How many providers rent each GPU
The count of providers with a current on-demand rate for each GPU, card by card, and what a short list of sellers means for a provider entering it.
The seller count, card by card
We hold a current on-demand GPU rate from 48 providers, covering 51 cards and 464 rate observations. Counted card by card, the number of providers with a current rate is: H100 SXM 80GB 13, H100 PCIe 7, H200 26, A100 SXM 80GB 9, L40S 20, RTX 4090 14, B200 22 and MI300X 4.
The remaining counts are: L4 7, A10 5, A40 3, T4 4, V100 4, GH200 4, RTX 6000 Ada 10 and RTX PRO 6000 20. Every count, and the rates behind it, is on the GPU price list.
Read the numbers rather than this paragraph for which card is most widely rented. The order changes as sellers list and drop cards, and a sentence ranking them would go stale the week it was written.
Some cards are missing from both paragraphs. The A100 with no form factor named, the L40, the RTX 4090's newer and older siblings and the Blackwell parts each have their own count on their card page, and the A100 post explains why one card can have more than one row.
What a seller count measures
A seller is counted for a card when its own pricing page carries an on-demand rate for that exact card and our last good read of the page is within 21 days. One seller counts once per card, however many regions or instance shapes it lists. Reserved and interruptible rates do not make a seller count.
So the count measures public on-demand listing, and nothing else: not capacity, not sales, not the size of any seller. A provider that sells a card only on contract, or only behind a sign-in, is not in it. A provider whose page could not be read for more than 21 days drops out until the next good read.
How the data is gathered sets out the reads. The count is the denominator behind every range on the card pages, which is why it is the first figure to look at.
It is also the figure most easily moved by our own reads. If a provider's page changes layout and cannot be read, its rates leave every row at once, and several counts fall by one on the same day. A drop across many cards together is more likely to be a page that could not be read than sellers leaving the market.
What a short list means for a seller entering
On a card with few sellers, a new listing becomes part of the range the day it is read. If it is the lowest, it is the figure buyers see first; if it is the highest, it sets the top. Either way the newcomer is visible in a way it would not be on a card with many sellers.
The same short list makes each listing easy to misread. A single per-node price taken as per-GPU, or an SXM card filed under a PCIe name, can be the whole range for a while. A seller entering a thin row should check its page for card name, unit and commitment before it checks its rate against anyone else.
A short list is not evidence of low demand. It may mean the card is new, that supply is held in contracts, or that sellers list it only on request. The count cannot tell those apart.
For a seller deciding whether to list a card publicly at all, the count is one input among several. A short row means a public rate will be seen; it does not mean buyers are waiting for it. Only your own enquiries can say that, and nothing on this site measures them for a card you do not yet list.
What a long list means
On a card with many sellers, one listing moves the range only if it lands at an end. The spread figure, highest rate over lowest, is shown on each card page where there are at least a pair of sellers, and across the 29 cards compared the median spread is 3.7×. The widest is the RTX-A4000-16GB, at 10.4×.
For a seller in a long row the question changes from being seen to being compared. Buyers line the rates up, and the GPU price index shows how cards sit against each other. What decides a sale after that is on the seller's own page: what the hour includes, the terms, and whether the rate is the one the buyer will pay.
A seller in a long row can also check its position over time. The card page shows the date each rate was read, so a seller that changes its price can see when the change was picked up and where it then sits, without asking anyone.
Reading the counts with care
The counts on this page fill from the data each time it loads, and they move. Each earlier post in this series reads a single card in depth, for example the H100 SXM post, and the card page is always the current view.
A count is also a statement about pages we read, not about every provider that exists. A provider with no page we read, or one we have not yet added, is not counted. A seller missing from a card it sells can claim its listing, which is free, and correct the record. The count then changes on the next read, and so does every range that depends on it.
What we could not answer
- ⚑The counts cover public on-demand listings only, so contract-only supply is invisible here.
- ⚑The post does not rank cards by seller count, because the order changes as listings change.
infratoapp, “How many providers rent each GPU”, read 27 October 2026. providers with a current GPU rate: 48. https://www.infratoapp.com/blog/how-many-providers-rent-each-gpu
Questions people ask about this
Which GPUs are rented by the most cloud providers?
The count of providers with a current on-demand rate is held card by card and changes as sellers list and drop cards. For the H100 SXM 80GB it is 13, and for the A100 SXM 80GB 9. The full list of counts is on the GPU price list.
How many GPU cloud providers have a published price?
We hold a current on-demand GPU rate from 48 providers, covering 51 cards. A provider is counted when its own pricing page carries an on-demand rate and the last good read is within 21 days. Contract-only and sign-in prices are not counted.
Does a card with few sellers mean low demand?
Not necessarily. Few public sellers can mean the card is new, that most supply is sold on contract, or that sellers quote it only on request. The seller count measures public on-demand listing and cannot tell those reasons apart.
These render as FAQPage structured data · answers are filled from the same live figures
Read next
Where H100 SXM 80GB on-demand rates sit across providers, read from each seller's page, and what a seller below or above the range should check.
L40S 48GB on-demand rates across sellers, set beside the L40 that shares its memory size, and the naming checks an L40S seller should run first.
The lowest and highest listed H200 141GB on-demand rates, how they compare with the H100, and the memory case a seller can make for the card.