A checklist for a GPU pricing page buyers can compare
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.
The short version
A GPU pricing page that buyers can compare does six things on the line where each rate appears: it names the card with form factor and memory, quotes the price per GPU per hour, labels the commitment, says how the card can be had today, says which region the rate is for, and points to the egress policy. Each item removes a guess a buyer would otherwise have to make about the seller.
The posts before this one take each item in turn. What follows is the order to check them in, and the reason each sits where it does.
The order is reasoning, not measurement. It follows the points at which a page can lose a reader: first when the card cannot be identified, then when the unit cannot be compared, then when the commitment is unclear, and only after that on region, access and network. A rate whose card cannot be identified gives the later items nothing to attach to.
The card and the unit
First, the name. "H100 SXM 80GB", not "H100" and not an instance code. The post on naming the card shows the two H100 pages and the unlabelled A100 page this site keeps because so many rates leave the form factor out.
Second, the unit. Price per GPU per hour first, node price beside it. A buyer will not divide, and a node figure pasted into a column of card figures sinks to the bottom of the sort. The post on units works the arithmetic with the H200.
Check both against your own page by copying a single rate into an empty spreadsheet cell with whatever text sits on its line. If the card and unit are not in what you copied, a buyer will not have them either.
The commitment and the reclaim terms
Third, the commitment, written beside each figure: on-demand, reserved with its term, or interruptible. Labelling every rate covers why a mixed column flatters whichever row is cheapest.
Fourth, if you sell interruptible capacity, the three facts behind it: reclaim notice, what happens to the job and its disk, and how the rate is set. Describing interruptible instances sets out what buyers look for before trusting the discount.
Test this pair by reading only the row for your cheapest offer. If a stranger could mistake it for on-demand capacity, the label is missing or in the wrong place, and that row is winning comparisons it will later lose as a complaint.
Region and access
Fifth, the region. One headline rate with its scope stated, and a regional table ordered by region rather than by price. The regional pricing post explains why this site compares headline rates and leaves regional rows out.
Sixth, access. On demand, on request, waitlist or reserved only, with the largest configuration a buyer can start alone and the date the line was checked. We do not measure availability, and the page is the only place a buyer can learn it before trying.
Egress sits beside these as a single sentence near the rates: whether data transfer out is charged, with a link to the schedule. A buyer who has been caught once reads it before the rate; the egress post covers what the sentence should already answer, from object storage to traffic between regions.
Why the checklist matters more than the rate
Across the 48 providers with a current GPU rate, the widest gap between the dearest and cheapest seller of one card is 10.4×, on the RTX-A4000-16GB. Where sellers of one card sit far apart, a buyer has reason to read the small print, and a page that answers the checklist gives them small print worth reading.
None of these items is about lowering a price. A provider can be the dearest seller of a card and still win the buyer who needed a full node today, in the right region, with no egress charge, provided the page says so.
The checklist is also short on purpose. Six items on one line is about as much as a row can carry before it stops being read, and every other fact a provider wants to make, such as support hours, interconnect or storage tiers, belongs on a page linked from that row rather than squeezed into it.
After the page is right
Claim the listing on this site, which is free, and check that the values read from your page match it. Where one does not, a dispute with the live page as evidence is the route; the post on corrections explains what a claimed vendor edits directly and what goes through review.
Then keep the page current. A rate older than 21 days is hidden rather than shown stale, so a page that changes without the old figure being replaced on the next read simply drops out of the comparison until it is.
Last, read the page as an assistant would. It lifts a line, not a layout. If the rate, the card, the unit and the commitment survive being lifted on their own, the page is comparable; if any of them lives in a header, a tab or a footnote, it can be lost when someone asks a model for the cheapest card.
What we could not answer
- ⚑The checklist reflects what makes a rate comparable, not what wins a sale; we hold no data on which items change a buyer’s decision.
- ⚑Egress, availability and regional rates are not collected here, so the checklist cannot be checked against data for those items.
What this rests on
infratoapp, “A checklist for a GPU pricing page buyers can compare”, read 27 October 2026. providers with a current GPU rate: 48. https://www.infratoapp.com/blog/gpu-pricing-page-checklist-for-buyers-to-compare
Questions people ask about this
What should a GPU pricing page include?
The exact card name with form factor and memory, the price per GPU per hour, the commitment beside each rate, the access mode such as on demand or on request, the region each rate applies to, and a sentence on whether data transfer out is charged, with a link to the detail.
Why do buyers skip some GPU pricing pages?
A page gets skipped when a rate cannot be placed: the card is unnamed, the unit is per node, the commitment is unlabelled or the region is unclear. A buyer comparing many providers can move to a page that answers those questions rather than write to sales to find out.
How can a GPU provider test whether its pricing page is comparable?
Copy one rate, with whatever text sits on its line, into an empty spreadsheet cell. If the card, the form factor, the unit and the commitment are not in what was copied, a buyer comparing providers will not have them either and will have to guess.
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.
On-demand, reserved and interruptible are three products. Put the label beside every rate, or buyers will compare your cheapest row with a rival’s dearest.