# One H100 is easy to rent, but an 8× node had stock less than half of October

The November 2026 gpu.compare GPU Index: the cheapest place to rent each GPU in October, and the price you can count on to find one.

- Updated hourly. Last scan 2026-11-16T22:30:26.764Z
- Canonical page: https://gpu.compare/trends/gpu-index-november-2026
- Machine-readable index: https://gpu.compare/llms.txt
- Author: Jack Britton, Product Growth (https://gpu.compare/authors/jack-britton)
- Published: November 4, 2026

Each month, the gpu.compare GPU Index tells you where to rent each GPU and the price you can count on: the lowest price that would have found you one in at least 9 of every 10 hours. It's for developers who need a GPU this week to train or fine-tune something. This edition covers October 2026.

In October, the price you can count on for a single H100 was $3.29, 79 cents above the cheapest one. A full 8× H100 node (one machine with eight GPUs) had no such price. You could find one in 43% of hours, and paying more didn't raise that. For most of September, one Scaleway listing in Paris was the only 8× H100 node in the index, and in October it ran dry. The nodes that came back rarely lasted long. Check at 13:00 and 21:00 UTC (9am and 5pm Eastern) every day, and you'd have missed two of every three.

## The index: where to rent a single GPU, from October data

Each row is one GPU, rented on demand from the four providers with stock readings for every hour of October: [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). [The methodology](https://gpu.compare/methodology#availability) explains how each provider reports stock.

| GPU                         | Memory       | Cheapest, and how often it had stock | Price you can count on  | September |
| --------------------------- | ------------ | ------------------------------------ | ----------------------- | --------: |
| [H100](https://gpu.compare/gpu/h100)           | 80 GB        | $2.50 at Hyperstack, 69% of hours    | $3.29                   |     $3.29 |
| [A100](https://gpu.compare/gpu/a100)           | 40 or 80 GB  | $1.35 at Hyperstack, 69%             | $1.99 (40 or 80 GB)     |     $1.99 |
| [L40S](https://gpu.compare/gpu/l40s)           | 48 GB        | $1.57 at DigitalOcean, 1%            | None (66% at any price) |      None |
| L40                         | 48 GB        | $1.00 at Hyperstack, 94%             | $1.00                   |     $1.00 |
| [RTX A6000](https://gpu.compare/gpu/rtx-a6000) | 48 GB        | $0.50 at Hyperstack, 78%             | None                    |      None |
| A10                         | 24 GB        | $1.29 at Lambda, 99%                 | $1.29                   |     $1.29 |
| L4                          | 24 GB        | $0.88 at Scaleway, 87%               | None                    |     $0.92 |
| [H200](https://gpu.compare/gpu/h200)           | 141 GB       | $4.47 at DigitalOcean, 38%           | None                    |      None |
| AMD MI300X or MI325X[^rocm] | 192 / 256 GB | $2.59 at DigitalOcean (MI300X), 7%   | $3.80 (MI325X)          |     $3.80 |

If you need an 80 GB A100, there was no price you could count on. Hyperstack's $1.35 card had stock 69% of hours.

If your model fits in 48 GB, read the L40 and RTX A6000 rows before the H100 one. They cost a fraction of an H100, and the L40 had stock more often than any H100 instance did. Skip the cheapest L40S listing, too: DigitalOcean's almost never had stock, so Scaleway's is the one you'll get.

[^rocm]: Yes, we know this row only helps if your stack runs on ROCm. If it does, enjoy being the one developer in this post who could rent their GPU every single hour of October.

## If you need a full node: October data

Here's the same index for 8× nodes:

| 8× node           | In stock at any price | Price you can count on | September |
| ----------------- | --------------------: | ---------------------- | --------: |
| H100              |          43% of hours | None                   |     $4.41 |
| A100, 80 GB       |                   90% | $2.79                  |      None |
| A100, 40 or 80 GB |                   93% | $1.99                  |     $2.79 |
| H200              |                    2% | None                   |      None |
| [B200](https://gpu.compare/gpu/b200) |              under 1% | None                   |      None |
| [B300](https://gpu.compare/gpu/b300) |                    2% | None                   |      None |

In September, the price you can count on for an 8× H100 node was $4.41. In October, there wasn't one. A100 nodes went the other way. In September, no 80 GB A100 node had a price you could count on. In October, $2.79 did. If 40 GB per GPU is enough, $1.99 found you a node in 93% of hours. If your training run fits on A100s, that's the easy node this month.

## One provider held up September


The H100 node fell off a ledge in the first week of October, from 85% of hours to 24%.

The setup happened a month earlier. DigitalOcean, Lambda, and Hyperstack made their last 8× H100 restocks between September 17 and 20. For two-thirds of September's hours, the only node any of the four had was one Scaleway listing in Paris. Nodes looked easy to get because a single provider kept restocking.

Then Scaleway ran dry too. In the week of October 5, its node had stock about one hour in eight, and nobody else had come back yet. Hyperstack returned on October 8. Lambda had a node for 12 hours all month, and DigitalOcean for 3.


Scaleway didn't have a node once in the last week of October, and the odds of finding one anywhere fell to 31%.

## How to rent an 8× H100 node anyway

The cheapest 8× H100 node on any list was Hyperstack's $2.50 PCIe node. It never had stock in October, and neither did Hyperstack's $3.20 SXM node. When you could get a node, the cheapest one cost $3.03 per GPU-hour on average, and most of those hours went to Hyperstack's $2.60 PCIe node with NVLink. If your job needs SXM (most large training jobs want all eight GPUs on one NVLink fabric), the floor was Scaleway at $3.53, and an SXM node had stock about one hour in five.

Nodes don't stay long. October had 56 separate windows when at least one node was in stock. Half of them closed within 3 hours, and three in four within 8. The longest, at Scaleway, opened on October 2 and lasted 49 hours. The gaps between windows were usually just as short, but the longest ran 51 hours, from October 28 into October 31.

That speed changes how you hunt. If you checked at 13:00 and 21:00 UTC (9am and 5pm Eastern) every day, you would've seen 19 of those 56 windows.

To rent one this month:

- script the check, so you catch the windows that close within a few hours.
- decide up front whether a PCIe node will do. Most of October's node hours were PCIe, and SXM started at $3.53.
- price an A100 node before you wait on an H100 one.

Check [the H100 page](https://gpu.compare/gpu/h100) for what's in stock right now.

## The newest GPUs barely show up on demand

B200 and B300 nodes barely existed on demand in October. An 8× B300 node had stock for 16 hours, all of them at Scaleway, and Hyperstack's $7.40 one never did. An 8× B200 node had stock for 5 hours.

If your job can take an interruption, rent spot. (Spot capacity is cheaper because the provider can take it back at any time.) DigitalOcean's single-GPU B300 spot droplet had stock 80% of October, and its price fell from $11.19 per GPU-hour in August to $6.17.

## In October, a listed node was usually a sold-out node

A listing gives you a price. Here's how often the listings from these four providers had stock in October, by size:

| Size           | Listings | Never in stock | In stock under 10% of hours |
| -------------- | -------: | -------------: | --------------------------: |
| 1 GPU          |       39 |             5% |                         13% |
| 2 to 4 GPUs    |       40 |            10% |                         38% |
| 8 or more GPUs |       47 |            36% |                         64% |

List prices didn't move while nodes sold out. My colleague Joel Hans covers that in [the launch post](https://gpu.compare/trends/introducing-gpu-compare). A listed price tells you what a node would cost if you could get one. The price you can count on tells you what it takes to get one.

## What I'm checking in December

In December, I'm checking:

- whether an 8× H100 node gets a price you can count on again.
- whether Hyperstack's $2.50 H100 node ever has stock, because it didn't once in October.
- whether single H200s come back.
- RunPod, once we have a full month of its hourly data (we started on October 29).

## How the index works

Every hour, gpu.compare reads each provider's API and records whether each instance is in stock. The index:

- names the provider behind each cheapest price, so you can go rent it.
- counts an hour as in stock when a provider would have taken an on-demand order for that instance, in any region, at the top of the hour.
- sets the price you can count on as the lowest per-GPU price where, in at least 90% of the month's hours, at least one instance at or below that price had stock.
- uses the four providers that report stock every hour. Hyperstack and Scaleway report it directly. DigitalOcean and Lambda drop a region from an instance when it sells out.
- counts one node and fifty nodes the same, since most provider APIs only say whether you could order at least one.
- leaves out spot, reserved, and marketplace capacity. It only sees the orders you could place right now with a credit card.

Read [our methodology](https://gpu.compare/methodology#availability) for the details on each provider. Point your agent at [the Markdown version](https://gpu.compare/trends/gpu-index-november-2026.md).

The December edition comes out in the first week of December. Follow [the RSS feed](https://gpu.compare/rss.xml) to get it.

To cite this edition, use "gpu.compare GPU Index, November 2026" and link to this page.

## 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.
