GPU Offtake Agreements: Underwriting the Cash Flow
By Commertize Desk · Published 5 November 2026
Take a hypothetical 1,000-GPU cluster sold forward under a three-year offtake agreement at a fixed price per GPU-hour. On paper it is a contracted cash flow: a creditworthy counterparty, a fixed rate, a known term. The numbers below are illustrative, not a quote, but the structure is the one lenders and allocators are now being asked to underwrite.
The contract on paper
Assume the cluster costs $40 million installed, with the buyer committing to 85% of available hours at a fixed rate for 36 months. Contracted revenue is visible from day one, which is why compute capacity increasingly gets compared to a net lease. A lender sees a counterparty, a term and a payment stream.
Three things sit underneath that picture, and each can break it: the useful life of the hardware, the hours the buyer actually takes, and the electricity the building can deliver.
Stress one: the hardware outlives the contract, or does not
A net lease sits on land and a building that last decades. A GPU cluster is a depreciating machine. If a newer generation arrives during the term, the market rate for the old hardware can fall below the contracted rate. That helps the owner while the buyer stays in the contract, and hurts the moment the buyer's credit weakens or the contract is renegotiated.
The underwriting question is therefore residual value at term end. In our example, if the cluster is fully depreciated over its 36-month term, the residual is the reserve against reuse, resale or re-contracting. A structure that assumes a flat re-lease rate at maturity has assumed away the main risk.
Stress two: take-or-pay versus take-if-needed
Utilization is where an offtake agreement differs most from a lease. A take-or-pay clause obliges the buyer to pay for committed hours whether or not it uses them. A capacity reservation with usage flexibility does not. In our example, if the buyer's real consumption is 60% instead of 85%, a take-or-pay contract still pays in full, while a flexible one cuts revenue by nearly 30%. The cash flow is only as firm as the weakest clause governing hours.
Underwriters should also read the termination rights: what the buyer owes on early exit, and whether performance credits for downtime can erode the contracted payment.
Stress three: power, not silicon
The cluster earns nothing without delivered power. The Lawrence Berkeley National Laboratory's Queued Up report, 2024 edition, found roughly 2,600 GW of generation and storage capacity waiting in US interconnection queues at the end of 2023, with multi-year waits. The same lab's United States Data Center Energy Usage Report of December 2024 estimated that data centers used about 4.4% of US electricity in 2023 and could reach 6.7% to 12% by 2028. The International Energy Agency's Energy and AI analysis of April 2025 put global data center demand at around 415 TWh in 2024, rising toward 945 TWh by 2030.
For an offtake underwriter, a power agreement with energization dates is as important as the compute contract. Our full treatment of how power sits in the deal is in the pillar on tokenization for energy and digital infrastructure.
What the checklist looks like
An allocator reading a GPU offtake agreement should be able to answer, from documents and data rather than from a sales deck:
- Counterparty: who pays, what is its credit, and is the obligation take-or-pay?
- Hardware: which generation, what warranty, and what is the assumed residual?
- Utilization: committed hours, measured hours and who bears the gap?
- Power: signed interconnection, energization date, and curtailment rights.
- Metering: is usage recorded in a way a third party can verify?
That last item is where on-chain structures are relevant. Metered, timestamped usage is a verifiable data stream, and cash-flow assets with verifiable data can be fractionalized at lower minimums than a $40 million cluster allows. The how it works overview describes how this connects to digital capital markets. The contract still has to be good before the wrapper matters.
Source: https://www.iea.org/reports/energy-and-ai (published April 2025)
What to do next
If you have digital infrastructure you are considering bringing to market, start at Commertize's energy and digital infrastructure page.
Prefer email? Send a short asset summary to deals@commertize.com. We reply within one business day, with no obligation to engage us.
Asset class: GPU Compute