---
title: Where to Rent an A100 SXM at the Cheapest Price (October 2026)
url: https://gpubank.io/rent-a100
description: A100 SXM 80 GB prices checked on 16 October 2026: Vast.ai from $0.94, GPUBank $0.95 all-in, CloudRift $1.05, AWS $3.43, and cost per job against the H100.
generated: 2026-11-02T00:20Z
publisher: GPUBank (GPU Compute OÜ, Tallinn, Estonia)
summary_for_ai: https://gpubank.io/llms.txt
full_text_for_ai: https://gpubank.io/llms-full.txt
---

> Markdown version of https://gpubank.io/rent-a100, generated from the live page. GPUBank is a GPU cloud billed by the second at one all-in price per GPU (vCPU, RAM, NVMe, a public IPv4 and traffic included), in Paris, Frankfurt, Ashburn and Singapore. Prices are in US dollars, excluding VAT.

Updated 16 Oct 2026 · 36 offers checked

# Where to rent an A100 SXM at the cheapest price

We read every A100 SXM 80 GB offer we could find on the provider's own page, worked out what each hour includes and what a month really costs, and measured the ageing card against the H100 per finished job rather than per hour.

**GPUBank Research** 20 min read · Prices read on 16 Oct 2026

**80 GB**HBM2e

**2.0 TB/s**memory bandwidth

**600 GB/s**NVLink

**2020**launched

**No FP8**Ampere tensor cores

The short answer, 16 Oct 2026

1. $0.95 /hr **Cheapest A100 SXM machine on demand.** GPUBank at $0.95 an hour, billed per second, with 16 vCPU, 125 GB of RAM, a 1 TB NVMe disk, a public IPv4 address and all traffic in the price; $0.76 reserved for a week or a month and about $0.59 effective after the bonus on a $1,000 top-up. The next managed offers were CloudRift at $1.05 and Thunder Compute at $1.09.
2. $0.53 /hr **Cheapest interruptible hour.** A Vast.ai bid of about $0.53 on an Iowa host, the lowest minimum bid among the live A100 SXM4 80 GB offers on 16 October 2026; a higher bidder can stop it at any time and storage is billed on top. GPUBank's interruptible tier, with five minutes of notice, costs $0.57.
3. $0.94 /hr **Cheapest marketplace listing.** The same Iowa host on demand at $0.937 with 32 CPU cores, before its storage at about $0.33 per GB-month and its bandwidth; the cheaper A100 SXM4 rows that day were all 40 GB cards.

[A100 SXM $0.95 /hr all-in Launch an A100 SXM Spot $0.57 · reserved $0.76
Billed per second](https://console.gpubank.io/launch?gpu=a100-sxm)

The A100 SXM is six years old and still one of the most rented 80 GB GPUs there is. That longevity has produced an odd market: the same card is listed at $0.94 an hour by a marketplace host in Iowa, at $1.05 to $1.79 by specialised clouds that sell one card with its CPU, memory and disk, at $2.79 per GPU by Lambda in an eight-GPU instance, and at $3.43 to $5.07 by the large clouds. GetDeploying, which tracks 39 priced providers, put the median on-demand A100 hour at $1.88 on 15 October 2026, a figure that mixes 40 GB, 80 GB, PCIe and SXM rows and therefore reads better as a market temperature than as the price of this exact card.

The spread is not noise. Each price buys a different machine, a different meter, a different minimum unit and a different promise about what happens when a host breaks, and many listings that say only "A100 80GB" never state whether the card is the SXM module or the PCIe board. This page compares like with like, keeps the 40 GB and PCIe versions apart, and then asks the question that matters for an ageing card: whether it is still cheaper per finished job than an H100 that costs about twice as much per hour. It is published by GPUBank, a GPU cloud run by GPU Compute OÜ in Tallinn and one of the providers compared, so read our own entry with that in mind: it is priced on the same terms as every other offer, and the text says where a marketplace host, a spot bid or a faster card beats it.

01 / Still cheap, per job?

## An ageing card that is still cheap

The A100's price history is less of a straight decline than its age suggests. The large clouds set the ceiling and have barely moved it: Azure's ND96amsr A100 v4 has listed at $4.10 per GPU since December 2021, and AWS's p4de fell from $5.12 per GPU in 2022 to $3.43 only after a price cut in June 2025. The specialised clouds did not fall in a line either: Lambda's eight-GPU A100 80 GB went from $1.50 in May 2023 to $1.79 in 2024 and 2025 and costs $2.79 today, and Runpod's Secure Cloud SXM went from $1.84 in 2023 to $2.29 in 2024 and back to $1.79. What did fall is the floor: in October 2026 a whole machine costs $0.95 and a marketplace host $0.94, while the card itself has gone from roughly $10,000 new in 2023 to about $5,000 in realised used sales.

Renting one A100 80 GB, 2021 to 2026 Large clouds, per GPU of 8 Lambda, 8 GPUs Runpod, 1 GPU Vast.ai GPUBank

USD per GPU-hour, on demand. Every point on the chart: Dec 2021 **$4.10** Azure ND96amsr A100 v4, list price per GPU (8x SXM4 80 GB), unchanged since Aug 2022 **$5.12** AWS p4de.24xlarge on demand, $40.96 for 8x A100 80 GB May 2023 **$1.50** Lambda 8x A100 80 GB on demand Jul 2023 **$1.84** Runpod Secure Cloud, 1x A100 SXM 80 GB Jun 2024 **$2.29** Runpod Secure Cloud, 1x A100 SXM 80 GB Mar 2025 **$1.79** Lambda 8x A100 SXM 80 GB on demand Oct 2026 **$3.43** AWS p4de.24xlarge on demand, after the June 2025 cut Oct 2026 **$2.79** Lambda 8x A100 SXM 80 GB on demand Oct 2026 **$0.937** Vast.ai cheapest A100 SXM4 80 GB listing, on demand Oct 2026 **$0.95** GPUBank A100 SXM on demand, whole machine

Buying one, 2020 to 2026 8-GPU system, per GPU One module, new then sold used Used module, quoted or asked

USD per GPU; whole systems are divided by eight. Every point on the chart: May 2020 **$24,875** NVIDIA DGX A100 launch price (8x A100 40 GB), $199,000 for eight Feb 2023 **$10,000** A100 chip, "roughly $10,000" (CNBC, variant not stated) Mar 2026 **$7,000** Used A100 SXM 80 GB, $5,000 to $9,000 (Jarvislabs guide) Aug 2026 **$5,050** Used A100 SXM4 80 GB, realised sales over 12 months, down about 35 percent, range $4,700 to $5,400 Sep 2026 **$9,925** Used A100 SXM4 80 GB, median US asking price, 4 listings, range $8,300 to $12,200 Oct 2026 **$11,900** Used HGX A100 8x 80 GB node, valuation, $95,200 for eight

### Cheaper per hour, but not always per job

Cheap per hour is not the same as cheap per job, and the measured ratios decide which card to rent. On GPUBank the H100 SXM costs 1.95 times the A100's rate, so the A100 stays the sensible choice wherever the H100 is less than about twice as fast: long-sequence and memory-bound inference, ResNet-style training, BF16 stacks that cannot use FP8, QLoRA runs that spend real time loading data and writing checkpoints, notebooks that sit idle, and any session billed while you think. The H100 is cheaper per job on short, compute-heavy work and on anything that can run FP8, as the BERT result shows, and its advantage grows with concurrency. The practical test costs little: run your own job for half an hour on each card at per-second rates, divide the price by the throughput you measured, and keep the cheaper card.

A100 SXM **$0.95 /GPU-hr**

H100 SXM **$1.85 /GPU-hr**

The H100 must be this much faster to cost less **1.95×**

BERT-large training, 8 GPUs (MLPerf Training v2.1)

H100 measured **2.65×** faster, needs **1.95×**

A100 8×, 16.88 min **$2.14**

H100 8×, 6.38 min **$1.57**

per run

H100 cheaper

ResNet-50 training, 8 GPUs (MLPerf Training v2.1)

H100 measured **1.95×** faster, needs **1.95×**

A100 8×, 28.69 min **$3.63**

H100 8×, 14.68 min **$3.62**

per run

Tie

Llama 3.1 8B, 1,000 in and 1,000 out tokens, 1 GPU (E2E, TensorRT-LLM)

H100 measured **1.98×** faster, needs **1.95×**

A100 1×, 3,747 tok/s **$0.070**

H100 1×, 7,425 tok/s **$0.069**

per million tokens

Tie

Cost of each run at GPUBank on-demand rates: minutes to target × GPUs × price per GPU-hour, or price per GPU-hour divided by tokens per hour for serving. A gap under 5 percent is called a tie. The meter marks the speed-up the H100 needs (1.95×, its price over the A100's) and the speed-up measured.

02 / The card

## What the A100 SXM is, and who still rents it

The A100 SXM4 80 GB is NVIDIA's Ampere data-centre accelerator in its mezzanine form, a module that sits on an HGX baseboard rather than in a PCIe slot. NVIDIA's specification gives it 80 GB of HBM2e at 2,039 GB/s, NVLink at 600 GB/s between the cards of a board, a 400 W thermal design power in the standard configuration and up to seven MIG instances of 10 GB each, with 312 TFLOPS of dense BF16 and FP16 tensor throughput. It has no FP8, which is the main architectural gap against Hopper, where FP8 roughly doubles tensor throughput again for the models and frameworks that use it. The first A100s shipped in 2020 with 40 GB of memory, the 80 GB SXM module followed late that year and the 80 GB PCIe card later still; the PCIe board has a lower power limit, slightly less bandwidth and, in most servers, no full NVLink mesh, so the three versions are treated as different products in every price comparison on this page.

The variant problem is real when shopping. AWS's p4d.24xlarge carries eight 40 GB cards and only the p4de.24xlarge carries 80 GB, Lambda's single-GPU A100 is a 40 GB card while its 80 GB SXM comes only in an eight-GPU instance, and Thunder Compute, Jarvislabs, Koyeb and Modal list an "A100 80GB" without stating the form factor. The people who still rent this card are mostly fine-tuning 7B to 70B models with LoRA or QLoRA, running batch inference in BF16 on stacks that do not use FP8, keeping Ampere-era CUDA code alive, or splitting one card into MIG slices for several small services. The QLoRA authors fine-tuned a 65B model on a single 48 GB GPU, so 80 GB leaves room for a 70B model at 4-bit with a useful context and batch size.

03 / The price ladder

## How to read the A100 SXM market

A managed cloud sells a fixed machine, the GPU with its vCPUs, memory and local disk, at one hourly price that may or may not include the public IP, the persistent storage and the traffic. For the A100 SXM that group runs from GPUBank's $0.95 through CloudRift at $1.05, Thunder Compute at $1.09, Beam at $1.36, Massed Compute at $1.38, Jarvislabs at $1.49 and Hyperstack at $1.60 to Runpod Secure Cloud at $1.79 and Verda at $1.996. A peer-to-peer marketplace lets independent hosts post their own GPU price and bills storage and bandwidth as separate lines set by each host, which is why Vast.ai's $0.937 and a managed $0.95 are not two prices for the same product even though they sit a cent apart.

The other three kinds of offer trade something away for the price. Spot or interruptible capacity can be taken back: on Vast.ai you bid and can be outbid, GPUBank gives five minutes of notice and stops rather than deletes the machine, and on Azure the eight-GPU ND96amsr A100 v4 spot instance is evicted when the capacity is wanted elsewhere. Reserved capacity is paid for whether you use it or not, and for this card it ranges from GPUBank's $0.76 for a week or a month and CloudRift's $0.89 for three months to Hyperstack's $1.36. Serverless offers bill only the seconds in which a request or job runs, at an hourly equivalent of $1.60 to $2.72, which pays off only when the GPU would otherwise sit idle most of the time.

Two traps recur on this card. The first is the minimum unit: Lambda, Crusoe and every large cloud except Google sell the 80 GB SXM only as an eight-GPU instance, so a per-GPU figure of $2.79 is in practice a bill of $22.32 an hour. The second is the static price page: Vast.ai's A100 SXM4 page showed "No current offers" when we fetched it, while its live offers API listed eight rentable 80 GB machines at the same moment, and Lium showed a reference price of $0.70 with no pod available, so a marketplace figure is only real at the moment you can actually launch on it.

*GetDeploying median, 39 providers $1.88*

**Managed on demand** A fixed machine at one hourly price.

*GPUBank**$0.95***

*CloudRift**$1.05***

*Thunder Compute**$1.09***

*Beam**$1.36***

*Massed Compute**$1.38***

*Jarvislabs**$1.49***

*Hyperstack**$1.60***

*Runpod**$1.79***

*Verda**$2.00***

*Crusoe**$2.30***

*Lambda**$2.79***

**Marketplace** Hosts set their own price; machines vary.

*Lium**$0.70***

*Vast.ai**$0.937***

*Runpod Community**$1.39***

*TensorDock**$1.42***

**Spot and interruptible** Cheaper hours that can be taken back.

*Vast.ai**$0.534***

*GPUBank**$0.57***

*TensorDock**$0.67***

*Microsoft Azure**$1.06***

*AWS**$2.59***

*Google Cloud**$3.04***

**Reserved** Committed for a term, paid either way.

*GPUBank**$0.76***

*CloudRift**$0.89***

*CloudRift**$0.95***

*Hyperstack**$1.36***

*AWS**$1.46***

*Thunder Compute**$1.49***

*Microsoft Azure**$1.80***

**Serverless** Billed by active seconds.

*Koyeb**$1.60***

*Koyeb**$2.15***

*Modal**$2.50***

*Runpod Serverless**$2.72***

USD per GPU-hour, read on 16 Oct 2026. Hover a dot for the offer; crowded dots keep their label in the table below. The large clouds are listed separately.

04 / Provider by provider

## What each price buys, and what it leaves out.

The captures below were taken on 16 Oct 2026 from each provider's own A100 or pricing page, so every figure can be read next to the conditions printed beside it.

gpubank.io/pricing

**GPUBank** All-in

$0.95/hr

Managed on demand · Paris, Frankfurt, Ashburn, Singapore

16 vCPU 125 GB RAM 1 TB NVMe IPv4 All traffic Disk $0.03/GB-mo

GPUBank lists the A100 SXM at $0.95 an hour on demand, $0.000264 a second, in machines of one to eight cards joined by NVLink, and the rate covers 16 vCPU, 125 GB of RAM and a 1 TB local NVMe disk per GPU, one public IPv4 address and every byte sent or received. The only other line a bill can carry is a persistent disk at $0.03 per GB-month. Billing is per second with no minimum and starts only once the machine has passed its health check, a failed launch is not billed, and a host failure during a job restarts the machine on a healthy host with the same disk and credits the session in full. The interruptible tier costs $0.57 with five minutes of notice and a stop rather than a deletion, and a week or a month reserved costs $0.76. Payment is a prepaid balance topped up in crypto from 25 dollars, with a bonus that brings the effective hour to between $0.79 and $0.59; cards are not accepted for now.

Checked 16 Oct 2026

cloud.vast.ai/create/

**Vast.ai**

$0.94/hr

Marketplace · host-set prices, 1 to 8 GPUs

Host's CPU and RAM share Storage at host rate Bandwidth per GB Interruptible from about $0.53

Vast.ai's live offers API listed eight rentable A100 SXM4 80 GB machines on 16 October 2026, the cheapest at $0.937 per GPU-hour on demand on a host in Iowa with 32 CPU cores and a reliability score of 0.97, the same host in one, two, four and eight-GPU sizes, and a French host at about $1.00. Interruptible minimum bids on those machines started at about $0.53. The headline is not the whole bill: that host charged about $0.33 per GB-month for storage and $0.004 per GB of bandwidth, and Vast bills storage for as long as an instance exists, running or stopped. The cheaper A100 SXM4 rows in the same search, from $0.47, were 40 GB cards, and the public price page showed "No current offers" because it does not render live listings. Read a listing's reliability, its storage rate and its uplink before comparing it with a managed price.

Checked 16 Oct 2026

cloudrift.ai/pricing

**CloudRift**

$1.05/hr

Managed on demand · up to 8 GPUs per instance

800 GB local storage Networking $0.95 one month $0.89 three months

CloudRift is the closest managed competitor on price. Its pricing page lists the A100 SXM4 80 GB at $1.05 an hour on demand, $0.95 on a one-month commitment and $0.89 on three months, with 800 GB of local storage per GPU, instances of up to eight cards and billing by the second. The page does not state the vCPU and RAM that come with each card, the regions or live stock, nor a price for storage that survives an instance, so a workflow that keeps datasets between runs has to plan for re-uploading them or for an extra line we could not price. For a buyer who wants a managed single-GPU machine paid by card, it is the cheapest option we found, ten cents above GPUBank's list price, and its three-month rate is the cheapest reservation outside GPUBank's own.

Checked 16 Oct 2026

thundercompute.com/pricing

**Thunder Compute**

$1.09/hr

Managed on demand · variant not stated

8 vCPU 64 GB RAM 100 GB disk Extra vCPU $0.04/hr Reserved $1.49 from 8 GPUs

Thunder Compute lists an "A100 80GB" at $1.09 an hour, billed per minute, with 8 vCPU, 64 GB of RAM and the first 100 GB of disk per GPU included, and lets you add vCPUs at $0.04 an hour each up to 16 and disk up to 500 GB. Its page labels its H100 as SXM but does not say whether the A100 is the SXM module or the PCIe card, so this row is not strictly comparable with the SXM offers around it. Reserved capacity is quoted at $1.49 per GPU-hour from 8 up to more than 512 GPUs, which is above its own on-demand rate and is aimed at teams that need guaranteed capacity rather than a lower price. The machine is smaller than GPUBank's or Beam's, so a data-loading-heavy fine-tune may need the paid vCPU upgrade, which narrows the price gap with the managed offers above it.

Checked 16 Oct 2026

vm.massedcompute.com/pricing

**Massed Compute**

$1.38/hr

Managed on demand · United States

14 vCPU 100 GB RAM 625 GB storage 8 GPUs at $11.04/hr

Massed Compute lists the A100 SXM4 80 GB at $1.38 an hour for a single card with 14 vCPU, 100 GB of RAM and 625 GB of storage, and at $11.04 an hour for an eight-GPU machine, which is the same $1.38 per GPU. It also lists a plain "A100 (80GB)" from $1.35 and a DGX A100 at $1.38 per GPU, so check which line you launch. The page does not show live stock, and Massed Compute capacity is also resold by brokers and other clouds, which is why the same machine shape appears under several names in aggregator tables. Among single-GPU managed offers it sits in the middle of the market, about 45 percent above GPUBank and 23 percent below Runpod Secure Cloud, with one of the few published eight-GPU A100 SXM prices that does not rise per GPU when you scale up.

Checked 16 Oct 2026

runpod.io/pricing

**Runpod**

$1.79/hr

Secure Cloud on demand · per second

16 vCPU 125 GB RAM No traffic charges Volume $0.10 to $0.20/GB-mo Serverless $2.72/hr

Runpod's pricing page, updated on 8 October 2026, lists the A100 SXM 80 GB at $1.79 an hour on Secure Cloud with 16 vCPU and 125 GB of RAM, billed per second with no traffic charges, up from $1.59 at the end of September. Disks are extra: container and volume disks cost $0.10 per GB-month while the pod runs, an idle volume $0.20, and network storage $0.07 per GB-month under 1 TB and $0.05 above. The structured data in the page still gives $1.39 for an A100 SXM on Community Cloud, its peer-hosted tier, but the visible table did not show that row, and Runpod Serverless charges $2.72 an hour for an A100 80 GB worker. Runpod costs nearly twice GPUBank's rate for a similar machine; what it offers in return is card payment, a very large template ecosystem and the serverless product on the same account.

Checked 16 Oct 2026

### Where GPUBank wins, and where it does not

GPUBank rents the A100 SXM at $0.95 per GPU-hour on demand, $0.000264 a second or $693.50 for a 730-hour month, with 16 vCPU, 125 GB of RAM and a 1 TB local NVMe disk per GPU, in machines of one to eight cards joined by NVLink at 600 GB/s. The price includes the vCPU, the RAM, the local NVMe, one public IPv4 address and all inbound and outbound traffic, and the only other line a bill can carry is a persistent disk at $0.03 per GB-month. Billing is per second with no minimum and starts only once the machine has passed its health check, a failed launch is not billed, a stopped machine costs only its persistent disk, and a host that fails during a job restarts the machine on a healthy host with the same disk and credits the interrupted session in full. The interruptible tier costs $0.57 with five minutes of notice, and a week or a month reserved costs $0.76, prepaid and cancellable before it starts.

That makes GPUBank the cheapest A100 SXM we found that comes as a whole machine, a cent above the cheapest marketplace listing, and its effective hour falls to $0.79 with a 25 dollar top-up and $0.59 with a 1,000 dollar top-up thanks to a balance bonus on credit that never expires. It loses in three places. A Vast.ai interruptible bid of about $0.53 undercuts its $0.57 spot tier for a buyer who accepts being outbid without notice. For throughput-bound jobs its own H100 SXM at $1.85 is often cheaper per finished job, as the measurements further down show. And GPUBank takes only prepaid top-ups in crypto, from 25 to 5,000 dollars, with cards not accepted for now, so a team that must pay by card should look at CloudRift, Thunder Compute or Runpod instead.

### The rest of the market

Beyond the six offers shown as cards, several managed clouds sell a single A100 80 GB at prices between $1.36 and $2.00. Beam lists an A100 80 GB SXM4 machine from $1.36 an hour with 16 vCPU, 128 GB of RAM and 1 TB of NVMe, one of the most generous machines in this price range. Jarvislabs charges $1.49 with 16 vCPU and 112 GB of RAM, billed per minute, without saying SXM or PCIe on its pricing page, and GetDeploying showed Geodd's A100 SXM with 16 vCPU and 120 GB and Sesterce's unlabelled A100 80 GB, both at $1.49 and in stock on 15 October; and Hyperstack lists its A100 SXM at $1.60 on demand and $1.36 reserved, with up to 24 pCPU and 120 GB of RAM, next to a PCIe A100 at $1.35. Verda, formerly DataCrunch, now shows $1.996 per GPU-hour for its A100 SXM4 80 GB in one, two, four and eight-GPU sizes, up from the $1.79 we recorded at the end of September, with 22 CPUs and 120 GB of RAM listed against the card. Above that range, and only by the eight-GPU instance, Crusoe lists an A100 80 GB SXM at $2.30 per GPU-hour next to a PCIe version at $2.00, and Lambda sells its A100 SXM 80 GB as an eight-GPU instance at $2.79 per GPU with 240 vCPU, 1,800 GiB of RAM and 19.5 TiB of SSD.

Marketplace floors also say little about a reliable host: Thunder Compute's review of 9 October priced three Vast.ai A100 80 GB hosts at a median of $1.63 per GPU-hour, between $1.21 and $1.76, with 100 GB of storage counted and bandwidth on top, against the $0.94 floor we found a week later; the same review puts Paperspace's A100 80 GB at $3.18 and CoreWeave's eight-GPU node at $2.70 per GPU. On the marketplace and serverless side, TensorDock advertises A100 80 GB hosts from $1.42 to $1.63 an hour depending on which line of its page you read, and spot from $0.67, on a mix of SXM hosts in Prague and Oregon and PCIe hosts in Devon, so the variant depends on the listing. Lium had no A100 SXM pod available when we checked and showed only a reference price of $0.70, and Runpod's Community Cloud figure of $1.39 appears in its page's structured data but not in the visible table. For per-second work, Koyeb lists an A100 80 GB at $1.60 an hour and an A100 SXM at $2.15, both with 15 vCPU and 180 GB of RAM, Modal charges $0.000694 a second, or $2.50 an hour, with CPU and memory billed on top, and Runpod Serverless charges $2.72 an hour for an A100 80 GB worker. On Runpod's own prices a serverless worker beats a Secure Cloud pod left running only while it is busy less than about 66 percent of the time, $1.79 divided by $2.72.

### The large clouds

AWS

$3.43 /GPU-hr

8-GPU node · p4de.24xlarge, 96 vCPU, 1,152 GiB RAM

Oracle Cloud

$4.00 /GPU-hr

8-GPU node · BM.GPU.A100-v2.8 bare metal

Microsoft Azure

$4.10 /GPU-hr

8-GPU node · ND96amsr A100 v4, East US, Linux

Google Cloud

$5.07 /GPU-hr

1 GPU · a2-ultragpu-1g, 12 vCPU, 170 GB RAM, us-central1

The large clouds sell the A100 80 GB almost only by the eight-GPU instance, and their per-GPU price is the instance price divided by eight. AWS's p4de.24xlarge, with eight 80 GB cards, 96 vCPU and 1,152 GiB of RAM, costs $27.447 an hour on demand in Vantage's default view, or $3.43 per GPU, after the price cut AWS announced for its GPU instances in June 2025; in 2022 AWS's own page listed the same instance at $40.96. The older p4d.24xlarge, at $21.958, carries 40 GB cards and does not belong next to an 80 GB price. Vantage shows the p4de spot price at $20.736 an hour and a three-year reservation at $11.718, or $1.46 per GPU. Microsoft's own Retail Prices API lists the ND96amsr A100 v4, eight 80 GB SXM4 cards, at $32.77 an hour for Linux pay-as-you-go in East US, $4.10 per GPU and unchanged since December 2021, with spot at $8.44 an hour, about $1.06 per GPU, and a three-year reservation that works out to about $1.80 per GPU.

Google Cloud is the one large cloud that rents a single A100 80 GB, as a2-ultragpu-1g with 12 vCPU and 170 GB of RAM, at $5.07 an hour on demand and $3.04 on spot in us-central1, figures we took from gcloud-compute.com's update of 11 October because Google's own price tables did not render for us. Oracle's eight-GPU bare-metal A100 shape costs $32.00 an hour, $4.00 per GPU, read from Economize's update of 8 October because Oracle's price list refused our requests. These clouds also bill outbound traffic, block storage and, on most of them, public addresses on top. You pay their prices for quota-backed capacity, compliance coverage and integration with the rest of their services rather than for the GPU, which is why a team already committed to one of them may still rationally pay three to five times the specialised-cloud rate.

05 / Every offer

## The whole comparison, sorted by kind and price.

| Provider | USD / GPU-hr | Kind | Unit | What the price includes | Billing | Stock | Basis |
| --- | --- | --- | --- | --- | --- | --- | --- |
| GPUBank | $0.95 | Managed on demand | 1 GPU | 16 vCPU, 125 GB RAM, 1 TB NVMe, IPv4, traffic | Per second | In stock | Provider page |
| CloudRift | $1.05 | Managed on demand | 1 GPU | 800 GB local storage per GPU | Per second | Not stated | Provider page |
| Thunder Compute | $1.09 | Managed on demand | 1 GPU | 8 vCPU, 64 GB RAM, 100 GB disk; variant not stated | Per minute | Not stated | Provider page |
| Beam | $1.36 | Managed on demand | 1 GPU | 16 vCPU, 128 GB RAM, 1 TB NVMe | Not stated | Not stated | Provider page |
| Massed Compute | $1.38 | Managed on demand | 1 GPU | 14 vCPU, 100 GB RAM, 625 GB storage | Not stated | Not stated | Provider page |
| Jarvislabs | $1.49 | Managed on demand | 1 GPU | 16 vCPU, 112 GB RAM; variant not stated | Per minute | Not stated | Provider page |
| Hyperstack | $1.60 | Managed on demand | 1 GPU | CPU, RAM and ephemeral NVMe included | Not stated | Not stated | Provider page |
| Runpod | $1.79 | Managed on demand | 1 GPU | 16 vCPU, 125 GB RAM; disks extra | Per second | Not stated | Provider page |
| Verda | $2.00 | Managed on demand | 1 GPU | 22 CPUs, 120 GB RAM | Not stated | Not stated | Provider page |
| Crusoe | $2.30 | Managed on demand | 1 GPU | Machine not detailed on the price list | Not stated | Not stated | Provider page |
| Lambda | $2.79 | Managed on demand | 8-GPU node | 240 vCPU, 1,800 GiB RAM, 19.5 TiB SSD | Not stated | Not stated | Provider page |
| Lium | $0.70 | Marketplace | 1 GPU | Reference price only | Not stated | Sold out | Provider page |
| Vast.ai | $0.937 | Marketplace | 1 GPU | 32 CPU cores; storage $0.33/GB-mo, bandwidth extra | Per second | In stock | Provider page |
| Runpod Community | $1.39 | Marketplace | 1 GPU | Structured data only, not in the visible table | Per second | Not stated | Provider page |
| TensorDock | $1.42 | Marketplace | 1 GPU | Host-dependent; SXM and PCIe hosts mixed | Not stated | Not stated | Provider page |
| Vast.ai | $0.534 | Spot and interruptible | 1 GPU | Minimum bid; can be outbid | Per second | In stock | Provider page |
| GPUBank | $0.57 | Spot and interruptible | 1 GPU | As on demand, 5 min notice, disk kept | Per second | In stock | Provider page |
| TensorDock | $0.67 | Spot and interruptible | 1 GPU | Host-dependent; variant mixed | Not stated | Not stated | Provider page |
| Microsoft Azure | $1.06 | Spot and interruptible | 8-GPU node | ND96amsr A100 v4, East US, Linux | Per second | Not stated | Provider page |
| AWS | $2.59 | Spot and interruptible | 8-GPU node | p4de.24xlarge, Vantage default view | Per second | Not stated | Aggregator |
| Google Cloud | $3.04 | Spot and interruptible | 1 GPU | a2-ultragpu-1g, 12 vCPU, 170 GB RAM, us-central1 | Per second | Not stated | Aggregator |
| GPUBank | $0.76 | Reserved | 1 GPU | Week or month, prepaid | Per term | In stock | Provider page |
| CloudRift | $0.89 | Reserved | 1 GPU | Three months | Per term | Not stated | Provider page |
| CloudRift | $0.95 | Reserved | 1 GPU | One month | Per term | Not stated | Provider page |
| Hyperstack | $1.36 | Reserved | 1 GPU | Reservation | Per term | Not stated | Provider page |
| AWS | $1.46 | Reserved | 8-GPU node | p4de.24xlarge, three years | Per term | Not stated | Aggregator |
| Thunder Compute | $1.49 | Reserved | cluster | 8 to 512+ GPUs | Per term | On request | Provider page |
| Microsoft Azure | $1.80 | Reserved | 8-GPU node | ND96amsr A100 v4, three years, East US | Per term | Not stated | Provider page |
| AWS | $3.43 | Large clouds | 8-GPU node | p4de.24xlarge, 96 vCPU, 1,152 GiB RAM | Per second | Not stated | Aggregator |
| Oracle Cloud | $4.00 | Large clouds | 8-GPU node | BM.GPU.A100-v2.8 bare metal | Not stated | Not stated | Aggregator |
| Microsoft Azure | $4.10 | Large clouds | 8-GPU node | ND96amsr A100 v4, East US, Linux | Per second | Not stated | Provider page |
| Google Cloud | $5.07 | Large clouds | 1 GPU | a2-ultragpu-1g, 12 vCPU, 170 GB RAM, us-central1 | Per second | Not stated | Aggregator |
| Koyeb | $1.60 | Serverless | 1 GPU | A100 80 GB, 15 vCPU, 180 GB RAM; variant not stated | Per second | Not stated | Provider page |
| Koyeb | $2.15 | Serverless | 1 GPU | A100 SXM, 15 vCPU, 180 GB RAM | Per second | Not stated | Provider page |
| Modal | $2.50 | Serverless | 1 GPU | GPU only, CPU and RAM extra; variant not stated | Per second | Not stated | Provider page |
| Runpod Serverless | $2.72 | Serverless | 1 GPU | A100 80 GB worker | Per second | Not stated | Provider page |

"Aggregator" rows could not be read on the provider's own page that day; "Provider snapshot" rows come from a page dated by the provider.

06 / The real monthly bill

## What an A100 hour really costs

Storage is the line that bills while nothing runs. A 200 GB persistent disk costs $6.00 a month on GPUBank at $0.03 per GB-month and $14.00 on Runpod's network storage at $0.07, while a Runpod volume disk costs $0.10 per GB-month while the pod runs and $0.20 while it sits idle. Thunder Compute includes 100 GB of disk and charges $0.219 per GB-month beyond it, so its 200 GB would add about $21.90. On Vast.ai each host sets its own storage rate and Vast bills storage for as long as the instance exists, running or stopped; the cheapest 80 GB host on the day charged about $0.33 per GB-month, so the same 200 GB kept all month would cost about $66.60, eleven times the GPUBank disk, and its bandwidth was priced at $0.004 per GB. A stopped marketplace instance also keeps its data but not its claim on the GPU, so if another renter takes the card you either wait or move your data to another host.

Billing granularity matters most for short jobs. A job of 61 minutes on a per-second meter at $0.95 costs $0.97, while a provider that bills each started hour charges two hours, $1.90 at the same rate; Thunder Compute and Jarvislabs bill per minute, CloudRift, Runpod and GPUBank per second. On an eight-GPU-only cloud the unit itself is the cost: a quick test on Lambda's A100 80 GB instance bills $22.32 an hour for all eight cards even if the code uses one. Prepaid models add their own terms: GPUBank's balance never expires and its top-up bonus lowers the effective hour, but the money is committed up front, so the bonus is worth having only for credit you will actually spend.

The last cost is lost time, and the question is who pays for it. On GPUBank a launch that fails its health check is not billed and a host failure during a job is credited for the whole interrupted session. On a marketplace a host that disappears ends your session and the work since the last checkpoint is yours to lose, which is why Vast.ai publishes a reliability score for every listing and why the best-scored hosts are rarely the cheapest. A reclaimed spot instance costs whatever was done since the last checkpoint: with GPUBank's five-minute notice and checkpoint hook that can be close to nothing, while an outbid marketplace instance with no checkpointing can cost the whole run.

### Three workloads, priced for a month

**GPUBank interruptible**

40 h GPU $22.80

Storage $6.00

Other None

Month $28.80

Runs checkpoint; no reclaim losses counted

All-in

**CloudRift**

40 h GPU $42.00

Storage $0.00

Other Persistent storage not priced

Month $42.00

Data re-uploaded to the 800 GB local disk each run

**GPUBank**

40 h GPU $38.00

Storage $6.00

Other None

Month $44.00

On demand, list price

All-in

**Runpod Secure Cloud**

40 h GPU $71.60

Storage $14.00

Other None

Month $85.60

Network storage at $0.07/GB-mo

**Vast.ai**

40 h GPU $37.48

Storage $66.60

Other Bandwidth $0.004/GB

Month $104.08

Iowa host, instance kept all month to hold the disk

GPUBank's interruptible tier wins at $28.80, 40 hours at $0.57 plus $6.00 of disk, as long as each run writes checkpoints. CloudRift's $42.00 looks slightly cheaper than GPUBank's on-demand $44.00, $38.00 of GPU time plus $6.00 of disk, but only because it publishes no price for storage that survives an instance, so the $42.00 holds only if you re-upload 200 GB before every run. Runpod costs $71.60 plus $14.00 of network storage, $85.60. The Vast.ai host that is cheapest per hour becomes the most expensive here: $37.48 of GPU time is followed by about $66.60 of storage, because Vast bills the disk for as long as the instance exists and this host charges $0.333 per GB-month. With a 100 dollar top-up, GPUBank's on-demand bill falls to about $32.59 of cash.

07 / Measured per job

## Performance, measured, and the right tool

The A100 SXM is the sensible rental when a job needs 80 GB of fast memory but not FP8, when it is memory-bound or spends much of its time idle rather than saturating the tensor cores, or when the software stack predates Hopper: QLoRA and LoRA fine-tuning of 7B to 70B models, BF16 batch inference, notebooks left open for hours and MIG slices for small services. It is the wrong tool for FP8 serving at high concurrency, for pretraining at scale and for any long throughput-bound job where the H100's speed repays its higher rate. The measured gap depends heavily on the workload. In MLPerf Training v2.1, eight A100 SXM 80 GB took 16.88 minutes to train BERT-large against 6.38 minutes for eight H100 SXM, a ratio of 2.65, while ResNet-50 took 28.69 against 14.68 minutes, a ratio of only 1.95. The H100's BERT entry used FP8 through the Transformer Engine, which is part of why the gap is wider there.

Inference shows the same pattern. E2E Networks' TensorRT-LLM offline benchmark of Llama 3.1 8B on one GPU gives 6,051.62 tokens per second on the A100 against 15,861.63 on the H100 with 128 input and 128 output tokens, a ratio of 2.62, but only 3,747.46 against 7,425.07 with 1,000 input and 1,000 output tokens, a ratio of 1.98. That page does not state the precision, the software version or the A100's form factor, and it shows a third, unlabelled H100 column at 27,006.16 tokens per second that may be an FP8 run, so we use the like-for-like column and treat the figures as indicative. The results therefore agree on direction and disagree on size: the H100 is about twice as fast on bandwidth-limited or long-sequence work and two and a half to nearly three times as fast on short, compute-heavy work, and more again when it runs FP8 and the A100 cannot.

**BERT-large training, 8 GPUs** MLPerf Training v2.1 closed, NVIDIA DGX A100 80 GB vs DGX H100 (preview), minutes to target accuracy

A100 SXM 80 GB

16.9 min

H100 SXM

6.38 min

MLCommons v2.1 NVIDIA run logs, scored with the MLPerf rule (best and worst runs dropped); EE Times reported 16.8 and 6.4 minutes. H100 used FP8.

**ResNet-50 training, 8 GPUs** MLPerf Training v2.1 closed, NVIDIA DGX A100 80 GB vs DGX H100 (preview), minutes to target accuracy

A100 SXM 80 GB

28.7 min

H100 SXM

14.7 min

MLCommons v2.1 NVIDIA run logs, scored with the MLPerf rule; EE Times gave 28.2 minutes for the closest 8x A100 score.

**Llama 3.1 8B serving, 1 GPU** E2E Networks, TensorRT-LLM offline throughput, 128 input and 128 output tokens

A100

6,052 tok/s

H100

15,862 tok/s

E2E Networks TIR inference benchmarks, page updated 15 May 2026; precision, version and form factor not stated.

WHOLE MACHINE

$0.95

GPUBank on demand, CPU, RAM, NVMe, IPv4 and traffic included

MARKET MEDIAN

$1.88

GetDeploying, 39 priced providers, 15 Oct 2026

BERT RUN, 8 GPUS

$2.14

on A100s against $1.57 on H100s at GPUBank rates

08 / H100 or the others

## Per job against the other cards

On GPUBank the H100 SXM costs $1.85 an hour against $0.95 for the A100 SXM, 1.95 times as much, so the H100 is cheaper per job whenever it runs more than 1.95 times faster. On the measurements above that threshold is crossed on BERT, where a training run costs $2.14 on eight A100s and $1.57 on eight H100s, and on short-sequence LLM serving, where a million tokens cost $0.044 on the A100 and $0.032 on the H100. It is not crossed on ResNet-50, at $3.63 against $3.62, or on long-sequence serving, at $0.070 against $0.069 per million tokens, where the two cards tie. The top-up bonus and the interruptible tier apply the same factor to every card, so neither changes this ranking, and the result holds as long as the job keeps the GPU busy; idle time is billed at the hourly rate on both cards, which favours the cheaper one.

We found no measurement from a named source that ran the H100 PCIe, the L40S or the RTX PRO 6000 against the A100 SXM on the same test, so we do not compute a cost per job for them. From the specifications alone, the L40S at $0.55 has 48 GB of GDDR6 at 864 GB/s, less than half the A100's bandwidth, so it is unlikely to win on memory-bound decoding, but it has FP8 and costs 58 percent of the A100's rate for models that fit in 48 GB. The RTX PRO 6000 at $1.15 has 96 GB at 1.8 TB/s, more memory for 1.21 times the price, and the H100 PCIe at $1.35 has the same 80 GB of HBM2e at 2.0 TB/s with Hopper's FP8 but no NVLink between cards.

L40S

**48 GB** GDDR6 ECC · 864 GB/s

$0.55 /hr

Stable Diffusion, video, Whisper transcription and embeddings

A100 SXM

**80 GB** HBM2e · 2.0 TB/s

$0.95 /hr

LoRA and QLoRA fine-tuning, batch inference

RTX PRO 6000

**96 GB** GDDR7 ECC · 1.8 TB/s

$1.15 /hr

Serving 70B models in 4-bit and AI video generation

H100 PCIe

**80 GB** HBM2e · 2.0 TB/s

$1.35 /hr

Single-GPU training and inference on a budget

H100 SXM

**80 GB** HBM3 · 3.35 TB/s

$1.85 /hr

AI training, fine-tuning and production LLM inference

GPUBank on-demand prices, all-in. Memory bar scaled to 96 GB.

09 / Rent or buy

## Rent or buy an A100 SXM

The realistic purchase unit is a whole eight-GPU HGX system, because the SXM4 module has no PCIe connector and needs a baseboard with its power delivery and cooling. AmCompute values a used HGX A100 node with eight 80 GB cards at $95,200 in October 2026 and reports realised sales of single SXM4 80 GB modules at about $4,700 to $5,400 over the twelve months to August 2026, down about 35 percent on the year before. Asking prices are higher: Caladan Semi recorded a median ask of $9,925 from only four US listings on 9 September, and Jarvislabs gives $5,000 to $9,000 for a used module and $80,000 to $120,000 for a used DGX A100. We use $95,200 for the server and assume that it keeps losing about 35 percent of its value a year, which leaves roughly $40,000 of resale value after two years.

Depreciation is then $95,200 minus $40,000 divided by 24 months, $2,300 a month for the whole server or $287.50 per GPU. Power at an assumed 5 kW under load, below the 6.5 kW maximum NVIDIA gives for a DGX A100, at the US average of 18.19 cents per kWh adds $0.91 per machine-hour, about $0.11 per GPU-hour. Against GPUBank's $0.95 on-demand rate the break-even is $287.50 divided by $0.84, about 344 hours per GPU per month or 47 percent utilisation; against the $0.76 reserved rate it is 445 hours, and against the $0.59 effective rate with 1,000 dollar top-ups it is 599 hours, 82 percent. Rack space, cooling, networking, spare modules and the engineer who keeps the box running are not counted, and all of them push the break-even higher, so buying makes sense only for a team that already runs a data-centre rack and can keep eight cards busy most of every day for two years.

Monthly cost of one used 8-GPU HGX A100 80 GB server (8 GPUs) against hours of use a month, 0 to 730. Owning: price − $40,000 resale over 24 months, plus 5 kW at the wall under load. Rent: per GPU-hour × 8.

| Hours a month | Rent at $0.95 | Rent at $0.59 effective | Own | Cheaper |
| --- | --- | --- | --- | --- |
| 100 | $760.00 | $475.00 | $2,390.95 | Rent |
| 250 | $1,900.00 | $1,187.50 | $2,527.38 | Rent |
| 344 | $2,614.40 | $1,634.00 | $2,612.87 | Even at list price; rent with the bonus |
| 445 | $3,382.00 | $2,113.75 | $2,704.73 | Own beats list and reserved; rent with the bonus |
| 600 | $4,560.00 | $2,850.00 | $2,845.70 | Even with the bonus; own at list |
| 730 | $5,548.00 | $3,467.50 | $2,963.93 | Own |

AmCompute valuation of a used HGX A100 8x 80 GB node, October 2026

10 / What to choose

## The cheapest A100 depends on the job.

Fine-tuning a 7B to 70B model with LoRA or QLoRA

A100 SXM, interruptible if your runs checkpoint

80 GB fits a 70B model at 4-bit with room for context, and these runs spend enough time on data and checkpoints that the H100's speed is not fully used.

Serving an LLM at high concurrency or in FP8

H100 SXM

It was 2.6 times faster on short-sequence serving in E2E's test, so at 1.95 times the price it costs about a quarter less per million tokens.

Notebooks, debugging and legacy CUDA stacks

A100 SXM on demand

Idle time is billed at the hourly rate, so the cheaper card wins, and Ampere-era code runs without porting work.

Multi-GPU training for a few days

8x A100 SXM, or 8x H100 for transformer training

On MLPerf the A100 tied the H100 per job on ResNet-50 and lost on BERT; per-second billing on an NVLink machine avoids paying for an idle week.

11 / Questions

## What people ask before renting one.

The cheapest A100 SXM 80 GB hour on 16 October 2026 was an interruptible bid of about $0.53 on a Vast.ai host in Iowa, and the cheapest on-demand listing was the same host at $0.937 per GPU-hour. Those are marketplace prices: storage, at $0.333 per GB-month on that host, and bandwidth are billed separately, and the machine and its reliability depend on the host. The cheapest price for a whole managed machine was GPUBank's $0.95, with 16 vCPU, 125 GB of RAM, a 1 TB NVMe disk, an IPv4 address and traffic included, followed by CloudRift at $1.05 and Thunder Compute at $1.09. GPUBank's interruptible tier costs $0.57 and its top-up bonus brings the effective on-demand hour down to $0.59. Many cheaper-looking rows on aggregators are 40 GB or PCIe cards, so check the variant first.

Sources

## Sources

1. 01 [GPUBank pricing](https://gpubank.io/pricing) GPUBank A100 SXM rates in three modes, inclusions, disk price, billing rules, top-up bonuses, other cards · read 16 Oct 2026
2. 02 [CloudRift pricing](https://www.cloudrift.ai/pricing) A100 SXM4 on demand, one-month and three-month rates, 800 GB storage · read 16 Oct 2026
3. 03 [Thunder Compute pricing](https://www.thundercompute.com/pricing) A100 80GB rate, included machine, per-minute billing, reserved rate · read 16 Oct 2026
4. 04 [Vast.ai live offers API, A100 SXM4 search](https://console.vast.ai/api/v0/bundles/) Live 80 GB offers, on-demand prices, minimum bids, storage and bandwidth rates · read 16 Oct 2026
5. 05 [Runpod pricing (updated 8 Oct 2026)](https://www.runpod.io/pricing) Secure Cloud, Community and Serverless A100 rates, storage prices · read 16 Oct 2026
6. 06 [Lambda pricing](https://lambda.ai/pricing) 8x A100 SXM 80 GB instance, single-GPU A100 40 GB · read 16 Oct 2026
7. 07 [Vantage, p4de.24xlarge](https://instances.vantage.sh/aws/ec2/p4de.24xlarge) AWS on-demand, spot and reserved prices (aggregator) · read 16 Oct 2026
8. 08 [gcloud-compute.com, a2-ultragpu-1g and 8g (11 Oct 2026)](https://gcloud-compute.com/a2-ultragpu-1g.html) Google Cloud on-demand and spot prices (aggregator) · read 16 Oct 2026
9. 09 [GetDeploying, Nvidia A100 rental prices](https://getdeploying.com/reference/cloud-gpu/nvidia-a100) Market median across 39 priced providers, 15 Oct 2026 · read 16 Oct 2026
10. 10 [NVIDIA A100 Tensor Core GPU](https://www.nvidia.com/en-us/data-center/a100/) Specifications · read 16 Oct 2026
11. 11 [AWS, EC2 P4 instances](https://aws.amazon.com/ec2/instance-types/p4/) p4d and p4de specifications · read 16 Oct 2026
12. 12 [Jarvislabs pricing](https://jarvislabs.ai/pricing) A100 80GB rate and machine · read 16 Oct 2026
13. 13 [Koyeb pricing](https://www.koyeb.com/pricing) A100 80GB and A100 SXM serverless rates · read 16 Oct 2026
14. 14 [Modal pricing](https://modal.com/pricing) A100 80 GB per-second rate · read 16 Oct 2026
15. 15 [Dettmers et al., QLoRA (arXiv 2305.14314)](https://arxiv.org/abs/2305.14314) 65B model fine-tuned on a single 48 GB GPU · read 16 Oct 2026
16. 16 [Beam pricing](https://www.beam.cloud/pricing) A100 80 GB SXM4 machine from $1.36 · read 16 Oct 2026
17. 17 [Massed Compute pricing](https://vm.massedcompute.com/pricing) A100 SXM4 80 GB, 1 and 8 GPUs · read 16 Oct 2026
18. 18 [Hyperstack, A100 SXM](https://www.hyperstack.cloud/a100-sxm) On-demand and reserved rates · read 16 Oct 2026
19. 19 [Verda GPU instances](https://verda.com/gpu-instances) A100 SXM4 80GB rate and machine · read 16 Oct 2026
20. 20 [Azure Retail Prices API, Standard_ND96amsr_A100_v4, eastus](https://prices.azure.com/api/retail/prices?$filter=armSkuName%20eq%20%27Standard_ND96amsr_A100_v4%27%20and%20armRegionName%20eq%20%27eastus%27) Pay-as-you-go, spot and reservation prices · read 16 Oct 2026
21. 21 [Crusoe pricing](https://crusoe.ai/pricing) A100 80GB SXM and PCIe rates · read 16 Oct 2026
22. 22 [Vast.ai, Rent A100 SXM4 GPUs](https://vast.ai/pricing/gpu/A100-SXM4) Static page showing no current offers · read 16 Oct 2026
23. 23 [Lium, A100](https://lium.io/gpu/a100) Reference price, no pod available · read 16 Oct 2026
24. 24 [Thunder Compute, NVIDIA A100 pricing (reviewed 9 Oct 2026)](https://www.thundercompute.com/blog/nvidia-a100-pricing) Vast.ai A100 80 GB median $1.63 over three hosts ($1.21 to $1.76); Paperspace $3.18; CoreWeave $2.70 per GPU · read 15 Oct 2026
25. 25 [TensorDock, A100](https://www.tensordock.com/gpu-a100.html) On-demand and spot starting prices, host locations · read 16 Oct 2026
26. 26 [AWS News Blog, price reduction for NVIDIA GPU instances (5 Jun 2025)](https://aws.amazon.com/blogs/aws/announcing-up-to-45-price-reduction-for-amazon-ec2-nvidia-gpu-accelerated-instances/) June 2025 price cut · read 16 Oct 2026
27. 27 [AWS P4 page, Wayback Machine, 5 Aug 2022](https://web.archive.org/web/20220805235553/https://aws.amazon.com/ec2/instance-types/p4/) p4de.24xlarge at $40.96 an hour in 2022 · read 16 Oct 2026
28. 28 [Economize, OCI BM.GPU.A100 (8 Oct 2026)](https://www.economize.cloud/resources/oci/pricing/compute/BM_GPU_A100_8/) Oracle bare-metal A100 price (aggregator) · read 16 Oct 2026
29. 29 [Vast.ai documentation, Billing](https://docs.vast.ai/documentation/reference/billing) Storage billed while stopped, bandwidth per byte · read 15 Oct 2026
30. 30 [MLCommons, MLPerf Training v2.1 results (NVIDIA run logs)](https://github.com/mlcommons/training_results_v2.1) BERT and ResNet-50 time to train, DGX A100 80 GB and DGX H100 · read 16 Oct 2026
31. 31 [EE Times, Nvidia shows off Hopper MLPerf training benchmarks (29 Nov 2022)](https://www.eetimes.com/nvidia-shows-off-hopper-mlperf-training-benchmarks/) BERT 6.4 min on 8x H100, 16.8 and 28.2 min on 8x A100 · read 16 Oct 2026
32. 32 [NVIDIA Technical Blog, Leading MLPerf Training 2.1 (9 Nov 2022)](https://developer.nvidia.com/blog/leading-mlperf-training-2-1-with-full-stack-optimizations-for-ai/) FP8 Transformer Engine in the H100 BERT submission · read 16 Oct 2026
33. 33 [E2E Networks, TIR inference benchmarks (updated 15 May 2026)](https://docs.e2enetworks.com/docs/tir/benchmarks/inference_benchmarks/) Llama 3.1 8B TensorRT-LLM throughput, A100 and H100 · read 16 Oct 2026
34. 34 [AmCompute, A100 vs H100](https://www.amcompute.com/compare/a100-vs-h100) Used HGX A100 8x 80 GB node valuation · read 16 Oct 2026
35. 35 [AmCompute, A100 SXM4 80GB appraisal](https://www.amcompute.com/appraisals/a100-sxm4-80gb) Realised used sale prices to August 2026 · read 16 Oct 2026
36. 36 [Caladan Semi, used H100 and A100 buying guide 2026](https://caladansemi.com/blog/used-nvidia-h100-a100-gpu-buying-guide-2026) US asking prices on 9 Sep 2026 · read 16 Oct 2026
37. 37 [Jarvislabs, NVIDIA A100 price in 2026](https://jarvislabs.ai/blog/a100-price) New and used module and DGX prices · read 16 Oct 2026
38. 38 [NVIDIA DGX A100 datasheet (Colfax copy)](https://www.colfax-intl.com/downloads/nvidia-dgx-a100-datasheet.pdf) 6.5 kW maximum system power · read 16 Oct 2026
39. 39 [EIA, Electric Power Monthly, table 5.3](https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=epmt_5_3) US average electricity price used in the buy calculation · read 15 Oct 2026
40. 40 [Lambda GPU cloud, Wayback Machine, 12 May 2023](https://web.archive.org/web/20230512094726/https://lambdalabs.com/service/gpu-cloud) 8x A100 80 GB at $1.50 per GPU · read 16 Oct 2026
41. 41 [Lambda GPU cloud, Wayback Machine, 16 Mar 2025](https://web.archive.org/web/20250316063514/https://lambdalabs.com/service/gpu-cloud) 8x A100 SXM 80 GB at $1.79 per GPU · read 16 Oct 2026
42. 42 [Runpod pricing, Wayback Machine, 9 Jul 2023](https://web.archive.org/web/20230709173002/https://runpod.io/gpu-instance/pricing) A100 SXM 80 GB at $1.84 · read 16 Oct 2026
43. 43 [Runpod pricing, Wayback Machine, 13 Jun 2024](https://web.archive.org/web/20240613104312/https://runpod.io/gpu-instance/pricing) A100 SXM 80 GB at $2.29 · read 16 Oct 2026
44. 44 [CNBC, Nvidia's A100 is the $10,000 chip powering the race for AI (23 Feb 2023)](https://www.cnbc.com/2023/02/23/nvidias-a100-is-the-10000-chip-powering-the-race-for-ai-.html) New price around $10,000 · read 16 Oct 2026
45. 45 [NVIDIA press release, DGX A100 (14 May 2020)](https://nvidianews.nvidia.com/news/nvidia-ships-worlds-most-advanced-ai-system-nvidia-dgx-a100-to-fight-covid-19-third-generation-dgx-packs-record-5-petaflops-of-ai-performance) DGX A100 from $199,000 · read 16 Oct 2026

What we could not confirm

Several figures rest on aggregators because the provider's own page could not be read: AWS prices come from Vantage (the region of its default view is not shown, and AWS's page no longer lists prices), Google Cloud from gcloud-compute.com and Oracle from Economize, since Google's price tables did not render and Oracle refused our requests. Vast.ai's public price page showed no offers and a blank headline, so its figures come from the live offers API at one moment on 16 October, and marketplace prices move by the hour. Runpod's Community Cloud A100 SXM at $1.39 appears only in the page's structured data, Crusoe's page did not make clear whether $2.30 is the on-demand column, TensorDock's page quotes three different starting prices, and Lium had no pod to price. Jarvislabs no longer shows the $0.89 spot A100 the research pass reported, Thunder Compute shows no $0.79 tile, and Verda's spot A100 price is not displayed. The MLPerf minutes were computed from MLCommons run logs with the MLPerf scoring rule and match EE Times for BERT, but EE Times gives 28.2 rather than 28.69 minutes for the closest 8x A100 ResNet-50 score; E2E's benchmark page does not state precision, software version or form factor, and Microsoft's Azure vLLM benchmark could not be read. eBay listings for used modules and servers returned errors, so purchase prices rest on AmCompute, Caladan Semi and Jarvislabs; the 35 percent annual decline, the 5 kW draw and the electricity price in the buy calculation are assumptions. We found no measured comparison of the A100 SXM with the L40S, the RTX PRO 6000 or the H100 PCIe on a common test, and could not recheck the September figures for LeaderGPU, Prime Intellect, io.net, Denvr, Gcore, Oblivus, Spheron, Civo, E2E Networks, Vultr, OVHcloud, Paperspace, CoreWeave, Akash and Clore.ai, which are therefore left out.

A100 SXM on GPUBank

## Eighty gigabytes of Ampere, billed by the second.

- 16 vCPU, 125 GB RAM, 1 TB NVMe, IPv4 and traffic in $0.95/hr
- Billed by the second after the health check
- Interruptible $0.57, reserved $0.76

[Launch an A100 SXM · $0.95/hr](https://console.gpubank.io/launch?gpu=a100-sxm)

[Compare 63 GPU clouds](https://gpubank.io/compare)

More GPU price guides

## Where to rent the other cards at the cheapest price.

[B200 SXM 180 GB HBM3e From $2.49 spot · $4.15 all-in](https://gpubank.io/rent-b200)

[H200 SXM 141 GB HBM3e From $1.49 spot · $2.49 all-in](https://gpubank.io/rent-h200)

[H100 SXM 80 GB HBM3 From $1.11 spot · $1.85 all-in](https://gpubank.io/rent-h100)

[H100 PCIe 80 GB HBM2e From $0.81 spot · $1.35 all-in](https://gpubank.io/rent-h100-pcie)

[MI300X 192 GB HBM3 From $0.99 spot · $1.65 all-in](https://gpubank.io/rent-mi300x)

[RTX PRO 6000 96 GB GDDR7 ECC From $0.69 spot · $1.15 all-in](https://gpubank.io/rent-rtx-pro-6000)

[L40S 48 GB GDDR6 ECC From $0.33 spot · $0.55 all-in](https://gpubank.io/rent-l40s)

[RTX 5090 32 GB GDDR7 From $0.29 spot · $0.48 all-in](https://gpubank.io/rent-rtx-5090)

[RTX 4090 24 GB GDDR6X From $0.14 spot · $0.23 all-in](https://gpubank.io/rent-rtx-4090)