What is your GPU spend actually buying?
On-demand GPUs cost more than the sticker rate: shared pools shave throughput, interruptions eat recovery hours, and the meter premium compounds the more you run them. Reserved capacity wastes money the other way — idle hours you pay for anyway. Enter your setup to see your true cost per useful GPU-hour — and what dedicated capacity would save.
your current setup
what it really costs today
Sticker rate $2.99, but after the shared-pool throughput haircut and 4 interruptions a month, every useful GPU-hour costs $3.59 — a 20% premium you never see on the invoice.
what dedicated with Nestor would cost (estimated)
Estimated committed rate: $2.99/GPU-hr. Scenarios are anchored to published reserved-discount bands, not our pricing.
Utilization on dedicated, right-sized to your workload: 95% (nothing runs at 100 — we plan for that)
Estimate only. At your provider's rate, a 35% discount would land below what dedicated capacity realistically clears, so the estimate is held at $2.99. The saving above comes from full throughput and zero preemption at that rate, not from an unrealistic sticker. Actual pricing and sizing are confirmed during scoping.
your rate vs the market vs Nestor (estimated)
Published on-demand rates per GPU-hour from official pricing pages, as of July 19, 2026. Hyperscaler rates are per-8-GPU instance divided by 8 (list, US regions). Your effective rate includes the drag you entered above; the Nestor bar reflects the estimate you set.
full breakdown
| horizon | your spend | with Nestor (est.) | you save |
|---|---|---|---|
| 3 months | $28,704 | $25,169 | $3,535 |
| 6 months | $57,408 | $50,338 | $7,070 |
| 9 months | $86,112 | $75,507 | $10,605 |
| 12 months | $114,816 | $100,676 | $14,140 |
committed rates elsewhere (reference)
| program | $/GPU-hr |
|---|---|
| Together AI (91–180 day commit) | $3.09 |
| AWS Capacity Blocks | $4.33 |
| Azure 3-yr reservation | $5.39 |
| Azure 1-yr reservation | $7.87 |
| GCP 3-yr CUD (approx.) | $4.86 |
Published committed/reserved programs, for context. A Nestor dedicated environment runs single-tenant hardware at full throughput with no preemption, right-sized to your workload, with orchestration, monitoring, and operator support included. Actual pricing and sizing are scoped with your team — the figures above are directional, not a quote.
See the full breakdown
Your 12-month cost vs a right-sized Nestor deployment, the estimated monthly and annual savings, and committed-rate benchmarks for this workload.