Skip to content
InfratoApp
Tools / GPU cost calculator

What a GPU costs you a month

Live rates, your workload. Every rate cites the page it was read from, and the URL carries your inputs so the calculation is shareable.

LLM $/1M tokGPU $/hr
GPU · up to 4
Terms

Spot rates are interruptible · a reserved rate assumes its term is served

1 × H200-SXM-141GB for 730 hours a month costs $1,526 a month at Beam, against $4,603 at CoreWeave (on demand) — 3.0× the cheapest, across 32 rates.

GPU · seller$/mo ↓vs cheapestRead from
H200-SXM-141GBBeam$1,526—2026-10-27
H200-SXM-141GBShadeform$1,6431.1×2026-10-26
H200-SXM-141GBTheta EdgeCloud$1,6721.1×2026-10-26
H200-SXM-141GBSesterce$1,8101.2×2026-10-07
H200-SXM-141GBKoyeb$2,1901.4×2026-10-26
H200-SXM-141GBCivo (on demand)$2,5481.7×2026-10-27
H200-SXM-141GBDeepInfra$2,5481.7×2026-10-26
H200-SXM-141GBLium$2,8471.9×2026-10-27JSON
H200-SXM-141GBVast.ai (on demand)$2,8841.9×2026-10-27
H200-SXM-141GBVast.ai (on demand)$2,8911.9×2026-10-27
H200-SXM-141GBHyperbolic$2,9131.9×2026-10-26
H200-SXM-141GBHyperstack$2,9131.9×2026-10-26
H200-SXM-141GBOblivus (on demand)$2,9131.9×2026-10-26
H200-SXM-141GBCanopy Wave$2,9201.9×2026-10-27
H200-SXM-141GBCerebrium$3,0662.0×2026-10-27
H200-SXM-141GBDigitalOcean (on demand)$3,2632.1×2026-10-11
H200-SXM-141GBVast.ai (on demand)$3,2632.1×2026-10-27
H200-SXM-141GBModal$3,3142.2×2026-10-26
H200-SXM-141GBVast.ai (on demand)$3,3512.2×2026-10-27
H200-SXM-141GBVast.ai (on demand)$3,3582.2×2026-10-27
H200-SXM-141GBVast.ai (on demand)$3,3652.2×2026-10-27
H200-SXM-141GBRuncrate$3,3802.2×2026-10-26
H200-SXM-141GBSpheron$3,5042.3×2026-10-26
H200-SXM-141GBVast.ai (on demand)$3,6872.4×2026-10-27
H200-SXM-141GBRunPod (on demand)$3,8622.5×2026-10-27
H200-SXM-141GBReplicate$4,0082.6×2026-10-26
H200-SXM-141GBEnverge$4,0152.6×2026-10-26
H200-SXM-141GBCrusoe$4,3802.9×2026-10-27
H200-SXM-141GBFal.ai$4,3802.9×2026-10-26
H200-SXM-141GBVerda (DataCrunch) (on demand)$4,4272.9×2026-10-27
H200-SXM-141GBLyceum (on demand)$4,5193.0×2026-10-26
H200-SXM-141GBCoreWeave (on demand)$4,6033.0×2026-10-26

Count × hours × the rate each provider published · nothing here is averaged, and a provider we hold no current rate for is absent rather than estimated · every rate on the GPU listing

Frequently asked

What does one RTX-3090-24GB cost per month, 24/7?
$65.70 at today's cheapest posted on-demand rate — $0.09 an hour × 730 hours. That is a list price at one seller, not a quote: committed terms are usually cheaper where a seller publishes them, and we only list rates that are published.
When does buying a GPU beat renting one?
We do not answer that, because the honest version needs figures nobody publishes: your power cost, your rack, your utilisation and how long the card stays current. What the calculator gives you is the rental side of that comparison, priced from posted rates and dated — the half somebody selling you hardware has no reason to get right.
Are these prices what I would actually pay?
They are what 45 sellers publish across 51 cards, read from their own pages with the date. What they cannot include is your negotiated discount, your committed-use terms, egress, storage or support — so the figure is a list price for the machine you described, not a bill.
How many GPUs do I actually need?
Enough memory to hold the weights, plus room for the context and the batch. A model that does not fit in one card has to be split across two, and two cheap cards frequently cost more than one larger one — which is why the listing shows VRAM beside every rate rather than ranking on price alone.
Does the cheapest provider stay cheapest at scale?
Not reliably. The spread between sellers of an identical card is wide and moves, and a provider cheap on one part is often not cheap on another. The ranking here is recomputed from the current readings each time the page is built, so it reflects what is posted now rather than a position anybody holds.
Can a provider pay to come out cheapest here?
No. The ranking is the arithmetic of the rates we read, in one stated order, and nothing in it can be bought. How the site is paid for is set out under money and independence at /our-data#sec5.

Related

Buying the output instead of the card is on the LLM cost calculator, and every rate behind this one is on GPU prices.