GPU Price Trends
GPU prices rose +3.3% over the last 4 weeks.
Week of
- GPU price index
- 99.7
- Equal-weight · 27 models · Jan '25 = 100
- Past year
- +5%
- Sep '25 → Sep '26
- Past 4 weeks
- +3.3%
- Aug 17 → Sep 14
- Rose this week
- 7 /25 models
- 18 flat or down
Heads up: The index is a level, not a price: it tracks how a fixed panel of GPU models moves against its base week, and the cheapest listing on any given day sits well below it. Our data is informational only and not intended for reference in financial instruments or contracts. More on how we price.
How the index is built
Every week we record the on-demand price of every GPU listing we track, in US dollars per GPU per hour: an 8×H100 instance at $40/hr counts as $5.00. The data is our own: 51,031 weekly observations since July 2024, across 82 providers and 103 GPU models, scraped as often as every 15 minutes.
The index is the median of each model's price relative to its own price in the base week, across a fixed panel with every model weighted equally. Jan. 6, 2025 = 100. It is a median, not a floor: the lowest listing on the market any given day sits below it. For that price, see the cheapest cloud GPUs guide.
- A model's price is each provider's own median first, then the median across providers. Listing one card at eight instance sizes buys one vote, not eight. To see what that rate buys in memory and bandwidth, see best value GPUs.
- The panel is reviewed each January and July. A model joins after about three months of consistent pricing (13 qualifying weeks) and leaves after a sustained gap. Changes are linked so the level runs continuously; newer models appear in the per-model charts meanwhile.
- Left out: spot and reservation pricing, weeks where we captured only part of the market, and any model priced by fewer than four providers that week.
- The band is the middle 60% of panel models. When it widens, models are diverging rather than moving together.
- Early history is approximate. Coverage grew from roughly 30 providers in early 2025 to around 60 by February 2026. Where a correction to our own data moved the index rather than the market, that step is spliced out and the moves either side are left alone.
The panel: AMD MI300X, Nvidia A10, Nvidia A100, Nvidia A40, Nvidia A4000, Nvidia A5000, Nvidia RTX A6000, Nvidia B200, Nvidia B300, Nvidia H100, Nvidia H200, Nvidia L4, Nvidia L40, Nvidia L40S, Nvidia P100, Nvidia RTX 3070, Nvidia RTX 3080 Ti, Nvidia RTX 3090, Nvidia RTX 4000 Ada, Nvidia RTX 4080, Nvidia RTX 4090, Nvidia RTX 5090, Nvidia Quadro RTX 6000, Nvidia RTX 6000 Ada, Nvidia RTX PRO 6000, Nvidia T4, Nvidia V100. The panel has changed at January 2026 and July 2026. Departed: Nvidia GH200 (July 2026 review).
Accruing history: Nvidia RTX 5080 (qualifying now); Nvidia RTX PRO 4500 (7 of 13 qualifying weeks); Nvidia RTX 3080 (qualifying now). Each can join at the January 2027 review if it keeps being priced.
Common questions
What is the GPU Rental Price Index?
A weekly index of cloud GPU rental prices. Each model's median on-demand price across providers, in USD per GPU per hour, is tracked against its own price in the base week (January 2025 = 100). The panel holds 27 models, each weighted equally, and the published level is the median across them.
Are cloud GPU prices going up or down?
On-demand GPU rental prices are up about 5% over the last 12 months, with a 3.3% move up over the last 4 weeks. The index tracks the median price across providers, weighting each provider equally. It moves both when providers change prices and when cheaper or pricier offerings enter or leave the market.
How much more do hyperscalers charge for GPU rentals?
For an Nvidia H100, hyperscalers (AWS, Microsoft Azure, Google Cloud) currently charge about 100% more than dedicated GPU clouds. Median on-demand prices in USD per GPU per hour:
GPU GPU clouds Hyperscalers Difference Nvidia H100 $3.94 $7.89 +100% Nvidia A100 $2.00 $3.67 +84% Nvidia H200 $4.50 $10.62 +136% Nvidia B200 $7.88 $15.26 +94% Nvidia RTX PRO 6000 $2.15 $4.50 +109% How much cheaper are reserved GPU contracts?
A 1-year reservation currently costs about 31% less than on-demand, and a 3-year reservation about 50% less. Median discounts, comparing each provider's reserved price against its own on-demand price for the same GPU:
Commitment vs on-demand 1 mo -7% 3 mo -3% 6 mo -15% 1 yr -31% 2 yrs -25% 3 yrs -50% How much cheaper are spot GPUs than on-demand?
Spot GPU instances currently list at about 54% of the same models' on-demand price, a discount of roughly half. This is the median across the 16 GPU models priced both ways broadly enough to compare, with each provider's spot price measured against its own on-demand price. Spot capacity can be reclaimed by the provider at short notice, so the discount pays for interruption risk.
The discount varies by model. Spot price as a share of on-demand:
GPU Spot vs on-demand Nvidia H100 53% Nvidia RTX PRO 6000 52% Nvidia A100 55% Nvidia H200 55% Nvidia V100 39% Which GPU prices moved the most recently?
The biggest price moves in the week of September 14, 2026. Only models listed by at least 3 providers are included. Median on-demand prices in USD per GPU per hour:
GPU Week before Latest Change Nvidia A40 $0.49 $0.77 +57% Nvidia RTX 2070 $0.06 $0.08 +33% Nvidia GTX 1070 $0.04 $0.05 +25% Nvidia A5000 $1.24 $0.79 -36% Nvidia RTX 5070 $0.24 $0.17 -29% Nvidia RTX 4070 $0.20 $0.15 -25%