do GPU lenders recover?
Through a collateral waterfall: enforcement costs and the liquidation advisor’s fees first, then the senior secured lender on the SPV up to its claim, then any junior holder, then the SPV’s equity, usually the parent and its creditors.
The problem is the size of the proceeds. In StackedAI’s base case the lender advanced about 60% of hardware cost against collateral whose residual is modeled at about 15% of cost after five years (StackedAI analysis, AI Compute Partnership Returns Model, Jul 2026). Air-cooled PCIe servers sell into a broad market; liquid-cooled rack-scale systems only go to buyers with compatible cooling plant. The GPU depreciation page covers the pricing evidence.
The lender’s best outcome is usually a going-concern sale of the SPV with its contracts, which only works if they survive change of control.
Who buys what, and why is the power position the prize?
| Stakeholder | What it holds | What it typically recovers | Recovery driver |
|---|---|---|---|
| Colocation landlord | Lease, security, the powered and cooled hall | Security plus re-letting; often a net gain at 1% vacancy (JLL, Aug 2026) | Wholesale rate of $140 to 155 per kW per month (Data Center Knowledge, Apr 2026) |
| Senior GPU lender | Security over the GPU SPV and, if assigned, its contracts | Going-concern sale of the SPV, or hardware liquidation at a deep discount | Contract survival on change of control; generation; form factor |
| Hyperscaler or AI-lab customer | Take-or-pay contract, sometimes a lease guarantee | Continuity if the buyer assumes the contract | Whether it guaranteed the lease, as Google did for about $1.3B of TeraWulf-Fluidstack obligations (TeraWulf, Oct 2025) |
| Vendors and integrators | Unsecured trade claims | Usually pennies | Retained title or purchase-money lien |
| Parent equity | Common equity in the operating company | Nothing in most scenarios | Residual after all claims |
| Strategic buyer | Cash and a platform | The lease, the power, the fitted hall, GPUs at a discount | Speed to power in a market with no meaningful easing before 2030 (Cushman & Wakefield, Feb 2026) |
The strategic buyer turns a distressed situation into a transaction. Operators, hyperscalers and infrastructure funds want the utility capacity and the liquid-cooled hall because those take years to reproduce; they price on the power and the lease. In StackedAI’s phrase, the building carries the deal.
Key terms
- Covenant breach: failure to meet a loan test, giving the lender remedies even if payments are current.
- Forbearance: a lender’s agreement not to exercise remedies for a defined period, usually for fees or tighter terms.
- Collateral waterfall: the order in which sale proceeds go to costs, senior lenders, junior lenders and equity.
- Step-in rights: a right allowing a landlord, lender or customer to take over an operator’s obligations or assets on default.
How StackedAI applies this
StackedAI works with restructuring advisors, lenders and landlords when a neocloud tenant or borrower shows stress. It separates the deployment into its four asset layers, identifies which stakeholder holds each, and models recovery under going-concern and liquidation cases. For landlords it benchmarks re-letting against wholesale pricing; for lenders it tests whether contracts survive change of control and whether the form factor supports resale.
Frequently asked questions
Has any major neocloud gone bankrupt?
As of August 2026, StackedAI’s review of major outlets found no reported bankruptcy of a significant neocloud. The stress shows elsewhere: AI-linked credit default swaps surged in July 2026 (Bloomberg, Jul 2026) and wholesale colocation deals with neocloud tenants stalled over credit risk (Data Center Knowledge, Apr 2026).
What is the first sign a neocloud is in trouble?
Usually a covenant breach on a GPU-backed loan rather than a missed payment. In StackedAI’s model the GPU-only structure’s DSCR falls below 1.0x by year five at 64% loan-to-cost under a 20% annual price decay (StackedAI analysis, AI Compute Partnership Returns Model, Jul 2026). Vendors see unpaid invoices months earlier.
What can a colocation landlord do when a neocloud tenant defaults?
Declare default, apply the prepayment or deposit, exercise any step-in rights, and re-let. With North American vacancy at 1% (JLL, Aug 2026) and wholesale pricing at $140 to 155 per kW per month (Data Center Knowledge, Apr 2026), a landlord with a liquid-cooled hall is usually the strongest stakeholder.
How much do GPU lenders recover?
It depends on generation and form factor. A lender that advanced about 60% of hardware cost against collateral modeled at about 15% residual after five years (StackedAI analysis, AI Compute Partnership Returns Model, Jul 2026) can be under water within the term, especially on liquid-cooled rack-scale systems.
Why is the power and lease position the prize?
Because it is the only layer that cannot be reproduced quickly. GPUs can be ordered; a powered, cooled hall with utility capacity in a 1% vacancy market cannot (JLL, Aug 2026). Strategic buyers price the deal on the lease and the power and treat the silicon as a discount item.
Sources
- StackedAI analysis, AI Compute Partnership Returns Model, Jul 2026 (internal, illustrative)
- StackedAI analysis, Stranded Asset Report, Aug 2026 (internal)
- Data Center Knowledge, Neocloud Storm Gathers as Data Center Deals Stall Over Credit Risk, Apr 2026, https://www.datacenterknowledge.com/cloud/neocloud-storm-gathers-as-data-center-deals-stall-over-credit-risk
- JLL, North America Data Center Report Midyear 2026, Aug 2026, https://www.jll.com/en-us/insights/market-dynamics/north-america-data-centers
- Cushman & Wakefield, Americas Data Center Market Shifts to Managed Growth, Feb 2026, https://www.cushmanwakefield.com/en/united-states/news/2026/02/americas-data-center-market-shifts-to-managed-growth
- Bloomberg, Wall Street picks AI winners and losers as credit swaps surge, Jul 2026, https://www.bloomberg.com/news/articles/2026-07-29/wall-street-picks-ai-winners-and-losers-as-credit-swaps-surge
- TeraWulf, press release: TeraWulf Expands Strategic Partnership with Fluidstack Through New 168 MW AI Compute Joint Venture, Oct 2025, https://investors.terawulf.com/news-events/press-releases/detail/121/terawulf-expands-strategic-partnership-with-fluidstack-through-new-168-mw-ai-compute-joint-venture