8 things GPU rental prices tell you about the AI cycle (H100, H200, B200)
If you want to know whether AI compute is scarce, look at what it costs to rent. Our GPU prices page tracks live prices for ten GPUs and history back to 2023. Here is how to read it.
1Rising rents mean demand is outrunning supply
When rental prices go up even as new chips ship, buyers are still short of capacity.
2Old-chip prices matter most
If H100 and A100 rents hold up while B200s ramp, GPUs keep earning for longer, which supports neocloud lending and hyperscaler depreciation schedules.
3Reserved vs on-demand shows urgency
When 12-month reserved prices climb towards on-demand prices, buyers are paying up to lock in capacity.
4Spot prices are the first to crack
Spot and marketplace prices move daily. A falling spot price with steady list prices is the earliest sign of spare capacity.
5Few GPUs listed for rent means tight supply
We count how many GPUs are free to rent on open marketplaces each day. Thin listings for H100 and B200 point to a tight market.
6The H100's long decline has flattened
H100 on-demand prices fell from 2023 highs as supply caught up, then steadied. Watch whether they turn back up.
7New chips price at a premium
B200 and B300 rent for roughly twice an H100. How fast that premium shrinks shows how quickly new supply arrives.
8A composite index summarises it
A daily compute price index for a basket of data-center GPUs shows the overall direction in one number.
Who should watch GPU prices
Anyone who owns neoclouds such as CoreWeave, Nebius or IREN, lenders and investors exposed to GPU-backed debt, and NVIDIA and AMD shareholders. Rental prices drive neocloud revenue directly and decide how long a GPU keeps its value.
How to read the charts together
Start with the live table for today's prices. Then check the long history to see whether prices are trending up or down, the market-wide median to make sure one provider isn't skewing the picture, and the reserved-versus-on-demand chart to see whether buyers are rushing to lock in capacity.
Rental prices move before earnings do. Check them before neocloud and NVIDIA reports.
Frequently asked questions
How much does an H100 cost to rent?
On-demand H100 SXM prices were around $4 per GPU-hour across providers in October 2026, with spot and marketplace prices lower. The page shows today's figure.
Where do the prices come from?
Price of Compute (live prices, history and index), GetDeploying's CC BY 4.0 dataset (weekly medians across about 88 providers), and Vast.ai and RunPod public prices.
How often are GPU prices updated?
Live prices refresh hourly; weekly medians update daily.
Get the daily brief
Everything that happened across the AI supply chain, in a 3-minute email every weekday. Free.
More guides
Figures quoted are as of the publish date; the linked pages show live data. Educational research by Singularity Research, not investment advice.