GPU prices swing wildly between reserved and spot capacity. Packaging them as credits inside UsageBox makes margins predictable and lets customers choose reliability vs. savings.
Define the Credit Types
- Reserved credits: Priority scheduling, predictable p95 latency, and higher SLAs.
- Spot credits: Discounted, preemptible runs with automatic retries.
- Hybrid blocks: Blend 80/20 reserved-to-spot for balanced cost protection.
Pricing Levers
- Prepay discounts: Annual reserved blocks with upfront discounts.
- Burst pools: Temporary spot packs for launches and marketing campaigns.
- Auto-upgrades: When spot evictions exceed thresholds, upgrade to reserved and note credits on invoices.
Surface the Experience
Expose reserved vs. spot consumption in FinOps dashboards and customer portals. Show customers how many retries were absorbed and where SLAs held.
With creditized GPU capacity, both finance and engineering avoid late-night paging about runaway inference costs.