Reserved Instances
Reserved instances are GPU compute resources purchased via a time-bound commitment — typically 1, 6, 12, or 36 months — at a significant discount to on-demand pricing in exchange for guaranteed payment. Reserved pricing for B200-class GPUs ranges from $2.25/hr for 36-month commitments to $4.99/hr on-demand, representing discounts of 25-55% depending on provider and term length. CoreWeave publishes "up to 60% off on-demand for committed usage." Reservations provide revenue predictability for operators and cost savings for customers with sustained workloads.
Reservation contracts vary significantly in structure: some require upfront payment, others charge monthly; some guarantee specific GPU models, others offer flexibility across generations; some include SLA commitments, others are best-effort. The contract terms — duration, pricing mechanisms, indexation clauses, termination rights, and covenants — are critical variables in both customer procurement and operator revenue modelling.
Reserved instance pricing and discount structures are central to our due diligence work. Management teams frequently understate real-world discount concessions in their business plans. Our benchmark data consistently shows market discounts 2-3x deeper than management assumptions, which materially affects revenue projections.
This glossary is maintained by Disintermediate as a reference for GPU infrastructure professionals, investors, and operators. Each entry reflects terminology as used in active advisory engagements and market intelligence work.