What is take-or-pay?

A take-or-pay obligation requires the buyer to pay for a contracted quantity over a defined period whether or not it takes delivery. If it takes the output, it pays. If it does not, it still pays.

The condition on the other side is that the seller must stand ready to deliver. Take-or-pay is not an unconditional promise to pay money; it is a promise to pay for availability. Where the seller cannot make the output available — the plant is down, the capacity is not there, the equipment has not been delivered — the obligation typically abates. That symmetry is the whole design: the buyer accepts volume risk, the seller accepts performance risk, and each carries the one it can control.

Related formulations do the same work under different names. A capacity payment in a power contract pays for the right to call on generation regardless of dispatch. A minimum commitment sets a floor with consumption priced above it. A reserved instance or committed-capacity arrangement in a compute contract pays for capacity held aside. The label varies by layer; what matters is whether the payment survives the buyer not using the thing.

The reason this distinction dominates every underwriting conversation is simple, and it is worth stating in one line: a take-or-pay obligation removes the buyer's own demand from the analysis.

Under a consumption contract, whoever finances the asset is underwriting two things at once — that the buyer can pay, and that the buyer will want the output. The second is a forecast about somebody else's business, made by a party with no visibility into it and no ability to influence it. Under take-or-pay, only the first question survives. The analysis collapses from "will this market develop" to "can this named party meet a fixed obligation", which is a credit question with established tools.

That is also why the strength of a take-or-pay contract is capped by the payer. A firm obligation from a party that cannot meet it under stress is a well-drafted document rather than a cash flow, which is why credit support — a parent guarantee, a [letter of credit](letter-of-credit), a cash-funded reserve — so often appears alongside one. The obligation and the ability to meet it are two separate tests, and both have to pass.

The ranking below is the one every party financing an asset applies, whatever the contract is called on its cover page.

Most disappointment with take-or-pay comes not from contracts that lack the obligation, but from contracts that have it and then give it back somewhere else in the document. The clauses that do the giving back are consistent across layers.

• Conditions precedent. If the obligation only begins on an event that has not happened — energisation, acceptance testing, a permit, a delivery date — the contract is a conditional obligation and the condition is the real risk.
• Availability and performance tests. The threshold at which the seller is deemed to have made the output available, and what happens between the contractual threshold and actual delivery. A demanding availability standard shifts real risk back onto the seller and reduces what the contract supports.
• Termination for convenience. A right to exit on notice converts a multi-year obligation into the notice period. This is the single most common gap between how long a contract is described as running and how long it can be relied upon.
• Change of control and assignment. Whether the obligation survives the buyer being acquired or restructured, and whether it can be transferred to a weaker entity.
• Force majeure scope. Broadly drafted, it can excuse payment for events well beyond the physical ones the concept exists for.
• Set-off and netting. A right to net disputed amounts against payments turns an unconditional obligation into a contested one at exactly the moment it matters.
• Caps and liquidated damages. A total liability cap below the remaining contracted value silently converts the tail of the obligation into an unsecured expectation.

None of these makes a contract defective. They are ordinary commercial terms and each is negotiable. The failure is not reading them together and discovering, late, that the firm contract in the model was a twelve-month obligation with a two-year name.

Each layer of the stack has a take-or-pay analogue, and the layers do not automatically line up with one another.

Generation. A power purchase or tolling agreement with a capacity payment is the classic case: the buyer pays for firm availability whether or not it dispatches. This is the obligation that supports long-dated capital against generation equipment, and it is the reason the power layer is often the most conventionally financeable part of a site.

Shell. A build-to-suit or wholesale capacity lease is take-or-pay in substance — committed kilowatts paid for whether or not racks are ever installed behind them. Whether the obligation abates when the facility cannot deliver power or cooling is the clause worth reading twice.

Compute. Reserved capacity paid regardless of utilisation is the only form of compute revenue that behaves like credit. Because accelerators offer weak recovery, the contract has to carry more of the underwriting here than at any other layer — the full version of that argument sits at [compute offtake as credit](/compute/compute-offtake-as-credit).

The cross-layer failure is the one worth naming. A take-or-pay compute contract that runs longer than the facility's power commitment, or a facility lease that runs longer than the site control beneath it, is an obligation resting on something that expires first. Matching those durations is the subject of [tenor](tenor), and the consequence of failing to is the argument at [tenor mismatch](/compute/tenor-mismatch-compute-and-infrastructure).

Continuum's work at this layer is structural: establishing what each obligation actually obliges, where the abatement and termination rights sit, and arranging each layer against an obligation that survives the tests the layer above will put it through.