Capacity Reserve Agreement
Forward-purchase of compute or inference capacity
A customer or AI lab commits to a multi-year forward purchase of compute capacity (training tokens, inference throughput, reserved GPU hours) from a hyperscaler or specialized provider, often in exchange for price guarantees, priority access during scarcity events, and binding model-availability commitments. The pattern shows up two ways: (1) an enterprise reserves AI lab inference capacity for product launches (Salesforce/Anthropic, Apple/Anthropic, Snowflake/Mistral), and (2) an AI lab reserves compute capacity from a hyperscaler tied to a parallel strategic investment (the compute-backed-strategic-investment pattern's capacity leg). Distinguishing feature: the contract's economic engine is a take-or-pay forward purchase, not a one-shot licensing fee.
Hypothetical, illustrative. Not actual deal terms. Practitioners should not use these clauses verbatim; they illustrate structure and what to negotiate.
The intuitive moves that alliance research has documented as predictably failing for this pattern. Each one comes with a mitigation that addresses the underlying mechanism, not just the symptom.
- 1.
- 2.
- 3.
- 4.
- 5.
No partnerships tagged with this pattern yet. As the corpus matures, public-source partnerships that exhibit this pattern's structural moves will appear here, each scored on which kill-list moves are visible in their public terms.
Litigation and regulatory actions a thoughtful counsel reasons against when negotiating a deal in this structural shape. Cited for orientation, not for legal advice.
The primary-source research this pattern is grounded in.