Cloud GPU Rental Price Index
We track posted cloud GPU rental prices for 17 models across 75 providers, covering on-demand, spot, and reserved rates.
Last released has the largest price increase among the three groups. Its on-demand median rose from $2.12 per GPU-hour in October 2024 to $4.72 in September 2026. Over the same period, the Modern group moved from $1.35 to $1.45, while Legacy fell from $1.77 to $0.95.
Each median uses the listings present that month. Providers, GPU variants, and available models can change. GPU groups stay fixed for comparison:
Category | GPUs | Role |
|---|---|---|
Last released (2024 and later) | B200, B300, MI300X, RTX 5090 | Newest-generation cohort |
Modern (2020 through 2023) | H100, H200, A100, L40S, RTX 4090, A10G, T4, L4 | Mainstream workload-runners |
Legacy (pre-2020) | V100, P100, K80, M60, P40 | Still rentable, mostly community-tier neoclouds |
These labels identify the index groups. T4 remains in Modern despite its earlier release date. Each GPU contributes equally to its group median, regardless of listing volume.
See our GPU index methodology for how this is computed.
Price trends by GPU model
H100’s September on-demand median is $3.25 per GPU-hour, compared with $4.40 for H200 and $1.76 for A100. L40S is $1.29. RTX 4090 and RTX 5090 sit below a dollar, at $0.44 and $0.63 respectively.
IONOS prices H100 and H200 monthly instead of hourly, at a flat $3,990 per dedicated server with EU data residency. Its on-demand catalog covers T4, A10, and RTX PRO 6000 Blackwell.
B200 and B300 have medians of $6.52 and $7.87, respectively. MI300X is $2.91, below H100, while the V100 reference line is $0.95.
These prices combine physical variants sold under the same GPU model name. A100 includes both 40 GB and 80 GB cards. PCIe, SXM, and NVL listings can also differ. The cloud GPU pricing tables separate memory and interconnect variants where the source identifies them.
Hourly price alone does not measure the cost of completing a workload. The multi-GPU benchmark compares measured performance on H100, H200, B200, and MI300X.
Price trends by provider
Provider choice adds another source of variation. September’s H200 on-demand listings range from $2.09 per GPU-hour at Beam to $13.78 at Microsoft Azure. The range includes different instance configurations and stock states.
The cloud GPU provider comparison covers the companies behind these listings. A provider’s line shows its monthly median for the selected GPU and billing tier. Storage, networking, service terms, and GPU count still need checking against the actual instance.
Supply and availability
Confirmed availability is lowest for MI300X at 7% of listings, followed by B300 at 11% and B200 at 16%. H100 reaches 33%, while RTX 4090 and RTX 5090 reach 68% and 79%.
Unknown stock status accounts for 88% of MI300X listings. Its low confirmed share therefore cannot be read as 93% sold out. The denominator includes available, unknown, waitlisted, and unavailable listings, across billing tiers.
These percentages describe catalog entries, not physical GPU inventory. Marketplace listing turnover can change them even without a comparable change in installed capacity.
On-demand and reserved prices
In the September 14 weekly data, B300’s reserved median is $5.71 per GPU-hour against $7.87 on-demand. H100 has a smaller $0.13 gap between its $3.10 reserved and $3.23 on-demand medians.
MI300X reverses that relationship. Its reserved median is $3.39, above the $2.91 on-demand median. On-demand and reserved medians use separate provider pools. It does not measure the discount available from the same provider on an identical instance.
Reservation records include different commitment lengths. The plotted median combines those terms, which need checking against an individual quote.
Spot discounts
September’s spot discount is 47.7% for Modern GPUs, 47.5% for Last released, and 37.3% for Legacy. These percentages pair on-demand and spot prices from the same provider, GPU model, and month.
Pairs need both rates to enter the calculation. Differences in instance variants can remain within a GPU model, and a category with sparse spot coverage represents fewer comparisons. Spot capacity is interruptible. These rates do not include the cost of restarting work.
GPU index methodology
The analysis covers publicly listed GPU rental prices and weekly price history. The history spans July 2024 through September 2026. Additional models and providers in the source history fall outside the curated selection.
All prices are US dollars per GPU-hour. Monthly trends first take the median of weekly observations for each provider, GPU, and billing tier. The model chart then takes the median across providers. The category chart takes one more median across the GPUs in each group.
The category chart’s Average option is the arithmetic mean of its on-demand, spot, and reserved category medians. Months need all three rates to qualify. It is an analytical summary, not a purchasable rate.
The provider explorer stops before cross-provider aggregation. Lines connect observed months, including gaps. Availability uses the current offerings snapshot. Reservation prices use the latest week and require at least four providers with reservation records for a GPU.
Spot discounts use (on-demand − spot) / on-demand × 100 for each matched provider, GPU, and month. We take the median across providers, then across GPUs within each category.
Contact-sales quotes and negotiated enterprise contracts fall outside the index. It does not measure throughput, regional capacity, or total operating cost.
Add as preferred sourceFAQs
We publish a refreshed monthly median view each month. The numbers reflect data through the prior month.
The GPU is the same; the bundle is not. Hyperscalers price in compliance (HIPAA, SOC 2, FedRAMP), enterprise SLAs, identity and networking integration, and 24/7 support. Neoclouds price bare metal or VM access with optional managed orchestration. If you do not need the bundle, the Neocloud price is the right comparison.
Yes, if your workload checkpoints and tolerates 5-15 minute interruptions. Modern GPU spot discount sits near 50% over the past six months, and savings compound over multi-day training. Spot is the wrong choice for latency-sensitive inference, single-replica services without failover, or evaluation runs that need a clean wall-clock comparison.
Price trends by provider chart’s billing dropdown switches between on-demand, spot, and 1-year reserved tiers wherever providers publish those rates. Multi-year contracts and enterprise-negotiated discounts are not included. Request a quote directly from the provider for those.
Further reading
- Multi-GPU Benchmark: B200 vs H200 vs H100 vs MI300X
- Top 30 Cloud GPU Providers & Their GPUs
- GPU Concurrency Benchmark
- Top 25+ AI Chip Makers: NVIDIA & Its Competitors
Cite this research
Pick the format that matches where you're publishing. Pasting the link version into your CMS preserves the backlink.
@misc{sari2026,
author = {Sarı, Ekrem},
title = {{Cloud GPU Rental Price Index}},
year = {2026},
month = sep,
howpublished = {\url{https://aimultiple.com/gpu-index}},
note = {AIMultiple. Retrieved September 22, 2026}
}Results and timestamps of 3 data points. Download the summary data shown in this article's charts and tables as a ZIP file containing one CSV file.
Want the granular data behind it? Join Premium
Changelog
3 updatesUpdated the GPU index across 63 providers, with new medians, supply-confirmation rates, and reserved discounts, and IONOS added.
Replaced the medians and ranges in the Modern and Last released GPU sections with new values.
Added an "Average" billing tier option to the market summary chart in the methodology section.
Be the first to comment
Your email address will not be published. All fields are required. Comments are left in their original language.