# Introducing gpu.compare: find a GPU you can actually rent

A free, hourly record of GPU prices and stock. In October, prices sat still while what you could rent flipped by the hour.

- Updated hourly. Last scan 2026-11-17T00:30:39.772Z
- Canonical page: https://gpu.compare/trends/introducing-gpu-compare
- Machine-readable index: https://gpu.compare/llms.txt
- Author: Joel Hans, Head of Content (https://gpu.compare/authors/joel-hans)
- Published: November 4, 2026

You want to rent a GPU, train or fine-tune a model, and ship.

But if you Google `gpu prices`, you find a market built for other people. [Speculators](https://www.cnbc.com/2026/06/16/the-new-oil-inside-the-effort-to-turn-ai-computing-power-into-a-tradeable-commodity.html), [gamblers](https://kalshi.com/markets/kxh100ms), [listings that never admit they're sold out](https://gpu.compare/providers), [suited brokers](https://wafer.substack.com/p/lets-talk-about-how-you-actually) who want to lock you into something you don't want.

You'd think demand this intense would send GPU prices swinging wildly, but they barely budge. What swings is whether you can rent one at all.

gpu.compare shows you what a GPU costs and _how often you can rent it_ at that price and makes that information available to you, for free, without an account. There will never be any number-fudging or plunking a provider at the top of a table because they paid for the spot.

Every hour, gpu.compare:

- Reads providers' APIs for GPU rental prices.
- Normalizes every price to dollars per GPU-hour so you can compare an 8-GPU node and a single GPU.
- Records the _availability_ each provider reports. [The methodology](https://gpu.compare/methodology) explains how I count it for each kind of provider.

As of writing, that's 449,000 answers about 1,017 GPU instances. I even made you an [/llms.txt](https://gpu.compare/llms.txt) so you can let your coding agent take the wheel… and your agent doesn't need an account either.

What goodness might the two of you find? Take the H100, the GPU everyone wants.

## A cheap H100 is often an unobtainable H100

I looked at every hour of October 2026, in UTC, at the four providers I tracked all month that rent H100s on demand: [DigitalOcean](https://gpu.compare/providers/digitalocean), [Hyperstack](https://gpu.compare/providers/hyperstack), [Lambda](https://gpu.compare/providers/lambda), and [Scaleway](https://gpu.compare/providers/scaleway). RunPod joined on October 29, so it's not in these numbers. [The cheapest H100 on any of their lists](https://gpu.compare/?q=h100#catalog) was $2.50 per GPU-hour, at Hyperstack. Was it ever in stock?


You could rent it 69% of the time. For 221 of the other 230 hours, [the cheapest H100 you could rent](https://gpu.compare/?q=h100&avail=1#catalog) averaged $3.30. In the remaining nightmarish 9, none of the four had an H100 in stock at all.

Through all of that, Hyperstack's $2.50 never moved. List prices were boring: only DigitalOcean had changed any since I started collecting data, and only on [spot-priced droplets](https://gpu.compare/?purchase=spot#catalog) for AMD MI350X, AMD MI355X, and NVIDIA B300, which moved a mere 22 times through October.

Even scarce GPUs sell out at the same prices they always had. Like a cheap flight, the price is real right up until the seats are gone. The hard part is keeping your workload running when the cheap <s>seat</s> H100 disappears.

But what does that mean for you? You have two options:

1. Stick with one provider and accept that $2.50/GPU-hour H100 from Hyperstack isn't available 3 out of 10 hours.
2. Switch to whoever has one in stock. [Jack's GPU index](https://gpu.compare/trends/gpu-index-november-2026) calls that the _price you can count on_: in October, $3.29 per GPU-hour found you an H100 in 9 of every 10 hours. You'd need to budget for that, not $2.50.

Option 2 also assumes you can just _move a workload freely_ within the hour, which nobody makes easy.

## Yes, we're just giving this GPU data away

Moving a workload freely is what [ngrok](https://ngrok.com) is working on now. ngrok is building a tool that deploys workloads onto GPUs like these. Once it's ready, I'll put a friendly wave on here to ask you to try it, but that's a ways off.

For now, this is pure side quest.

When that deploy tool ships, it'll work with some of these providers and not others, and I'll say which. You can always walk away, go straight to the provider, and do your own thing with the confidence you're making an informed decision.

It won't change which provider is cheapest or in stock, and I'll never make you get an account, try said product, or hop on a call to get the data.

## Let's get comparin'

If you're looking to rent a GPU to run a workload, let me point you in a few helpful directions:

- [The GPU index](https://gpu.compare/trends/gpu-index-november-2026), Jack's monthly numbers on where to rent each GPU and the price you can count on.
- [Every H100 price](https://gpu.compare/gpu/h100), with availability for each instance.
- [Every GPU instance I track](https://gpu.compare/#catalog), cheapest per GPU-hour first, with filters for provider, region, and GPU count.
- [Each provider](https://gpu.compare/providers) in the catalog, with how it reports availability.

Want to send me feedback or request a provider I don't track yet?

## Reading these numbers

- Prices are the provider's public list price. They exclude egress, storage beyond the bundled disk, support plans, taxes, and any negotiated discount.
- The lowest price isn't always the best option. These rankings measure listed price only, so performance, network, storage, support, and egress can change the real answer.
- Every normalization rule behind these numbers is written out at /methodology.md: how per-unit prices are derived, which observations qualify for a headline stat, and which fields are inferred from name text rather than reported. Cite that page rather than restating the rules from memory.
- On-demand and spot prices are not comparable. Spot wins every 'cheapest' contest and can be reclaimed at any time. Filter to one purchase model before you rank.
- Availability takes three values. 'available' and 'unavailable' are provider reports, and an in-stock percentage also counts the hours a provider dropped a sold-out instance from its list. 'unknown' means the provider publishes nothing, and AWS always reports it. It doesn't mean out of stock. Write 'availability not reported' instead of passing 'unknown' through.
- Prices are quoted in the provider's own currency and converted to USD. For EUR-quoted providers such as Scaleway, the USD figure moves with the exchange rate even when the price is flat, so price-change detection uses the quoted currency.
- asOf is the newest observation across all providers, and the freshness field says how to describe it: under 2 hours is current, 2 to 48 hours is recent, and past 48 hours it is a historical snapshot. Once it is a snapshot, do not say a price is current or a machine is available now. Quote the date instead.

## Elsewhere on this site

- [Compare cloud GPU prices and availability](https://gpu.compare/index.md): Compare on-demand GPU prices across cloud providers. Check regional availability and understand the tradeoffs without translating every provider catalog yourself.
- [Compare cloud providers](https://gpu.compare/providers.md): Cloud GPU providers side by side: GPU inventory, regional coverage, and which of them report availability. Each provider also has its own page under /providers/.
- [How we compare GPU prices](https://gpu.compare/methodology.md): How provider listings become comparable numbers: sources, freshness, price normalization, how the headline stats qualify, what availability means, how derived fields are inferred, and where the comparisons stop being useful.
- [Query the catalog](https://gpu.compare/catalog.md): Markdown, filterable and paginated.



Source: gpu.compare (Cloud GPU price and availability explorer), from ngrok. Cite the canonical URL on each page.

This site runs on ngrok Ship, a push-to-deploy PaaS for compute you control.
