All GPU classes · GPU prices per hour
Google Cloud GPU prices per hour
What Google Cloud lists per GPU-hour right now for the classes it sells, each with the instance the price sits in, and what the list price did over the last weeks. Read four times a day; described, never forecast.
Against the week before, at Google Cloud: A100 0 %, H100 0 %, H200 0 %.
Radar Rent Index
What Google Cloud prices did
The same rule applied to Google Cloud's own daily series: the current window against the one before it, and against the first week collected. A class with too few days says so instead of showing a change.
| GPU | This week | vs. last week | Since start | Offers | Day band $/h |
|---|---|---|---|---|---|
| A100latest day Nov 19 | $3.48median of 7 days | 0 %flat | 0 %since Nov 3 · 17 days on record | —no comparison yet | —min–p75 of the latest day |
| H100latest day Nov 19 | $11.06median of 7 days | 0 %flat | 0 %since Nov 3 · 17 days on record | —no comparison yet | —min–p75 of the latest day |
| H200latest day Nov 19 | $10.60median of 7 days | 0 %flat | 0 %since Nov 3 · 17 days on record | —no comparison yet | —min–p75 of the latest day |
No day was excluded. A 30-day change appears once thirty days are on record, from Dec 2.
A day is the median of its hourly medians, counting only hours with at least five offers (verified hosts on Vast.ai, free hosts on Clore.ai). A week is the median of its days and needs at least four of them. Words follow fixed thresholds: up or down from ±8 %, sharply from ±20 %; offers up or down from ±15 %. How the index is made →
Price history
Reading Google Cloud's prices
As of today's readingsOn November 19, 2026 Google Cloud listed 3 of the 8 tracked classes: A100 $3.48, H100 $11.06, H200 $10.60 per GPU-hour. It was not the cheapest source on any class that day. Its widest distance to the cheapest reading was on the H100: $11.06 against $2.50 at Hyperstack, 4.4 times as much. Google Cloud has been read here since November 3, 2026.
Google Cloud sells its datacenter GPUs as machine types: a fixed bundle of one to sixteen cards with vCPUs, memory and, on the newer families, local SSD, billed by the hour on demand. The number on this site is the on-demand price of the cheapest machine type of a class divided by its GPU count, read from Google's accelerator-optimized pricing page as it is served for Iowa (us-central1), the region the page shows before any is chosen. Other regions list other prices and are not read.
Four times a day the site reads the GPU machine-type tables of that page and keeps the rows of the tracked families: A2 (the A100 with 40 or 80 GB), A3 (the H100, and the H200 in A3 Ultra) and A4 (the B200). A row counts only with a plain hourly on-demand price; a machine type that lists none, as the A4 machine with the B200 did when this source joined, only Dynamic Workload Scheduler and committed prices, gets no figure here rather than a price from another column. Spot, Flex-start, Calendar mode and committed-use prices are left out, as the spot prices of Azure and Nebius are.
Google Cloud's per-GPU figures sit with the other hyperscalers rather than with the GPU clouds, which is the product: a machine type in a region of a large cloud, with its services around it, at list price before any commitment. The A2 family is the one place here a renter can take a single card; the row still names the machine type with the lowest price per card, which can be a larger machine such as the sixteen-card a2-megagpu-16g sitting a little under the one-card a2-highgpu-1g per GPU-hour. The H100 and H200 come as eight-card machines on this page, so the row's instance price is the minimum spend, as with AWS and Oracle. The marketplaces show what the same hour costs from a private host.
Price moves at Google Cloud, last 30 days
Any change of the list price between complete days; the big clouds move on the scale of months.
No move on record in the last 30 days.