What buyers compare before they rent a GPU
Buyers start with the hourly rate and finish with availability, network and terms, and that second list is where a provider keeps or loses the sale.
The hourly rate is where the search starts
A buyer shopping for a GPU compares the hourly rate first, because it is the only figure every provider publishes in the same unit. We hold a current rate from 48 providers, covering 51 cards with an on-demand price, and a buyer with a spreadsheet and an afternoon will line them up in that order.
The rate does the first cut and nothing more. On the H100 SXM 80GB the lowest listed on-demand rate is $1.68 an hour at Vast.ai and the highest is $6.16, across 13 sellers. Nobody rents from the dearest seller because they misread the table, so something other than the rate is carrying that sale. Everything below is a guess at what.
The same card appears on its own page with every seller, the rate and the page it was read from, which is the view a buyer is comparing against when they open your pricing page in another tab.
What buyers check once the rate looks right
Availability comes next, and it is the question a price page cannot answer. A buyer who needs a full node for a training run wants to know whether the whole node is free today, not whether one card is cheap. A listed rate on a card that is never in stock is a rate nobody can pay.
After that comes the network and the storage attached to the machine. For multi-node work the interconnect decides whether the job scales, and for any work the question is where the data lives and what it costs to move it out. Buyers who have been burned once ask about egress before they ask about the card.
Terms come last and matter most to the people who sign. Minimum commitment, notice to cancel, what happens to an interruptible instance when it is reclaimed, and how quickly a person answers a support ticket are all things a buyer learns after the first invoice unless the page says them first.
What a per-hour rate hides
A rate is a price per unit of time on one card, and a bill is something else. Idle time while a job waits in a queue is billed. Storage attached to a stopped instance is billed. Billing increments differ: a provider that rounds up to the hour charges for a hour on a ten-minute test.
The figure that makes this concrete is the month. The lowest listed H100 SXM rate, run around the clock for a month of 730 hours, comes to $1,226.40. A buyer working that out will notice that a small difference in the hourly rate is a large difference in the invoice, and will go back and read the terms they skipped.
The cost calculator does this arithmetic for any card and any number of hours, using only the rates providers published. It adds nothing for what the rate leaves out, and it says so beside the result.
How a buyer reads your pricing page
They look for the unit first: per GPU, per node, or per instance with some CPU and memory bundled in. A price quoted per node next to a price quoted per card cannot be compared, and a buyer who cannot compare will usually move on rather than do the division.
They look for the commitment next. On-demand, reserved for a term, and interruptible are three different products, and a table that mixes them under one heading invites a comparison that flatters whichever row is cheapest. The same goes for the card itself: an SXM part and a PCIe part with the same name are not the same purchase.
We read each provider page on a schedule, and a rate older than 21 days is hidden rather than shown stale. If your page changes and the old figure is what we hold, the old figure is what a buyer sees until the next read. That is the reason to keep the page plain and current.
What to put on the page so a comparison is fair
State the card exactly, including the form factor and the memory size, in the name buyers will search for. State the unit as price per GPU per hour. State the commitment beside every rate, and the region if the rate is not the same everywhere.
Say what the rate includes. If the CPU, memory and local disk are part of the price, say how much of each, because a buyer comparing you with a provider that charges separately will otherwise assume the worst about one of you and it will not always be the other one.
Then claim the listing. Claiming a listing is free, it proves the page is yours, and it lets you correct a figure we read wrongly and answer the questions we ask of your category with a source. How a figure is gathered is written down, including what we refuse to publish.
What we could not answer
- ⚑We hold no data on how often buyers switch provider after a first rental, so this describes what buyers check, not what decides it.
- ⚑Availability is not in a rate and we do not measure it, so nothing here says which provider has cards free today.
What this rests on
infratoapp, “What buyers compare before they rent a GPU”, read 27 October 2026. providers with a current GPU rate: 48. https://www.infratoapp.com/blog/what-gpu-buyers-compare-before-they-rent
Questions people ask about this
What do buyers compare first when renting a GPU?
The hourly on-demand rate for the exact card, because it is the one figure every provider publishes in the same unit. After that they ask about availability for the number of cards they need, then network and storage, then terms such as commitment and cancellation.
Why does the same GPU cost so much more at some providers?
The rate is only part of what a buyer pays for. Availability, interconnect, support, how an interruptible instance is handled and what is bundled into the price all differ, and a dearer seller is usually selling some of those. The rate alone cannot say which.
Does paying infratoapp change where a GPU provider appears in a price table?
No. Listing is free and the query that orders a public table does not join the billing table. Paid placement is a labelled card outside the ranked list, and the pricing page says what no amount of money changes.
These render as FAQPage structured data · answers are filled from the same live figures
Read next
Before choosing an interruptible GPU a buyer needs the reclaim warning, what happens to the job and how the rate is set, not just the size of the discount.
Buyers who have paid to move data out of a cloud ask about egress first. A GPU page that is silent on it is read as hiding a charge, whether or not one exists.
A GPU pricing page buyers can compare names the card exactly, quotes per GPU per hour, labels the commitment and states egress, regions and access mode.