--- title: Where to Rent an H100 PCIe at the Cheapest Price (Oct 2026) url: https://gpubank.io/rent-h100-pcie description: H100 PCIe from $1.35/hr with the whole machine included, against $1.83 at Beam, $2.49 Civo, $2.89 Runpod and $3.29 Lambda. Spot, reserved, SXM vs PCIe. generated: 2026-11-02T00:21Z 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-h100-pcie, 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 · 31 offers checked # Where to rent an H100 PCIe at the cheapest price We read every H100 PCIe offer we could find on the provider's own page, kept the PCIe, NVL and SXM variants apart, and priced what each hour includes, what a month costs and the point at which buying the card would pay for itself. **GPUBank Research** 17 min read · Prices read on 16 Oct 2026 **80 GB**HBM2e **2.0 TB/s**memory bandwidth **350 W**passive, PCIe Gen5 **756 TFLOPS**BF16 dense The short answer, 16 Oct 2026 1. $1.35 /hr **Cheapest on-demand H100 PCIe.** GPUBank at $1.35 an hour, billed per second, with 16 vCPU, 188 GB of RAM, a 1 TB NVMe disk, a public IPv4 address and all traffic in the price; a week or a month reserved costs $1.08. The next managed offers we could read at source were Beam from $1.83, Civo at $2.49 and Hyperstack at $2.50. 2. $0.81 /hr **Cheapest interruptible hour.** GPUBank's interruptible tier at $0.81 was the lowest rate we could actually book on 16 October, with five minutes of notice and the machine stopped rather than deleted. Vast.ai bids have gone lower, down to a two-card listing at $0.67 per GPU on 15 October, but its H100 PCIe page showed no offers at all the next day. 3. $1.98 /hr **Cheapest marketplace list price.** GPU.ai at $1.98 and Runpod Community Cloud at $1.99, both billed per second on partner or peer hosts with the disk billed on top, but both shown out of stock on 15 October. The Vast.ai hosts tracked at about $2.00 that day had gone by the next morning. [H100 PCIe $1.35 /hr all-in Launch an H100 PCIe Spot $0.81 · reserved $1.08 Billed per second](https://console.gpubank.io/launch?gpu=h100-pcie) The H100 PCIe is the cheapest way to rent 80 GB of Hopper memory on a single card, and that is why it is sold one GPU at a time by so many specialised clouds. It is also the variant that price lists describe most loosely. A row that says only 'H100' can mean this PCIe card, the 94 GB H100 NVL or an SXM module inside an eight-GPU HGX server, three products with different memory, bandwidth and throughput, and a price comparison that mixes them compares nothing. On this page every row says which variant it is, and where the provider does not say, we say that too. The posted prices for an H100 PCIe hour ran from $0.81 to $3.50 at managed and marketplace providers when we checked on 15 and 16 October 2026, and the spread is mostly about what the hour buys rather than the silicon. Some prices cover the GPU alone, some include a full machine with disk, an IP address and traffic, some can be taken back with a few minutes of notice, and some require a three-year commitment. Stock was thin: Runpod, Lambda, GPU.ai and Oblivus were all shown out of stock, and Vast.ai's own page listed no offers on 16 October. GPUBank publishes this page and rents the H100 PCIe itself, so read our own entry with that in mind. It is measured on the same terms as every other offer, and the text says plainly where it loses: interruptible marketplace bids have undercut it, its own H100 SXM often costs less per finished job, and its payment runs only on crypto top-ups for now. Prices were read on the provider's page unless a row says it comes from an aggregator or from a provider's dated snapshot. 01 / Three H100s, one name ## Three H100s, one name on the price list NVIDIA sells three H100 boards and they are not interchangeable. The SXM module sits on an eight-GPU HGX board with 80 GB of HBM3 at 3.35 TB/s, up to 700 W and 900 GB/s of NVLink through NVSwitch; the NVL is a dual-slot PCIe card with 94 GB of HBM3 at 3.9 TB/s and 350 to 400 W; the PCIe card has 80 GB of HBM2e at 2.0 TB/s, 350 W and an NVLink bridge to one partner card at 600 GB/s. In dense FP8 that is about 1,979 TFLOPS for the SXM, 1,671 for the NVL and 1,513 for the PCIe card, half the sparse figures NVIDIA prints. The PCIe card loses 40 percent of the SXM's bandwidth and half its power budget, which is why measured throughput falls by about a third, and it cannot scale past two cards on NVLink. That is why it matters when a listing says only 'H100'. On the same Hyperstack page the on-demand row reads 'NVIDIA H100' and the spot row 'NVIDIA H100 PCIe', and a third row called 'NVIDIA H100 NVLink' is a machine of bridged PCIe cards, not the NVL. Koyeb, HyperAI and fal.ai print 'H100' with no variant at all, and Massed Compute's 'H100 (80GB)' is the PCIe card only by elimination. A buyer who compares a $2.50 unlabelled row with a $3.20 SXM row, or pays an SXM price for a PCIe card, loses about a third of the throughput. Ask the provider, or run nvidia-smi in the first minute: it reports 'NVIDIA H100 PCIe', 'NVIDIA H100 NVL' or, for the SXM, 'NVIDIA H100 80GB HBM3'. SXM5 module on an 8-GPU HGX board H100 SXM Memory **80 GB HBM3** Bandwidth **3.35 TB/s** BF16, dense **989 TFLOPS** FP8, dense **1,979 TFLOPS** Board power **700 W** NVLink 900 GB/s via NVSwitch; PCIe Gen5 128 GB/s Cheapest today **$1.39 /hr** *Lium (marketplace, Secaucus)* Dual-slot FHFL PCIe card, passive H100 PCIe This page Memory **80 GB HBM2e** Bandwidth **2.0 TB/s** BF16, dense **756 TFLOPS** FP8, dense **1,513 TFLOPS** Board power **350 W** NVLink bridge to one other card, 600 GB/s; PCIe Gen5 128 GB/s Cheapest today **$1.35 /hr** *GPUBank (managed)* Dual-slot PCIe card, air-cooled H100 NVL Memory **94 GB HBM3** Bandwidth **3.9 TB/s** BF16, dense **835 TFLOPS** FP8, dense **1,671 TFLOPS** Board power **400 W** NVLink bridge, 600 GB/s; PCIe Gen5 128 GB/s Cheapest today **$2.59 /hr** *Runpod Community Cloud* Tensor-core figures are dense: NVIDIA prints them with structured sparsity, and we halved them as its footnote instructs. ### Listings that do not say which H100 Hyperstack NVIDIA H100 · $2.50 per hour The spot table on the same page calls the same card 'NVIDIA H100 PCIe', and a third row, 'NVIDIA H100 NVLink' at $2.60, is bridged PCIe cards rather than the 94 GB NVL. Massed Compute H100 (80GB) · $2.73/hr No variant on the price list; separate H100 NVL and H100 SXM5 rows make it the PCIe card by elimination, and trackers list it as PCIe. Koyeb H100 · 80GB VRAM · $2.50 /hr Neither PCIe, SXM nor NVL appears anywhere on the pricing page, while its A100 row is labelled SXM. HyperAI H100 · $1.80/h The visible text gives no variant; only the link behind the card points to an H100 PCIe 80 GB page. fal.ai H100 · 80GB · $4.50/h No variant stated; the page also advertises the H100 'as low as $2.49/hr' without saying on which terms. Lium NVIDIA H100 80GB HBM3 · from $1.39 The marketplace uses the name nvidia-smi reports, so only the memory type tells you this is the SXM module; the PCIe card appears as 'NVIDIA H100 PCIe'. 02 / The card ## What the H100 PCIe is, and who rents it The H100 PCIe is NVIDIA's GH100 Hopper GPU on a full-height, full-length, dual-slot PCI Express Gen5 card with a passive heatsink, built to slot into an ordinary air-cooled server. NVIDIA's product brief gives it 80 GB of HBM2e on a 5,120-bit bus at 2,000 GB/s, a total board power of 350 W, both as default and as maximum, and three NVLink bridge connectors that join two cards at 600 GB/s. NVIDIA's datasheet lists 1,513 TFLOPS of BF16 and 3,026 TFLOPS of FP8 on the tensor cores, but those figures are shown with structured sparsity and the footnote says they are half without it, so the dense numbers that matter for most real work are about 756 and 1,513 TFLOPS; the same sheet gives 26 TFLOPS of FP64 and 51 of FP32. The brief also sets a 200 W floor for the power limit and a boost clock of 1,755 MHz, which is why the card holds its clocks well in an ordinary air-cooled chassis. The same datasheet calls the memory HBM3, which contradicts the product brief's HBM2e; the brief, the board partners and the bandwidth figure all point to HBM2e. The people who rent this card are mostly doing single-GPU work where 48 GB is not enough: serving a 30B to 70B model quantised to fit one card, LoRA and QLoRA fine-tuning, notebooks and research that want FP8 and the Transformer Engine, and batch jobs that never span more than one GPU. GPUBank positions it for single-GPU training and inference on a budget, which is a fair summary. The price range is wide because the card is sold by everything from peer-to-peer hosts to Lambda, because many listings bundle a whole machine and many others sell the GPU alone, and because demand for single Hopper cards shifts by the week while supply sits in small fleets that sell out quickly. 03 / The price ladder ## How to read the H100 PCIe market Five kinds of offer sell this card, and they are not the same product even when the price is close. Managed on-demand clouds such as GPUBank, Beam, Civo, Hyperstack, Runpod Secure Cloud and Lambda rent a fixed machine with stated vCPU, RAM and disk at one hourly price you can start and stop at will. Marketplaces such as Vast.ai, Runpod Community Cloud and GPU.ai pass through a host's own price, and the CPU, memory, disk and network around the card change from one listing to the next, with disk and bandwidth usually billed on top of the headline. Spot or interruptible capacity is the same hardware at a discount that can be taken back. Reserved capacity is where the terms vary most. GPUBank reserves a single card for a week or a month at $1.08, Hyperstack shows reservations 'starting from' $1.75 without stating the term, and Civo's $1.99 and Oblivus's $2.56 both require 36 months. Longer terms buy less than you might expect: ComputePrices listed Impossible Cloud at $2.09 per GPU-hour on 29 September, but for a whole eight-GPU server held for 12 months. Serverless is the fifth kind: Beam bills the H100 PCIe at $0.000972 per second of GPU time, about $3.50 an hour, with CPU and memory charged separately, which suits spiky traffic and costs too much for a model that serves all day. The comparison table keeps these groups apart, because a marketplace headline at $2.00 and a managed all-in price at $2.00 buy different things. *Average of 17 clouds, 29 Sep $2.71* **Managed on demand** A fixed machine at one hourly price. *GPUBank**$1.35*** *Beam**$1.83*** *Civo**$2.49*** *Hyperstack**$2.50*** *QuantaCloud**$2.59*** *Omega Gradient**$2.63*** *OVHcloud**$2.99*** *AtmosCompute**$3.01*** *Scaleway**$3.21*** *Lambda**$3.29*** *Oblivus**$3.50*** **Marketplace** Hosts set their own price; machines vary. *io.net**$1.49*** *GPU.ai**$1.98*** *Runpod Community Cloud**$1.99*** *Vast.ai**$2.00*** *OpenRelay**$2.20*** **Spot and interruptible** Cheaper hours that can be taken back. *GPUBank**$0.81*** *Hyperstack**$2.00*** *Spheron**$2.20*** **Reserved** Committed for a term, paid either way. *GPUBank**$1.08*** *Hyperstack**$1.75*** *Civo**$1.99*** *Oblivus**$2.56*** **Serverless** Billed by active seconds. *Beam**$3.50*** 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 H100 PCIe or pricing page, so each figure can be read next to the conditions printed beside it. gpubank.io/pricing **GPUBank** All-in $1.35/hr Managed cloud · per-second billing 16 vCPU 188 GB RAM 1 TB NVMe IPv4 and all traffic Frankfurt and Ashburn Spot $0.81 · reserved $1.08 GPUBank rents the H100 PCIe at $1.35 an hour on demand, $0.000375 a second and $985.50 for a 730-hour month, with 16 vCPU, 188 GB of RAM, a 1 TB local NVMe disk, one public IPv4 address and all inbound and outbound traffic included per GPU, in machines of one to eight cards. The only other line on a bill is a persistent disk at $0.03 per GB-month. The meter starts after the machine passes 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. Interruptible capacity costs $0.81 with five minutes of notice, and a week or month reserved costs $1.08. Payment is a prepaid balance topped up in crypto from $25, with bonuses that bring the effective on-demand hour to about $1.13 from $25 and $0.84 from $1,000. Checked 16 Oct 2026 beam.cloud/pricing **Beam** from $1.83/hr Managed cloud · machines and serverless 26 vCPU 200 GB RAM 1 TB NVMe No egress fees Regions and stock not stated Beam's pricing page lists an on-demand H100 PCIe machine from $1.83 an hour with 26 vCPU, 200 GB of RAM and a 1 TB NVMe disk, and states that it charges no egress or bandwidth fees. That makes it the cheapest managed offer after GPUBank that we could read at source, with more CPU per card, though 'from' leaves room for higher prices on some configurations and the page states neither regions nor stock. Beam also sells the card serverless at $0.000972 per second of GPU time, about $3.50 an hour, with CPU at $0.000105 per core-second and memory at $0.0000055 per GiB-second on top, so the serverless rate suits endpoints that scale to zero and costs about twice the machine price for anything that runs all day. A Beam blog post quotes $1.74 for the same card, which disagrees with the pricing page; we use the pricing page. Checked 16 Oct 2026 civo.com/ai/cloud-gpu **Civo** $2.49/hr Managed cloud · Kubernetes and VMs 1 H100 PCIe 80 GB No ingress or egress fees Billed by the hour, rounded up $1.99 on a 36-month commitment Civo lists a 'Small H100 PCIe' with one 80 GB card at $2.49 an hour on demand and $1.99 on a 36-month commitment, and it was marked in stock when we read the page. It is one of the few providers that labels both its H100s clearly, with a Small H100 SXM next to it at $2.99 on demand. The page does not list the vCPU and RAM that come with the card, so compare machine sizes in the console before choosing it. Civo's billing documentation says services are billed hourly and rounded up to the nearest full hour, which makes it a poor fit for many short runs: a 61-minute job is billed as two hours, $4.98. For a long-running workload in London with a three-year horizon, the committed rate is competitive with the managed offers above it, though still above GPUBank's one-month reservation. Checked 16 Oct 2026 hyperstack.cloud/gpu-pricing **Hyperstack** $2.50/hr Managed cloud · per-minute billing Up to 28 pCPU 180 GB RAM Root and ephemeral NVMe IPv4 $0.0067/hr Spot $2.00 · reserved from $1.75 Hyperstack lists the card as 'NVIDIA H100' at $2.50 an hour on demand with up to 28 physical CPU cores and 180 GB of RAM per GPU, and as 'NVIDIA H100 PCIe' at $2.00 in its spot table, with reservations starting from $1.75 on a term the page does not state. A separate 'NVIDIA H100 NVLink' row at $2.60 is an eight-card machine of bridged PCIe cards, not the 94 GB NVL. CPU, RAM and the root and ephemeral disks are included and egress is free, but a public IP costs about $0.0067 an hour and shared volumes about $0.07 per GB-month, and billing is by the minute on a prepaid balance. Its own H100 PCIe product page quotes both $1.75 and $2.50 in different places, so read the pricing table rather than the marketing copy. GetDeploying showed the single-card spot flavour in stock on 15 October. Checked 16 Oct 2026 runpod.io/gpu-models/h100-pcie **Runpod** $2.89/hr Secure Cloud · Community Cloud $1.99 16 vCPU 188 GB RAM Disks billed separately Out of stock on 15 Oct Runpod lists the H100 PCIe at $2.89 an hour on Secure Cloud and $1.99 on Community Cloud, where peer hosts supply the machines, both billed per second, with 16 vCPU and 188 GB of RAM on the card's listing. Its H100 PCIe page, dated 27 August 2026, contradicts itself twice: the FAQ gives $2.39 for Secure Cloud against $2.89 in the table, and the headline calls the memory HBM3 while the spec table says HBM2e. Storage is billed on top, at $0.10 per GB-month for a running volume disk, $0.20 while the Pod is idle and $0.07 for a network volume under 1 TB. ComputePrices marked both tiers out of stock on 15 October, so the Community price is a good number to watch rather than one you can count on, and the Secure Cloud price is more than twice GPUBank's for the same vCPU and RAM with every disk billed on top. Checked 16 Oct 2026 lambda.ai/pricing **Lambda** $3.29/hr Managed cloud · per-minute billing 26 vCPU 225 GiB RAM 1 TiB SSD Plus sales tax or VAT Out of stock on 15 Oct Lambda lists a single NVIDIA H100 PCIe instance at $3.29 an hour with 26 vCPU, 225 GiB of RAM and a 1 TiB SSD, plus applicable sales tax or VAT, billed by the minute with no egress fees. It is often the first price people check, and the most expensive of the managed single-card offers we confirmed, at 2.4 times GPUBank's rate for a machine of similar size. Lambda's own single-GPU H100 SXM costs $4.29 on the same page, so the PCIe card saves 23 percent an hour while Lambda's own benchmark puts the SXM about 50 percent faster, which makes the SXM the better buy there for any saturated job. ComputePrices marked the PCIe instance out of stock on 15 October, while GetDeploying showed it in stock the same day, so check the console before planning around it. Checked 16 Oct 2026 ### Where GPUBank wins, and where it does not GPUBank is cheapest where the comparison is like for like: a managed single card with the machine, the disk, the IP address and the traffic included. At $1.35 on demand it undercuts the next managed offer we could read at source, Beam from $1.83, by 26 percent, and Lambda's $3.29 by more than half; its reservation at $1.08 for a week or a month is the cheapest reserved hour we found and the shortest commitment, where Hyperstack starts at $1.75 on an unstated term and Civo asks 36 months for $1.99. Per-second billing from the health check, a full credit when a host fails mid-job and a persistent disk at $0.03 per GB-month make the monthly bill close to the rate times the hours. It loses in four places. Vast.ai interruptible bids have been well below its $0.81 interruptible rate, so a fine-tune that checkpoints often and tolerates host churn can cost less there. Its own H100 SXM at $1.85 finishes saturated training and serving jobs about 1.5 times faster, which makes it cheaper per job whenever the speed-up exceeds the price ratio of 1.37. It offers this card only in Frankfurt and Ashburn, so teams that need a machine in Canada, London, Paris or Warsaw should look at Hyperstack, Civo or Scaleway. And it takes only crypto top-ups from $25 for now, with no card payments, which some finance departments cannot use. ### The rest of the market Beyond the six providers above, the rest of the market is a long tail of smaller clouds whose prices we could read only through trackers. ComputePrices showed QuantaCloud at $2.59 in the US Midwest, Omega Gradient at $2.63 in Canada, OVHcloud at $2.99 for its h100-380 flavour, which pairs the card with 30 vCPU, 380 GB of RAM and 3.84 TB of passthrough NVMe and reports itself to nvidia-smi as 'NVIDIA H100 PCIe' in Spare Cores' test, and AtmosCompute at $3.01, with Thunder Compute at $3.20 in an older reading dated 5 September, and GetDeploying showed Scaleway's single-card H100-1-80G at $3.21 in stock in Paris and Warsaw, Oblivus at $3.50 out of stock and Enverge at $4.00. Scaleway's own page confirms that both its flavours carry the PCIe card, with 24 vCPU and 240 GB of RAM on the single-GPU one, but did not show a price when we read it. io.net's own price guide, dated April 2026, gives $1.49 for an H100 PCIe on its decentralised network, a snapshot we could not refresh. The marketplace readings move by the hour, so their history says more than any single number. GridCompute, which tracks Vast.ai hourly, counted four on-demand H100 PCIe listings with eight GPUs between them on 8 October, with a lowest ask of $1.25 and a median of $2.78, and the median had ranged from $1.63 to $3.52 since tracking began on 2 October. On the interruptible side GetDeploying showed a two-card Vast.ai listing in South Africa at $0.67 per GPU on 15 October, and Beam's own H100 article puts Vast.ai's interruptible offers 'around $0.35 per GPU-hour'. Lium, whose price includes the node's local disk, was tracked at $1.40 in Almaty on 29 September and at $2.15 in Boardman in the United States, with 26 vCPU and 221 GB of RAM, on 7 October. None of these hours comes with a fixed machine or a credit when the host goes away, and on Vast.ai the disk and bandwidth are billed on top of the ask. Massed Compute sells an 'H100 (80GB)' at $2.73 an hour with 20 vCPU, 128 GB of RAM and 1,250 GB of storage, next to separate rows for the H100 NVL and the H100 SXM5, so by elimination it is the PCIe card even though the page never says so. On the marketplace side, Lium had no H100 PCIe free when we looked, only SXM listings, and OpenRelay was tracked at $2.20. None of these is a reason to skip the providers above, but QuantaCloud and Massed Compute are worth a look if you need a US machine with a large local disk and the cheaper clouds are sold out. ### The large clouds AWS p5.48xlarge $6.88 /GPU-hr 8-GPU node · H100 SXM, not PCIe; $55.04 per node Azure NC40ads H100 v5 $6.98 /GPU-hr 1 GPU · H100 NVL 94 GB, not PCIe; 40 vCPU, 320 GiB Oracle BM.GPU.H100.8 $10.00 /GPU-hr 8-GPU node · H100 SXM bare metal, not PCIe Google a3-highgpu-8g $11.06 /GPU-hr 8-GPU node · H100 SXM, not PCIe; $88.49 per node Azure ND96isr H100 v5 $12.29 /GPU-hr 8-GPU node · H100 SXM, not PCIe; about $98.32 per node None of the four large clouds rents an H100 PCIe 80 GB. Their H100 offers are SXM servers sold by the eight-GPU instance: AWS's p5.48xlarge lists at $55.04 an hour, $6.88 per GPU, Google's a3-highgpu-8g costs $88.49 in us-central1, $11.06 per GPU, Oracle's BM.GPU.H100.8 is $10.00 per GPU on its list price, and Microsoft's ND96isr H100 v5 is about $98.32 a node in East US, $12.29 per GPU. Azure's only single-card H100 VM, the NC40ads H100 v5, costs $6.98 an hour in East US with 40 vCPU and 320 GiB of RAM, and it carries an H100 NVL with 94 GB, not this card. These rows are useful only to show the gap. A large-cloud H100 hour costs five to nine times GPUBank's PCIe hour, for a faster card you must usually rent by eight, with egress billed on top and quotas to request before the first launch. Teams that already hold committed spend at one of these clouds, or that need the compliance paperwork and private networking that come with it, have reasons to stay; anyone paying list price for one GPU of single-card work does not. We read these prices through trackers rather than the clouds' own calculators, which require a region and purchase option to be chosen first. 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 | $1.35 | Managed on demand | 1 GPU | 16 vCPU, 188 GB RAM, 1 TB NVMe, IPv4, traffic | Per second | In stock | Provider page | | Beam | $1.83 | Managed on demand | 1 GPU | 26 vCPU, 200 GB RAM, 1 TB NVMe; price 'from' | Not stated | Not stated | Provider page | | Civo | $2.49 | Managed on demand | 1 GPU | Machine size not listed; no egress fees | Hourly, rounded up | In stock | Provider page | | Hyperstack | $2.50 | Managed on demand | 1 GPU | Up to 28 pCPU, 180 GB RAM; IPv4 extra; labelled 'NVIDIA H100' | Per minute | Not stated | Provider page | | QQuantaCloud | $2.59 | Managed on demand | 1 GPU | 20 vCPU, 128 GB RAM, 1,250 GB disk | Not stated | In stock | Aggregator | | OOmega Gradient | $2.63 | Managed on demand | 1 GPU | Not stated; Canada | Not stated | Not stated | Aggregator | | Massed Compute | $2.73 | Managed on demand | 1 GPU | 20 vCPU, 128 GB RAM, 1,250 GB disk; labelled 'H100 (80GB)' | Not stated | Not stated | Provider page | | Runpod Secure Cloud | $2.89 | Managed on demand | 1 GPU | 16 vCPU, 188 GB RAM; disks extra | Per second | Sold out | Provider page | | OVHcloud | $2.99 | Managed on demand | 1 GPU | h100-380: 30 vCore, 380 GB RAM, NVMe passthrough | Hourly | Not stated | Aggregator | | AAtmosCompute | $3.01 | Managed on demand | 1 GPU | Not stated | Not stated | Not stated | Aggregator | | Scaleway | $3.21 | Managed on demand | 1 GPU | H100-1-80G: 24 vCPU, 240 GB RAM; IPv4 extra | Hourly, 1-hour minimum | In stock | Aggregator | | Lambda | $3.29 | Managed on demand | 1 GPU | 26 vCPU, 225 GiB RAM, 1 TiB SSD; plus tax | Per minute | Sold out | Provider page | | Oblivus | $3.50 | Managed on demand | 1 GPU | Not stated | Not stated | Sold out | Aggregator | | io.net | $1.49 | Marketplace | 1 GPU | Decentralised network; snapshot dated April 2026 | Not stated | Not stated | Provider snapshot | | GGPU.ai | $1.98 | Marketplace | 1 GPU | Partner host's machine at its rate | Not stated | Sold out | Aggregator | | Runpod Community Cloud | $1.99 | Marketplace | 1 GPU | 16 vCPU, 188 GB RAM as listed; disks extra | Per second | Sold out | Provider page | | Vast.ai | $2.00 | Marketplace | 1 GPU | Host's share of CPU and RAM; disk and bandwidth extra | Per second | Sold out | Aggregator | | OOpenRelay | $2.20 | Marketplace | 1 GPU | Not stated | Not stated | Not stated | Aggregator | | GPUBank | $0.81 | Spot and interruptible | 1 GPU | Same machine; 5 min notice, stopped not deleted | Per second | In stock | Provider page | | Hyperstack | $2.00 | Spot and interruptible | 1 GPU | Labelled 'NVIDIA H100 PCIe'; notice not stated | Per minute | In stock | Provider page | | Spheron | $2.20 | Spot and interruptible | 4 GPUs | Hyperstack capacity resold, CANADA-1 | Not stated | In stock | Aggregator | | GPUBank | $1.08 | Reserved | 1 GPU | Week or month, prepaid, same machine | Per term | On request | Provider page | | Hyperstack | $1.75 | Reserved | 1 GPU | 'Starting from'; term not stated | Per term | On request | Provider page | | Civo | $1.99 | Reserved | 1 GPU | 36-month commitment | Per term | In stock | Provider page | | Oblivus | $2.56 | Reserved | 1 GPU | 36-month term | Per term | Not stated | Aggregator | | AWS p5.48xlarge | $6.88 | Large clouds | 8-GPU node | H100 SXM, not PCIe; $55.04 per node | Per second | Not stated | Aggregator | | Azure NC40ads H100 v5 | $6.98 | Large clouds | 1 GPU | H100 NVL 94 GB, not PCIe; 40 vCPU, 320 GiB | Per hour | Not stated | Aggregator | | Oracle BM.GPU.H100.8 | $10.00 | Large clouds | 8-GPU node | H100 SXM bare metal, not PCIe | Per hour | Not stated | Aggregator | | Google a3-highgpu-8g | $11.06 | Large clouds | 8-GPU node | H100 SXM, not PCIe; $88.49 per node | Per second | Not stated | Aggregator | | Azure ND96isr H100 v5 | $12.29 | Large clouds | 8-GPU node | H100 SXM, not PCIe; about $98.32 per node | Per hour | Not stated | Aggregator | | Beam | $3.50 | Serverless | 1 GPU | $0.000972 per GPU-second; CPU and RAM extra | Active time | 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 H100 PCIe really costs beyond the hourly rate Storage is the first line that changes the bill. GPUBank charges $0.03 per GB-month for a persistent disk and nothing else while a machine is stopped. Runpod charges $0.07 per GB-month for a network volume under 1 TB and $0.20 for the volume disk of an idle Pod, and Hyperstack's shared volumes work out to about $0.07. Keeping 500 GB of checkpoints for a month therefore costs $15 at GPUBank, about $35 on a Runpod or Hyperstack volume and $100 on an idle Runpod Pod disk, and on a light month that can rival the compute. Hyperstack also bills a public IP at about $0.0067 an hour, $4.91 for a full month, where GPUBank, Beam and Lambda include traffic and GPUBank includes the address. Billing granularity decides what a short job costs. A run of 61 minutes costs $1.37 on GPUBank's per-second meter, $2.54 on Hyperstack and $3.35 on Lambda, both billed by the minute, and $4.98 on Civo, whose billing rounds each service up to the next full hour, so the 61 minutes, worth $2.53 at its rate, are billed as two. Scaleway also bills by the hour with a one-hour minimum per resource, according to GetDeploying. On a month of evaluations, restarts and quick tests, that rounding can be worth more than the gap between two hourly rates, so a team that launches many short jobs should weigh the meter before the headline price. The last cost is time. A marketplace host that disappears mid-run, a spot instance reclaimed without warning or a listing slower than its card suggests costs the GPU-hours already spent and the hours to redo them, and on a marketplace the renter pays for both. GPUBank credits a session interrupted by a host failure in full, gives five minutes of notice before reclaiming interruptible capacity and credits the last 30 minutes if the notice is missed. Prepaid balances add a smaller risk: GPUBank's minimum top-up is $25 and the credit never expires, and its bonus lowers the effective rate only if the credit is actually spent. Running dry is treated very differently from one cloud to the next: Hyperstack's FAQ warns that resources on a drained balance are scheduled for termination after a short grace period and may be deleted for good after a week. ### Three workloads, priced for a month **GPUBank** 220 h GPU $297.00 Storage $6.00 Other None Month $303.00 On demand; $189.38 in cash from a $1,000 top-up All-in **Beam** 220 h GPU $402.60 Storage $0.00 Other Storage not priced Month $402.60 Machine 'from' $1.83; persistent storage price not on page **Runpod Community** 220 h GPU $437.80 Storage $14.00 Other None Month $451.80 Network volume $0.07/GB-month; card in stock (it was not on 15 Oct) **Vast.ai** 220 h GPU $440.00 Storage $0.00 Other Disk and bandwidth set by host Month $440.00 $2.00 host held all month, extras not counted **Hyperstack** 220 h GPU $550.00 Storage $14.13 Other IPv4 $1.48 Month $565.61 IP billed while running GPUBank wins this one on every assumption we tried. 220 hours at $1.35 is $297.00, and 200 GB of persistent disk at $0.03 adds $6.00, for $303.00; paid from a $1,000 top-up credited as $1,600, the same month costs $189.38 in cash. Beam's machine from $1.83 comes to $402.60 before any persistent storage, a Vast.ai host at $2.00 to $440.00 before the host's disk and bandwidth, Runpod Community to $437.80 plus $14.00 of network volume, $451.80, and Hyperstack to $550.00 plus $14.13 of volume and $1.48 of IP, $565.61. Only a marketplace host below about $1.30 kept for the whole month would undercut GPUBank's list price, and none was listed when we checked. 07 / Per job ## What the H100 PCIe does per job Every measurement we found puts the H100 SXM about half again as fast as the PCIe card. Lambda measured a 49 to 51 percent premium for the SXM on a geomean of PyTorch FP16 and TF32 training throughput across SSD, BERT, GNMT, ResNet-50, Tacotron 2 and WaveGlow, and Runpod's vLLM test on Qwen2.5-7B-Instruct at one request per second measured 165 tokens a second on the SXM against 109 on the PCIe card, a ratio of 1.51, with a median time to first token of 43 ms on the PCIe card. The gap follows the hardware: 3.35 against 2.0 TB/s of memory bandwidth and 700 against 350 W of power to sustain clocks. Against the A100 the picture is closer. In llama.cpp on Llama 3, the H100 PCIe generated 25.01 tokens a second on 70B Q4_K_M against 24.33 for the A100 SXM, almost the same, but processed prompts much faster, 7,760 against 5,864 tokens a second on 8B Q4_K_M, and 984 against 797 on 70B Q4_K_M, where the L40S managed 649. On the small 8B model at Q4_K_M it generated 144.49 tokens a second, against 138.31 on the A100 PCIe, 133.38 on the A100 SXM and 113.60 on the L40S. llama.cpp does not use Hopper's FP8 path, so these numbers understate what the H100 gains in vLLM or TensorRT-LLM with FP8 weights. Runpod's vLLM test, which does run on a modern serving stack, put the PCIe card 14 percent ahead of the A100 SXM at 109 against 96 tokens a second. Turned into cost at GPUBank's on-demand prices, a million single-stream tokens in Runpod's test costs $3.44 on the H100 PCIe, $3.11 on the H100 SXM, $2.75 on the A100 SXM, $3.18 on the L40S and $3.59 on the RTX PRO 6000, and in the llama.cpp 70B test a million generated tokens costs $14.99 on the H100 PCIe, $10.85 on the A100 SXM and $9.98 on the L40S. These are single-stream figures, and a batched server makes every token far cheaper on all of these cards, but the ranking carries over. A training run that takes 10 hours on the PCIe card ($13.50) takes about 6.7 hours on the SXM ($12.33). The PCIe card earns its hour when the GPU is partly idle, as in notebooks, development sessions, low-traffic endpoints and pipelines held back by CPU or storage, where wall-clock time rather than throughput sets the bill, and when a model needs more than the L40S's 48 GB. **vLLM decode speed, one user** Qwen2.5-7B-Instruct, vLLM, fixed load of 1 request per second, measured 15 Sep 2026 H100 SXM 165 tok/s H100 PCIe 109 tok/s A100 SXM 96 tok/s RTX PRO 6000 89 tok/s L40S 48 tok/s Runpod, Best GPU for LLM inference (2026); vLLM version and precision not stated; per-user speed, not saturated throughput. **Llama 3 70B at 4-bit, text generation** llama.cpp (cuBLAS), llama-bench, Q4_K_M, all layers on one GPU H100 PCIe 25 tok/s A100 SXM 24.3 tok/s A100 PCIe 22.1 tok/s L40S 15.3 tok/s XiongjieDai, GPU-Benchmarks-on-LLM-Inference, single rented GPUs; llama.cpp does not use the FP8 path. **Llama 3 8B at 4-bit, prompt processing** llama.cpp (cuBLAS), llama-bench, Q4_K_M, all layers on one GPU H100 PCIe 7,760 tok/s L40S 5,909 tok/s A100 SXM 5,864 tok/s A100 PCIe 5,800 tok/s XiongjieDai, GPU-Benchmarks-on-LLM-Inference, single rented GPUs. **Cost per million tokens on GPUBank** Runpod's single-user vLLM speeds against each card's on-demand hour; lower is cheaper A100 SXM 2.75 USD/M tok H100 SXM 3.11 USD/M tok L40S 3.18 USD/M tok H100 PCIe 3.44 USD/M tok RTX PRO 6000 3.59 USD/M tok Our arithmetic: tokens a second times 3,600, priced at $0.95, $1.85, $0.55, $1.35 and $1.15 an hour; single stream, so a batched server costs far less per token on every card. WHOLE MACHINE $1.35 GPUBank on demand, CPU, RAM, NVMe, IP and traffic included SXM SPEED-UP 1.5x Lambda training and Runpod vLLM, same verdict BUY BREAK-EVEN 904 h a month to beat $1.35; a month has 730 08 / Against other cards ## H100 PCIe or another card, job by job For single-stream decoding of a model that fits in 48 GB, the L40S at $0.55 is cheaper per token than the H100 PCIe even though it is less than half as fast: 48 against 109 tokens a second in Runpod's test, and 15.31 against 25.01 on Llama 3 70B at 4-bit in llama.cpp. The A100 SXM at $0.95 decodes at nearly the PCIe card's speed for 30 percent less, which makes it the cheaper choice for plain inference and fine-tunes that do not use FP8. The RTX PRO 6000 at $1.15 brings 96 GB, enough for a 70B model at 8-bit on one card, and CloudRift measured it slightly ahead of an H100 SXM under heavy concurrent load, 3,140 against 2,987 tokens a second serving GLM-4.5-Air at 4-bit with 256 to 512 requests in flight, though it trails the PCIe card at one request per second, 89 against 109. That leaves the H100 PCIe a specific place: 80 GB of HBM, FP8 and fast prefill on one card, at a lower hour than the SXM, for work that does not keep the GPU saturated. If the job runs flat out for hours, rent the H100 SXM; if it fits in 48 GB, rent the L40S; if it needs more than 80 GB on one card, rent the RTX PRO 6000. On GPUBank all four sit on the same terms, per-second billing with the machine included, so switching cards between jobs costs nothing but a new launch. 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 H100 PCIe The realistic purchase is one card in a standard 2U server. GPUDojo's tracker recorded a Newegg listing at $29,500 on 9 October 2026, a 90-day low after $34,989 a week earlier, and a used eBay offer at $27,999.99 in September, and Caladan Semiconductor recorded three used US asks between $28,000 and $32,399, with a median of $32,000, on 9 September. We take $29,500 for the card and assume $10,000 for the server, a resale value after two years of 35 percent of the card price and $2,000 for the server, 0.65 kW at the wall under load and $0.15 per kWh. The fixed cost is then $39,500 less $12,325 of resale over 24 months, $1,132.29 a month, plus $0.0975 of power for every hour of load. Against GPUBank's $1.35 on demand, ownership breaks even only at 1,132.29 divided by 1.2525, about 904 hours a month, and a month has 730. At full use owning costs $1,203.47 a month against $985.50 on demand, $788.40 reserved and $615.94 at the effective rate from a $1,000 top-up, before any colocation fee; a rack slot at $150 a month would lift the fixed cost to $1,282.29 and the break-even against on demand to about 1,024 hours. With a used card at $25,000, an $8,000 server and 50 percent card resale, ownership falls to about $842 a month at full use, below the on-demand month but still above the reserved and top-up prices. Buy only if your data cannot leave your premises, the rack and power are already paid for and the card will be busy every hour for two years. Monthly cost of one H100 PCIe in a 2U server (1 GPU) against hours of use a month, 0 to 730. Owning: price + $10,000 host − $12,325 resale over 24 months, plus 0.65 kW at the wall under load. Rent: per GPU-hour × 1. | Hours a month | Rent at $1.35 | Rent at $0.84 effective | Own | Cheaper | | --- | --- | --- | --- | --- | | 100 | $135.00 | $84.38 | $1,142.04 | Rent | | 200 | $270.00 | $168.75 | $1,151.79 | Rent | | 400 | $540.00 | $337.50 | $1,171.29 | Rent | | 730 | $985.50 | $615.94 | $1,203.47 | Rent | Newegg listing recorded by GPUDojo on 9 Oct 2026; used asks $28,000 to $32,399 (Caladan, Sep 2026) 10 / What to choose ## The cheapest H100 PCIe depends on the job. A team serving a 30B to 70B model to a few users H100 PCIe on demand or reserved 80 GB holds a 70B model at 4-bit with room for context, and the endpoint rarely saturates the card, so the cheaper hour wins over the SXM. Fine-tuning runs that keep the GPU busy for hours H100 SXM, interruptible if you checkpoint About 1.5 times faster for $1.85 against $1.35, so each finished run costs less on the SXM. Notebooks, development and evaluation sessions H100 PCIe billed per second The card sits idle much of the time, so wall-clock price and per-second billing matter more than peak throughput. Models that fit in 48 GB, served at the lowest cost per token L40S or A100 SXM At $0.55 and $0.95 an hour they cost less per generated token than the H100 PCIe in every single-stream test we found. 11 / Questions ## What people ask before renting one. The cheapest managed H100 PCIe we could verify on a provider's own page was GPUBank at $1.35 an hour on demand, $1.08 reserved and $0.81 interruptible, with 16 vCPU, 188 GB of RAM, a 1 TB NVMe disk, a public IP and all traffic included. Beam's machine from $1.83, Civo at $2.49 and Hyperstack at $2.50 came next. Marketplaces can list lower hours, and Vast.ai's interruptible bids have been well under a dollar, but on 16 October its H100 PCIe page showed no offers, and Runpod Community Cloud at $1.99 and GPU.ai at $1.98 were marked out of stock. Which is cheapest for you depends on whether your job survives an interruption and whether you want to manage hosts and storage yourself. Sources ## Sources 1. 01 [GPUBank pricing](https://gpubank.io/pricing) H100 PCIe on demand, interruptible and reserved rates, inclusions, disk price, top-up bonuses, other GPUBank cards · read 16 Oct 2026 2. 02 [Beam pricing](https://www.beam.cloud/pricing) H100 PCIe machine from $1.83, specs, serverless per-second rate, no egress fees · read 16 Oct 2026 3. 03 [Civo cloud GPU](https://www.civo.com/ai/cloud-gpu) Small H100 PCIe at $2.49, $1.99 on 36 months, in stock · read 16 Oct 2026 4. 04 [Hyperstack GPU pricing](https://www.hyperstack.cloud/gpu-pricing) H100 on demand, spot and reserved rates, labels, public IP, volumes · read 16 Oct 2026 5. 05 [GetDeploying, Vast.ai pricing](https://getdeploying.com/vast-ai) Two-card interruptible H100 PCIe listing at $0.67 per GPU on 15 Oct 2026 · read 15 Oct 2026 6. 06 [Vast.ai, Rent H100 PCIE GPUs](https://vast.ai/pricing/gpu/H100-PCIE) No current offers on the check date · read 16 Oct 2026 7. 07 [GetDeploying, NVIDIA H100 rental prices](https://getdeploying.com/gpus/nvidia-h100) Aggregator rows and stock for Scaleway, Oblivus, GPU.ai, Spheron, Hyperstack NVLink machine · read 16 Oct 2026 8. 08 [ComputePrices, H100 PCIe](https://computeprices.com/gpus/h100pcie) Aggregator rows dated 15 Oct 2026, stock flags for Runpod and Lambda · read 16 Oct 2026 9. 09 [NVIDIA H100 PCIe GPU product brief, PB-11133-001_v02](https://www.nvidia.com/content/dam/en-zz/Solutions/gtcs22/data-center/h100/PB-11133-001_v01.pdf) HBM2e, 2,000 GB/s, 350 W, NVLink bridges at 600 GB/s, form factor · read 16 Oct 2026 10. 10 [NVIDIA H100 PCIe datasheet (October 2022, partner copy)](https://e4company.com/wp-content/uploads/nvidia-h100-datasheet-nvidia-us-2287922-web-sep22-partner-r6.pdf) Sparse tensor TFLOPS and the note that dense figures are half · read 16 Oct 2026 11. 11 [Vast.ai documentation, Pricing](https://docs.vast.ai/documentation/instances/pricing) Host-set prices, storage and bandwidth billed on top, per-second billing · read 15 Oct 2026 12. 12 [Spare Cores, OVHcloud h100-380](https://sparecores.com/server/ovh/h100-380) h100-380 specs and nvidia-smi reporting 'NVIDIA H100 PCIe' · read 15 Oct 2026 13. 13 [GPUPerHour, rent H100 PCIe](https://gpuperhour.com/rent/h100-pcie) QuantaCloud and Massed Compute offers in stock on 15 Oct 2026 · read 16 Oct 2026 14. 14 [Scaleway H100 PCIe GPU instance](https://www.scaleway.com/en/h100/) H100-1-80G and H100-2-80G carry H100 PCIe cards, vCPU and RAM · read 16 Oct 2026 15. 15 [io.net, How much does an H100 cost per hour](https://io.net/p/fhow-much-does-an-h100-cost-per-hour) H100 PCIe at $1.49, page dated April 2026 · read 16 Oct 2026 16. 16 [GridCompute, H100 PCIe rental price per hour (Vast.ai tracker)](https://gridcompute.ai/gpu/h100-pcie) Vast.ai on-demand listings on 8 Oct 2026: lowest $1.25, median $2.78, median range since 2 Oct · read 15 Oct 2026 17. 17 [Beam, NVIDIA H100 pricing](https://www.beam.cloud/blog/nvidia-h100-pcie-pricing) Vast.ai interruptible H100 PCIe 'around $0.35 per GPU-hour' · read 15 Oct 2026 18. 18 [ComputePrices, Lium GPU pricing](https://computeprices.com/providers/lium) Lium H100 PCIe at $1.40 (29 Sep) and $2.15 (7 Oct) · read 15 Oct 2026 19. 19 [Massed Compute pricing](https://vm.massedcompute.com/pricing) H100 (80GB) at $2.73, H100 NVL and SXM5 rows · read 16 Oct 2026 20. 20 [Lium, H100 listings](https://lium.io/gpu/h100) SXM listings from $1.39, no H100 PCIe free · read 16 Oct 2026 21. 21 [Vantage, p5.48xlarge](https://instances.vantage.sh/aws/ec2/p5.48xlarge) AWS p5.48xlarge on-demand price, 8 H100 SXM · read 16 Oct 2026 22. 22 [gcloud-compute.com, a3-highgpu-8g](https://gcloud-compute.com/a3-highgpu-8g.html) Google a3-highgpu-8g on-demand price in us-central1 · read 16 Oct 2026 23. 23 [Spheron, Oracle Cloud GPU pricing 2026](https://www.spheron.network/blog/oracle-cloud-oci-gpu-pricing-2026/) Oracle BM.GPU.H100.8 at $10.00 per GPU (third-party) · read 15 Oct 2026 24. 24 [Spheron, Azure H100 pricing](https://www.spheron.network/blog/azure-h100-pricing/) Azure ND96isr H100 v5 node price (third-party) · read 15 Oct 2026 25. 25 [cloudprice.net, Standard_NC40ads_H100_v5](https://cloudprice.net/vm/Standard_NC40ads_H100_v5) Azure single H100 NVL VM at $6.98 in East US · read 16 Oct 2026 26. 26 [Lambda pricing](https://lambda.ai/pricing) H100 PCIe at $3.29 and H100 SXM single-GPU at $4.29, machine specs · read 16 Oct 2026 27. 27 [Lambda, new 1x, 2x and 4x H100 SXM instances](https://lambda.ai/blog/creating-more-options-for-ai-developers-new-1x-2x-and-4x-sxm) 49 to 51 percent SXM premium over PCIe on PyTorch training throughput · read 16 Oct 2026 28. 28 [Runpod, Best GPU for LLM inference: what the benchmarks show in 2026](https://www.runpod.io/articles/guides/best-gpu-for-llm-inference) vLLM single-user decode speeds, measured 15 Sep 2026 · read 16 Oct 2026 29. 29 [Runpod pricing](https://www.runpod.io/pricing) H100 PCIe, SXM and NVL rates on Secure and Community Cloud, storage prices · read 16 Oct 2026 30. 30 [Civo documentation, Billing](https://civo.com/docs/account/billing) Hourly billing rounded up to the full hour · read 16 Oct 2026 31. 31 [NVIDIA H100 Tensor Core GPU product page](https://www.nvidia.com/en-us/data-center/h100/) H100 SXM and H100 NVL specifications, sparsity footnote · read 16 Oct 2026 32. 32 [XiongjieDai, GPU-Benchmarks-on-LLM-Inference](https://github.com/XiongjieDai/GPU-Benchmarks-on-LLM-Inference) llama.cpp generation and prompt processing on Llama 3 · read 16 Oct 2026 33. 33 [CloudRift, RTX PRO 6000 vs H100, H200 and L40S](https://www.cloudrift.ai/blog/benchmarking-rtx6000-vs-datacenter-gpus) Saturated vLLM throughput, RTX PRO 6000 against H100 SXM · read 15 Oct 2026 34. 34 [GPUDojo, H100 PCIe price history](https://gpudojo.com/h100-pcie) Newegg $29,500 on 9 Oct 2026, eBay $27,999.99 in Sep · read 16 Oct 2026 35. 35 [Caladan Semiconductor, used H100 and A100 buying guide 2026](https://caladansemi.com/blog/used-nvidia-h100-a100-gpu-buying-guide-2026) Used H100 PCIe asks of $28,000 to $32,399 · read 15 Oct 2026 36. 36 [Koyeb pricing](https://www.koyeb.com/pricing) Unlabelled H100 at $2.50 · read 16 Oct 2026 37. 37 [HyperAI pricing](https://hyper.ai/en/pricing) Unlabelled H100 at $1.80 · read 16 Oct 2026 38. 38 [fal.ai pricing](https://fal.ai/pricing) Unlabelled H100 at $4.50, 'as low as $2.49' · read 16 Oct 2026 39. 39 [Hyperstack, NVIDIA H100 PCIe](https://www.hyperstack.cloud/h100-pcie) Conflicting $1.75 and $2.50 figures on the product page · read 16 Oct 2026 40. 40 [Runpod, H100 PCIe GPU rental](https://www.runpod.io/gpu-models/h100-pcie) Page dated 27 Aug 2026, conflicting FAQ price and memory type · read 16 Oct 2026 41. 41 [GPUBank home page](https://gpubank.io/) Launch time and the $2.71 market average from GPUBank's 29 Sep survey · read 16 Oct 2026 What we could not confirm Vast.ai's interruptible H100 PCIe price, recorded at $0.364 on 29 September, could not be checked because its page listed no offers on 16 October; the $0.67 two-card listing and the 'around $0.35' figure come from GetDeploying and a Beam article, the $1.25 lowest ask from GridCompute on 8 October, and Lium's rows from ComputePrices, not from the marketplaces themselves; the $2.00 Vast.ai row comes from ComputePrices on 15 October. Scaleway's euro price did not appear on its own pages when we read them, so its $3.21 is GetDeploying's conversion, and OVHcloud's $2.99, QuantaCloud, Omega Gradient, AtmosCompute, OpenRelay, GPU.ai, Oblivus and Spheron rows are aggregator readings. io.net's $1.49 is its own snapshot dated April 2026. Beam states no regions or stock, and Hyperstack's reservation term, Civo's machine size and Massed Compute's variant are not shown on their pages. The four large-cloud prices were read through trackers, Azure's ND96isr and Oracle's through third-party articles, and none is for an H100 PCIe 80 GB. NVIDIA's PCIe datasheet calls the memory HBM3 where its product brief says HBM2e; we follow the brief. Whether GPUBank's top-up bonus applies to reservations is not stated. The server price, power, resale and electricity figures in the rent-or-buy section are our assumptions, not quotes. H100 PCIe on GPUBank ## Eighty gigabytes of Hopper, by the second. - 16 vCPU, 188 GB RAM, 1 TB NVMe, IPv4 and traffic in $1.35/hr - Billed by the second after the health check - Interruptible $0.81, reserved $1.08 [Launch an H100 PCIe · $1.35/hr](https://console.gpubank.io/launch?gpu=h100-pcie) [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) [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) [A100 SXM 80 GB HBM2e From $0.57 spot · $0.95 all-in](https://gpubank.io/rent-a100) [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)