What Is a Take-or-Pay Contract? How It Works in GPU Financing
Summary
- Definition: A take-or-pay contract obliges the buyer to take a minimum quantity or pay for it anyway, and paying for the shortfall counts as performing the contract.
- Why it matters: It moves volume risk to the buyer, which turns a GPU deployment's revenue into something a lender can size a loan against.
- In GPU compute: Large reserved-capacity GPU contracts use it: CoreWeave reports that committed take-or-pay contracts made up 98% of its Q2 2026 revenue.
Introduction
A GPU cloud operator that signs a customer and orders servers has to finance most of the cost before the first GPU-hour is billed. Whether that customer's offtake contract is take-or-pay decides how much of its revenue a lender will count toward repaying the loan. The structure is long established in gas and LNG supply. It now also appears in reserved GPU capacity contracts, such as CoreWeave's committed contracts and the NexGen Cloud deal below.
For an operator, the contract is a financing input as much as a sales document. This article explains the payment mechanics, take-or-pay vs. take-and-pay, enforceability, how the clause shapes a GPU loan and its limits.
What a Take-or-pay Contract Is and How It Works
A take-or-pay contract obliges the buyer to take and pay for a minimum quantity each period, or pay the agreed price for the shortfall anyway. The Oxford Institute for Energy Studies (OIES) says the buyer must take or "otherwise pay an agreed price for any gas or LNG not taken."
In standard energy contract practice, the buyer has an alternative obligation, so paying for capacity it didn't use is a permitted way to perform the contract. Akin Gump calls the buyer's non-offtake "an exercise of a contractual right," with the payment widely understood as a debt owed to the seller. OIES sets out the resulting risk split: "the seller assumes the price risk while the buyer assumes the volume risk." In GPU terms, the operator fixes a price for its capacity, and the customer carries the risk of leaving that capacity idle.
The minimum commitment, the shortfall payment and make-up rights
OIES reports that gas take-or-pay levels often sit at 70% to 95% of the annual contract quantity (November 2022). Levels are highest in new projects that need financing. OIES notes that minimums are tied to a timeframe, such as "annual, monthly, quarterly, or daily," and in LNG can be set cargo by cargo.
Energy contracts, particularly LNG and gas contracts, often add make-up rights, which let the buyer later take volumes it has already paid for. Herbert Smith Freehills Kramer says LNG buyers that pay for untaken volumes usually have this entitlement, though it has seen contracts without one. OIES notes that the obligation may be excused for volumes the seller didn't make available, volumes rejected as off-specification and volumes force majeure kept the buyer from taking.
Take a gas buyer with an annual contract quantity of 100 units and a 90% take-or-pay level. If it takes 80 units, it pays for 90, and the 10-unit shortfall may be recoverable later as make-up gas.
In compute, the quantity is reserved GPUs or GPU-hours for the term, and the period is commonly a month. A customer that reserves 512 GPUs for 36 months owes the monthly charge for all 512, whether its jobs fill them or not. Whether any make-up right applies is a point to settle in the contract itself.
Take-or-pay vs. take-and-pay
In energy-law commentary, a take-and-pay contract obliges the buyer to both take and pay for the minimum. Norton Rose Fulbright says failure to take delivery entitles the seller to remedies for breach, and recovery will likely require showing efforts to mitigate losses. On-demand GPU revenue bills only for the hours a customer actually uses.
| Take-or-pay | Take-and-pay (energy-law sense) | On-demand GPU revenue | |
|---|---|---|---|
| Payment obligation | Minimum quantity, taken or not | Minimum quantity, taken and paid for | Hours actually used |
| If the buyer doesn't take | Pays the shortfall at the contract price | Breaches the contract | Pays nothing for unused hours |
| Who carries volume risk | Buyer | Buyer, through damages exposure | Operator |
| Seller's remedy | Claim for the agreed shortfall payment | Claims damages and will likely need to show it mitigated its loss | None |
| Lender view | Revenue sized on the customer's credit | Recovery depends on proving loss | Priced for a shorter commitment |
For most GPU operators, the practical choice sits between the first and third columns. A lender reads take-or-pay revenue as contracted cash flow and on-demand revenue as a utilization forecast.
Why Lenders Treat Take-or-pay Revenue as Financeable
OIES names long-term security of offtake as a key priority for sellers and lenders. Herbert Smith Freehills Kramer describes take-or-pay clauses as essential to finance large greenfield LNG developments, because they secure a level of revenue for the contract's duration. The contract converts customer demand into a payment obligation. The lender's main question then becomes whether the customer can pay, ahead of how busy the cluster will be.
A GPU loan has the same shape as a gas project loan. Principal and interest fall due every month whatever the cluster's utilization, and a take-or-pay charge falls due on the same monthly basis.
CoreWeave states that its committed contracts "are take-or-pay" and that they made up 98% of its revenue in the quarter ended June 30, 2026. It also reported $103.7 billion of remaining performance obligations (RPO), contracted revenue not yet recognized, as of June 30, 2026. CoreWeave expects 41% of that RPO to be recognized within 24 months.
CoreWeave reported that its $8.5 billion delayed-draw term loan facility, DDTL 4.0, closed in March 2026 and was rated A3 by Moody's. The company called it the first investment-grade financing secured by HPC infrastructure and an associated customer contract, with a fixed-rate tranche at about 5.9%. Ratings weigh many factors, so the customer contract is one input among several.
How take-or-pay changes a GPU loan
Compute IQ scores every financing through USD.AI on the operator, the customer and the contract. Customer credit and contract terms carry most of the review, because contracted revenue repays the loan, and a fixed, take-or-pay payment structure is one of the contract terms the Compute IQ scoring rubric weighs.
Better customer credit and stronger contract terms place a deal in a lower pricing tier. On-demand revenue still qualifies, priced to reflect its shorter commitment.
In an illustrative example, a $40 million deployment with GPU financing through USD.AI at 80% loan-to-value gives a $32 million loan and $8 million of operator equity. On a 36-month term with straight-line monthly amortization, principal is about $889,000 a month before interest, and a three-month debt-service reserve is funded at closing.
Assume the customer reserves the full cluster for $1.5 million a month, and that on-demand customers would pay the same hourly rate. In a month when the customer uses 70% of the cluster, the two revenue models diverge:
| Illustrative month at 70% usage | Take-or-pay contract | On-demand revenue |
|---|---|---|
| Capacity billed | 100% of the cluster | 70% of the cluster |
| Revenue | $1.5 million | $1.05 million |
| Principal due | About $889,000 | About $889,000 |
| Left for interest and operating costs | About $611,000 | About $161,000 |
Under take-or-pay, debt service is unaffected by the slow month. On-demand revenue leaves a thin margin for interest and operating costs, and the reserve cushions a run of weak months without replacing lost revenue. That gap is why the contract weighs so heavily in loan size and pricing.
A middle-market example
NexGen Cloud's $106M facility, a four-year senior secured term loan financed through USD.AI in October 2026, shows the structure at Compute Middle Market scale. It finances 216 NVIDIA HGX B300 servers, 1,728 GPUs, in a Tier III data center in Finland. The capacity is fully contracted under a take-or-pay agreement with a leading AI inference platform, and the loan amortizes over the same term.
Matching the contract term to amortization means every principal payment falls due while the customer's commitment is still running. A month of light usage lands on the customer's bill and leaves the operator's debt service intact. The facility was arranged through the GLX institutional marketplace, where institutions can take a position at origination.
Is a Take-or-pay Contract Enforceable?
Whether a take-or-pay clause holds up depends on the governing law and how it is drafted, so involve counsel on both. Mayer Brown reports that in M&J Polymers v Imerys (2008), the English High Court tested a take-or-pay clause against the penalty rule. The court held the freely negotiated, commercially justified clause wasn't a penalty on its facts. The UK Supreme Court reset the general penalty test in its 2015 Cavendish decision, which wasn't a take-or-pay case.
What Take-or-pay Doesn't Protect Against
Take-or-pay moves volume risk to the customer and leaves credit and delivery risk in place. The payment is only as good as the customer making it. The obligation also depends on the operator delivering available capacity, and outages or missed delivery dates reduce the minimum the customer owes.
Market conditions decide how readily customers accept these terms. CoreWeave's own filing warns that customers and the industry may not continue to support a take-or-pay model. In SemiAnalysis's October 2026 weekly episode, Jordan Nanos said one-year H100 deals are "approaching the prices that they were signed in the first half of 2023 today, like three bucks an hour."
In a tight market like that, customers accept commitments to secure capacity. When supply loosens, we expect them to push for shorter terms and make-up rights.
For the customer, a commitment buys capacity and a fixed price, and the charge stays owed even if its workloads shrink. Contract terms also affect what the provider pays for financing, and that cost feeds into the price of compute. In our view, a firm commitment from a creditworthy buyer gives its provider cheaper capital to build with.
Conclusion
A take-or-pay contract is a payment obligation that turns reserved GPU capacity into revenue a lender can count, with the customer carrying the volume risk. Its value still rests on the customer's credit, the operator's delivery and a clause drafted to hold up under its governing law.
Financing through USD.AI is underwritten on contracted cash flow and secured by the GPUs. Compute IQ scores the operator, the customer and the contract, with the customer and contract carrying most of the review. A creditworthy customer on a take-or-pay offtake contract therefore shows up directly in a deal's score, pricing tier and loan size.
If you have a signed or near-signed offtake contract for a new neocloud deployment, bring it to the first conversation. Talk to USD.AI to see how the contract shapes your financing.
FAQ
What is a take-or-pay contract in simple terms?
A take-or-pay contract is an agreement in which the customer commits to buy a minimum amount and pays for it even if it uses less. In GPU compute, the minimum is usually a block of reserved GPUs billed monthly for the contract term.
What is the difference between take-or-pay and take-and-pay?
Under take-or-pay, not taking the minimum is an allowed option, and the buyer simply pays for the shortfall. Under take-and-pay in its energy-law sense, failing to take is a breach, and the seller claims damages and will likely need to show it mitigated its loss.
What are make-up rights in a take-or-pay contract?
Make-up rights let the buyer take, in a later period, volumes it has already paid for under a shortfall payment. They are usually included in LNG contracts but aren't universal; some contracts have no such entitlement.
Why do lenders prefer take-or-pay contracts?
Take-or-pay revenue doesn't depend on utilization, so monthly payments line up with monthly debt service. That lets the loan rest on the customer's credit and the contract's terms.
Does on-demand GPU revenue qualify for financing through USD.AI?
Yes, documented on-demand revenue qualifies for financing through USD.AI. It is priced to reflect its shorter commitment, and better customer credit and stronger contract terms place a deal in a lower pricing tier.
