Neocloud vs hyperscaler offtake

The building is the same, the power is the same, the capacity is the same. What differs is the party obliged to pay for it — and everything the structure can support flows from that.

The mechanism is the one that governs every contracted asset: capital underwrites the party obliged to pay, not the party expected to succeed.

A hyperscaler obligation is a claim on a large, diversified, externally assessed business. Its payment does not depend on any particular workload succeeding, and its ability to pay in a downturn is not in serious question. That is why the tenor can be long — capital is willing to look out ten or fifteen years at a counterparty whose survival over that period is not the analysis. The lower price is the direct cost of that credit: the sponsor is being paid partly in certainty.

A neocloud obligation is a claim on a business whose own revenue comes from contracts it may still be winning. The obligation may be perfectly genuine and the company well run, but the recovery analysis has to work through a further layer: what pays this, and what happens to it if the market for compute softens. That is why terms are shorter, why the pricing is better, and why credit support appears — a parent guarantee, a letter of credit, a deposit or prepayment mechanic exists precisely to bridge the gap between the contract's quality and the counterparty's.

The specification and negotiating rows are the ones sponsors feel most and mention least. A counterparty that can place its capacity requirement anywhere sets the terms of the building it will occupy — density, redundancy, commissioning standards, service levels, and the remedies attached to them. Those requirements are not unreasonable, but they consume the sponsor's flexibility and they cost money that does not appear in the rate.

The last row on each side is the honest summary. Hyperscaler offtake fails at the point where the price and the specification together make the project not worth building. Neocloud offtake fails at the point where the counterparty's own customers do not renew. Those are different risks with different mitigations, and neither is smaller than the other in the abstract.

What separates a bankable contract from a signed one, on either side, is set out on compute offtake as credit.

Hyperscaler offtake wins wherever the structure behind the site needs a long, unquestioned obligation.

The financing is long-dated and ring-fenced. A project structure that intends to look out over the life of the building needs a counterparty capable of being looked at over that period. Where the plan is a limited-recourse structure standing on the contract alone, counterparty quality is not one factor among several — it is the credit, and it is the gate examined on corporate credit vs project finance.

The sponsor has no balance sheet to lend. A developer whose corporate capacity cannot absorb a shortfall needs the contract to carry the structure without help. That is exactly what a strong counterparty provides.

Forward funding or an early exit is the plan. A funder committing capital during construction, or a buyer contracted to take the completed asset, is buying the cash flow. The stronger the covenant, the earlier and cheaper that commitment is available — the mechanics are on construction loan vs forward funding.

The specification is deliverable. Where the requirements match what the site can economically provide, the trade is straightforward: accept the price, get the credit.

It is the wrong answer where the specification and the price together leave no margin. A long contract at a rate that does not cover the cost of the building it requires is a long problem, and the strength of the counterparty makes it harder to escape rather than easier. It is also the wrong answer where the sponsor's whole portfolio would then depend on one group — concentration in the strongest available counterparty is still concentration, and the perimeter question it raises is on single-asset vs portfolio financing.

One term deserves reading before any of the commercial ones: what the counterparty may do if the sponsor underperforms. Service-level remedies, abatement rights and termination triggers are where an apparently unimpeachable contract turns out to be conditional, and capital reads them before it reads the rate.

Neocloud offtake wins where the sponsor can absorb the credit gap, or where something in the structure closes it.

The sponsor has credit capacity of its own. A sponsor financing on corporate credit is standing behind the shortfall anyway, so the counterparty's credit matters commercially rather than structurally. That sponsor can take the better pricing and manage the risk directly.

Credit support is genuinely available. A parent guarantee from a substantial entity, a letter of credit from a bank, a meaningful deposit or a prepayment mechanic can move a contract from unfinanceable to financeable. The support has to be real: a guarantee from an entity with no assets, or a letter of credit that expires long before the contract does, closes nothing. Reading the support instrument as carefully as the contract is the whole of the work here.

The capital behind the site is shorter. A structure that does not need to look out fifteen years does not need a counterparty that survives fifteen years. Matching capital tenor to contract tenor is the general principle, and it is why tenor mismatch governs counterparty selection as much as it governs asset selection.

Speed is the binding constraint. Where a power position, a queue slot or an equipment allocation expires on a date, a counterparty that can sign this quarter may be worth more than a better one that signs next year. That is a real trade and it should be made explicitly rather than by drift.

Diversification is the objective. Several independent counterparties can be a better credit position than one, provided they are genuinely independent and genuinely differently exposed.

It is the wrong answer where the sponsor is relying on a limited-recourse structure with no support, where the term is materially shorter than the capital behind the building, or where several apparently separate counterparties turn out to share one underlying source of demand. That last failure is the one that looks like diversification and is not — and it is the same analysis, applied at the level above, that financing a neocloud against an enterprise cluster applies to the cluster owner itself.

The counterparty decision is usually presented as a choice between offers. It is more useful to treat it as a set of tests that both offers have to pass.

Is it an obligation to pay, or a right to consume? The single most important distinction, and it survives no matter who signed. A commitment that lets the counterparty reduce or stop taking capacity is not offtake, however impressive the name at the top of the page.

Does the term cover the capital behind it? A contract shorter than the structure it supports leaves a recontracting gap, and that gap is a position someone is taking. It should be identified and allocated at the outset rather than discovered at expiry.

What is the payment obligation actually conditioned on? Availability standards, service levels, commissioning milestones and acceptance tests all convert an unconditional-looking obligation into a conditional one. Capital reads these closely because they are where the contract's real strength sits.

Who is the contracting entity, and what does it own? A subsidiary formed for the purpose is a different counterparty from its parent, and the difference is the guarantee. This is the most common gap between how a contract is described and what it is.

What happens on a default — and can anything be re-let? Capacity built to one counterparty's specification may not suit another's without work. A site whose configuration is generic re-lets; one built to a single requirement may not, and that is a real difference in recovery regardless of who the original counterparty was.

Does the site itself clear the other gates? A counterparty cannot rescue a site with an unresolved power position, control that expires before energisation, or an environmental question nobody has asked. The gates are conjunctive, and they are set out on what makes a site financeable.

The useful discipline is to write down what capital would recover if the counterparty stopped paying in year three. Under a hyperscaler contract the answer is usually that it would not stop. Under a neocloud contract the answer is a chain — its customers, its equity, its support instruments — and every link in that chain should be nameable before the structure is built on it.