What is a stranded asset?

A stranded asset is one that still functions but can no longer earn enough to service the capital committed against it. The equipment runs. The building stands. The capacity exists. And none of it produces a return sufficient to carry what was borrowed or invested to create it.

Three distinctions make the term usable rather than merely pejorative:

• Obsolete means the asset no longer works, or no longer works well enough to be used. A stranded asset frequently works perfectly.
• Impaired is an accounting determination — a carrying value written down. Impairment often follows stranding and is not the same event.
• Idle is temporary. Stranding is the condition of having no path back to earning within the life of the capital behind it.

The most important property of the term is that it is relational, not intrinsic. An asset is not stranded in the abstract; it is stranded relative to a structure and a set of counterparties. The same generation set is a strong asset under a matched-tenor lease with a redeployment path and a stranded one under a facility that assumed it would run continuously for fifteen years at a site that no longer needs it. The same accelerators are a productive fleet inside an operator with contracted demand and dead weight inside one without.

Which is the reason this term belongs on a structuring site rather than an engineering one: strandedness is usually created at the point the deal is arranged, and only observed years later.

Each layer of a data-center stack has a characteristic failure. They look different, and each one is the same event — an asset outliving its ability to earn.

The causes above are visible with hindsight. The structural decisions that produce them are visible in advance, and there are only a handful.

Tenor mismatch. Capital with a term unrelated to the asset's economic life is the largest single source. A long facility against a short-lived asset outlives its collateral; a short facility against a long-lived one forces a refinancing at a moment nobody chose. Where a single instrument spans layers with different lives, both happen at once — the argument worked through in full at [tenor mismatch: financing compute inside infrastructure](/compute/tenor-mismatch-compute-and-infrastructure).

Contract shorter than the asset. Equipment contracted for three years and financed over eight has five years of assumption in it. That assumption is legitimate and it should be labelled as one, because it is the point at which the structure depends on a market rather than on a contract.

Single-purpose specification. Every increment of customisation narrows the set of parties who could use the asset next. A shell built to one tenant's configuration, a substation sized to one load profile, a cluster networked for one workload — each trades residual value for present efficiency, often without anyone pricing the trade.

Location lock. An asset whose removal costs more than it fetches is immobile whatever its nominal market value — which is why permits and installation at the *destination* matter to a residual.

Cross-collateralisation across layers. Tying a healthy layer to a failing one so the failure propagates. A functioning compute business inside a structure secured on a facility whose power position lapsed is not a functioning business any more — the seam did the damage, not the asset.

Counterparty concentration. An asset whose only realistic user is one counterparty holds that counterparty's credit regardless of what the security documents say. This is the specific exposure a [neocloud](neocloud) tenant creates for a landlord, and the specific exposure a single anchor customer creates for the neocloud.

The mitigations are the mirror image of the causes, and each is a term in a document rather than a property of the equipment.

Separate the layers before financing them. Capital matched to each layer's life prices that layer's risk and nothing else. A structure that separates land, shell, generation and compute has four assets each of which can fail without the others — which is the whole argument of [the data-center capital stack](the-data-center-capital-stack).

Match tenor to economic life, and say what happens at the end. Every asset in the stack reaches a point at which it must be re-contracted, refinanced, redeployed or realised. A structure that names that point and the intended action at it is describing an exit; one that does not is deferring the discovery.

Price the residual honestly, and separately. Residual value is a judgement about a future market, not an extrapolation of a purchase price. On the compute layer it is the dominant variable and the least stable — the subject of [GPU residual value and depreciation](/compute/gpu-residual-value-and-depreciation).

Keep the asset redeployable, and keep access to it. Generic specification, transferable contracts, documented removal rights and a realistic destination. Every remedy over physical equipment also requires the ability to reach it, so where the party controlling the site is the counterparty in difficulty, a perfect security position with no access is stranded in the most literal sense — see [collateralizing compute](/compute/collateralizing-compute).

Match the seams. Power contracted for less time than the compute contracts above it, site control shorter than the build, or a hosting agreement shorter than the equipment life each guarantee an event that the structure has no answer for. A seam that expires first strands everything above it.