FIELD GUIDE /
GPU Rental Pricing: Hourly Rates and Total Cost
Understand GPU rental prices, per-GPU versus whole-plan billing, regional differences and the full cost of an AI workload. Compare dated public listings.
How much does it cost to rent a GPU?
A GPU rental price pays for access to a configured cloud machine for a period of time. The useful comparison is the cost of completing your workload, including the whole plan, runtime and additional fees. Chip model alone does not identify an equivalent service.
SiliconTicker compares public listings in USD per GPU-hour and retains the original plan price. A lower per-GPU rate can still require a more expensive multi-GPU plan. Prices are advertised observations; they do not confirm available capacity or a negotiated quote.
Observed public GPU rental rates
These ranges cover the standard public-list and on-demand offers in our comparison catalog. Spot offers are excluded. Each row shows its underlying observation dates and sample size; this is a covered sample, not the entire rental market.
| GPU | Advertised range | Coverage | Observed / status |
|---|---|---|---|
| RTX 4000 Ada | $0.76–$0.76 | 1 offer / 1 provider | 2026-10-20 · fresh |
| RTX 6000 Ada | $1.57–$1.57 | 1 offer / 1 provider | 2026-10-20 · fresh |
| L40S | $1.57–$1.67 | 2 offers / 2 providers | 2026-10-20 · fresh |
| MI300X | $2.59–$2.59 | 2 offers / 1 provider | 2026-10-20 · fresh |
| H100 | $3.37–$4.41 | 4 offers / 3 providers | 2026-10-20 · fresh |
| MI325X | $3.80–$3.80 | 2 offers / 1 provider | 2026-10-20 · fresh |
| H200 | $4.47–$4.47 | 2 offers / 1 provider | 2026-10-20 · fresh |
| RTX PRO 6000 | $6.00–$8.25 | 4 offers / 1 provider | 2026-10-20 · fresh |
| B300 | $16.00–$16.00 | 1 offer / 1 provider | 2026-10-20 · fresh |
Data: GPU Cloud Compare, CC-BY-4.0, normalized under our methodology. View exact plans, locations and upstream provider links in the GPU comparison. Read structured rows at /api/compare.
Calculate the whole-plan cost
Compute cost = number of rented plans × plan price per hour × billed hours. The number of plans is your requested GPU count divided by GPUs per plan, rounded up. Keep the per-GPU unit separate from the billable plan unit.
Suppose a four-GPU plan costs $12 per hour. Its normalized rate is $3 per GPU-hour. A job needing five GPUs requires two plans: eight billed GPUs, $24 per hour and $240 for ten hours. Multiplying five by $3 would understate the bill.
Add storage, data transfer, software, support and taxes where applicable. Include idle allocation and restart time. Check the provider’s minimum billing interval and whether a stopped machine remains billable. For example, DigitalOcean documents that GPU Droplet billing continues while powered off and ends when the Droplet is destroyed. DigitalOcean billing details.
Compare reserved blocks by useful work
A reserved block has different billing terms from an hourly rental. AWS charges the EC2 Capacity Block reservation price upfront and fixes it after reservation. Operating-system charges can apply separately while instances run; Savings Plans and Reserved Instance discounts do not apply to Capacity Blocks. AWS pricing and billing documentation.
For your own cost comparison, divide the upfront reservation price by the GPU-hours your workload will actually use, then add applicable fees. Low utilization raises that effective cost even when the advertised hourly equivalent looks attractive. Compare that result with the whole-plan hourly estimate above, using the same GPU count and useful workload. Do not assume a Capacity Block has the billing or discount rules of an ordinary Capacity Reservation.
Compare the configuration before the rate
- Hardware: exact GPU variant, memory per GPU, dedicated versus shared access and interconnect.
- Plan: GPU count, CPU, RAM, storage and network capacity. A plan total is not a single-GPU quote.
- Location: region, data residency, transfer cost and distance to your data.
- Terms: interruption, commitment length, minimum spend, support and capacity confirmation.
- Performance: measure runtime or throughput with your model, precision, batch size and software stack.
A GPU that costs twice as much per hour must complete comparable useful work more than twice as fast to lower compute cost, before other costs. Compare H100 and A100 by cost per result and consider interruption risk for Spot offers.
Common pricing questions
Is the cheapest listing the cheapest workload?
Only if it can run the same workload with comparable output quality, runtime and service requirements. A smaller memory allocation, slower storage or extra restarts can erase a lower hourly rate.
Does a listed price guarantee a GPU is available?
No. Confirm region, quota, delivery time and allocation with the provider. SiliconTicker marks comparison availability as unconfirmed.
Are rental rates a futures settlement benchmark?
No. Public listings, the CL-FCR reference index and a venue’s settlement benchmark measure different things. See the current compute futures status.
Primary sources
Technical and billing details were reviewed on . Provider terms and capacity can change; price tables show their own observation dates.
- GPU Cloud Compare: licensed public price dataset ↗
- DigitalOcean: GPU Droplet pricing and billing ↗
- Microsoft: Spot VM pricing and capacity ↗
- AWS: Capacity Blocks pricing and billing ↗