Add our Chrome extension
GPULEADERBOARDby Marktechpost●
⌘K
A10 $1.29/GPU/hr ●A100 40GB $0.89/GPU/hr ●A100 80GB $1.35/GPU/hr ●A30 $0.35/GPU/hr ●A40 $0.59/GPU/hr ●B200 $5.43/GPU/hr ●B300 $6.25/GPU/hr ●GB200 $10.50/GPU/hr ●GB300 $13.72/GPU/hr ●GH200 $6.50/GPU/hr ●H100 $1.68/GPU/hr ●H200 $3.00/GPU/hr ●L4 $0.44/GPU/hr ●L40 $0.79/GPU/hr ●L40S $0.69/GPU/hr ●MI300X $2.59/GPU/hr ●MI325X $3.80/GPU/hr ●MI350X $4.00/GPU/hr ●MI355X $5.99/GPU/hr ●RTX 3080 Ti $0.20/GPU/hr ●RTX 3090 $0.20/GPU/hr ●RTX 3090 Ti $0.22/GPU/hr ●RTX 4000 Ada $0.76/GPU/hr ●RTX 4080 $0.27/GPU/hr ●RTX 4090 $0.44/GPU/hr ●RTX 5080 $0.32/GPU/hr ●RTX 5090 $0.60/GPU/hr ●RTX 6000 Ada $0.79/GPU/hr ●RTX A4000 $0.10/GPU/hr ●RTX A5000 $0.27/GPU/hr ●RTX A6000 $0.35/GPU/hr ●RTX PRO 4500 $0.92/GPU/hr ●RTX PRO 6000 $1.85/GPU/hr ●T4 $0.20/GPU/hr ●V100 16GB $0.99/GPU/hr ●V100 32GB $0.42/GPU/hr ●

Cheapest Cloud GPUs in November 2026: Verified Prices by VRAM Tier

The cheapest cloud GPU at every VRAM tier on Nov 13, 2026: RTX A6000 at $0.35/hr, A100 80GB at $1.09/hr and H100 at $1.68/hr. Every price is checked on the provider's own pricing page, with interruptible and dedicated options labeled.

GPU Leaderboard Team5 min read
Cheapest cloud GPUs by VRAM tier, Nov 13 2026: RTX A6000 $0.35/hr, A100 80GB $1.09/hr, H100 $1.68/hr, H200 $3.99/hr

The cheapest cloud GPU you can rent on demand depends on how much VRAM you need. On November 13, 2026, the cheapest dedicated options were an RTX A6000 (48 GB) at $0.35/hr, an A100 80GB at $1.09/hr and an H100 80GB at $1.68/hr. Interruptible consumer cards go lower: an RTX 3090 (24 GB) is $0.20/hr at High priority on SaladCloud, or $0.09/hr at its Lowest tier. We checked every price here on the provider's own pricing page.

Last verified: November 13, 2026. Prices are public on-demand list prices in USD per GPU-hour.

TL;DR

  • Cheapest 48 GB GPU: RTX A6000 at $0.35/hr on Thunder Compute.
  • Cheapest 80 GB GPU: A100 80GB at $1.09/hr on Thunder Compute.
  • Cheapest H100: $1.68/hr on Latitude, for 1 H100 80GB.
  • Cheapest H200 (141 GB): $3.99/hr on Hyperstack (SXM).
  • Cheapest 24 GB GPU: RTX 3090 on SaladCloud, $0.20/hr at High priority or $0.09/hr at Lowest. Its nodes are interruptible.

What is the cheapest cloud GPU by VRAM tier?

VRAM decides which models fit, so we rank by memory tier first. "Dedicated" means a reserved cloud instance. "Interruptible" means the node can go offline mid-job.

Memory tierGPUVRAMCheapest price/hrProviderTypeNext cheapest we verified
24 GBRTX 309024 GB$0.20SaladCloud (High priority)Interruptible—
32 GBRTX 509032 GB$0.60SaladCloud (High priority)Interruptible$0.65 CloudRift
48 GBRTX A600048 GB$0.35Thunder ComputeDedicated$0.50 Hyperstack
48 GBL40S48 GB$0.69CloudRiftDedicated$1.57 DigitalOcean
80 GBA100 80GB80 GB$1.09Thunder ComputeDedicated$1.35 Hyperstack
80 GBH10080 GB$1.68LatitudeDedicated (bare metal)$2.50 Hyperstack
141 GBH200 (SXM)141 GB$3.99HyperstackDedicated$4.47 DigitalOcean
192 GBMI300X192 GB$2.59DigitalOceanDedicated$3.45 Crusoe
BlackwellB200 (SXM6)180 GB~$5.43Massed Compute (8-GPU node)Dedicated$6.25 Daytona
Blackwell UltraB3002.1 TB per 8-GPU DGX$6.25DaytonaDedicated~$8.00 Latitude (8-GPU node)

Our cheapest GPU page tracks these floors automatically across 24 providers. The full list of offers is on the leaderboard.

Notes on each pick

RTX 3090 and RTX 5090 on SaladCloud

SaladCloud runs containers on a distributed network of mostly desktop computers. Its High tier is $0.20/hr for an RTX 3090 and $0.60/hr for an RTX 5090. High priority is never preempted by other workloads. But SaladCloud says its nodes are interruptible like spot, and owners can take them back. The Lowest tier is $0.09/hr for a 3090 and can be reallocated. Use it for retryable batch jobs and stateless workers that run on several replicas.

RTX A6000 on Thunder Compute

At $0.35/hr, this is the cheapest 48 GB GPU we verified. Thunder bills per minute while the instance runs. Each GPU includes 100 GB of disk. Extra disk is $0.219/GB/month, and extra vCPUs are $0.04/hr.

L40S on CloudRift

CloudRift lists the L40S at $0.69/hr. DigitalOcean charges $1.57/hr and Daytona $1.99/hr for the same card. That makes CloudRift less than half the price of the next 2 offers.

A100 80GB on Thunder Compute

Thunder's A100 80GB is $1.09/hr, the cheapest 80 GB GPU we found. Hyperstack follows at $1.35/hr for its A100 80GB. Hyperstack's A100 SXM is $1.60/hr.

H100 on Latitude

Latitude's g3.h100.small is 1 H100 80GB starting at $1.68/hr, or $1,230/month. It is a bare-metal server with 2 × 3.8 TB NVMe and 20 TB of free outbound traffic. Latitude does not state whether the GPU is PCIe or SXM. Other single-GPU H100 offers we checked ran $2.50 to $4.41/hr. See Latitude's provider page.

H200 on Hyperstack

The H200 is mainly a memory upgrade. NVIDIA lists 141 GB of HBM3e at 4.8 TB/s, versus 80 GB on the H100 SXM. Hyperstack's H200 SXM is $3.99/hr. DigitalOcean is $4.47/hr and Daytona is $4.54/hr. Hyperstack bills per minute and charges no egress fees.

MI300X on DigitalOcean

DigitalOcean's MI300X is $2.59/hr for 192 GB. That is more memory than an H200 for less money. It runs AMD's ROCm stack, not CUDA, so confirm your serving engine supports it.

B200 and B300

Massed Compute lists its B200 as an 8-GPU node at $43.46/hr, or about $5.43 per GPU. Daytona lists the B200 and B300 at $6.25/hr. Its page does not say whether that is per GPU. Latitude's B300 is an 8-GPU node at $64/hr, or $8.00 per GPU.

How much cheaper are spot and preemptible GPUs?

Preemptible capacity costs less if your job can restart from a checkpoint:

  • Daytona H100: $2.27/hr preemptible vs $3.95/hr standard (43% off).
  • Daytona B300: $4.08/hr vs $6.25/hr (35% off).
  • SaladCloud RTX 3090: $0.09/hr at Lowest priority vs $0.20/hr at High (55% off).

Discount = 1 − (preemptible price ÷ standard price).

What does a real job cost?

Say you serve a model 8 hours a day, 22 working days a month. That is 176 GPU-hours. Monthly cost = price per GPU-hour × 176. These are estimates at list price. They exclude storage, extra vCPUs and taxes.

SetupPrice/hrMonthly estimate
1× H100, Latitude$1.68$295.68
1× H100, Hyperstack$2.50$440.00
1× H100, DigitalOcean$4.41$776.16
1× H200, Hyperstack$3.99$702.24
2× H100, Latitude$3.36$591.36

The same H100 hours cost 2.6x more on DigitalOcean than on Latitude. If your model needs about 140 GB, 2 Latitude H100s cost less than 1 Hyperstack H200. That only holds if your serving engine splits the model well across 2 GPUs. Note too that Latitude's single-H100 servers are separate machines, not 1 linked node.

Also account for idle time. Effective cost per used hour = list price ÷ utilization. An H100 at $1.68/hr that sits idle half the time costs $3.36 per used hour. Run your own numbers with the cost calculator.

What the hourly price leaves out

  • Storage and CPU: some providers charge extra for disk, snapshots or added vCPUs.
  • Networking: egress fees and public IPs vary by provider.
  • Form factor: PCIe, SXM and NVL versions of the same GPU differ in interconnect and memory.
  • Availability: we did not test whether each GPU was in stock in every region.
  • Taxes and minimums: list prices exclude tax. Some offers are only sold as 8-GPU nodes.

Methodology and disclaimers

  • Sources: We took the floor price for each GPU from gpuleaderboard.com. Then we checked each one on the provider's pricing page: Latitude, Thunder Compute, Hyperstack, CloudRift, SaladCloud, DigitalOcean, Massed Compute, Daytona and Crusoe.
  • Conflicts: where our tracker and a provider page differed, we used the provider's number.
  • Scope: public, self-serve list prices only. Reserved contracts, negotiated discounts and marketplace bids are excluded.
  • Per-GPU math: multi-GPU node prices are divided by GPU count. Where a page does not state the billing unit, we read the hourly rate as per GPU and say so above.
  • Specs: VRAM comes from NVIDIA datasheets or the provider's own listing.
  • Ranking: order is by verified list price only.
  • Estimates: monthly costs are calculations, not quotes.
  • Accuracy: prices change often, and automated price tracking can be outdated or wrong. Confirm the price, region and stock with the provider before you commit. This is not purchasing advice.
  • Accessed: November 13, 2026.

Compare all 24 tracked clouds on our providers page.

Frequently asked questions

What is the cheapest cloud GPU right now?+

On Nov 13, 2026, the cheapest dedicated cloud GPU we verified was Thunder Compute's RTX A6000 (48 GB) at $0.35/hr. SaladCloud's RTX 3090 (24 GB) was lower at $0.20/hr, but its nodes are interruptible.

What is the cheapest H100 per hour?+

Latitude lists 1 H100 80GB starting at $1.68/hr on its pricing page (checked Nov 13, 2026). Hyperstack was next at $2.50/hr.

What is the cheapest 80 GB GPU to rent?+

Thunder Compute's A100 80GB at $1.09/hr was the cheapest 80 GB GPU we verified on Nov 13, 2026. The cheapest H100 80GB was $1.68/hr on Latitude.

Are spot or preemptible GPUs worth it?+

They cost 35–55% less in the examples we checked, such as a Daytona H100 at $2.27/hr preemptible vs $3.95/hr. Use them only for jobs that can restart from a checkpoint.

Does the hourly GPU price include everything?+

No. Storage, extra vCPUs, egress, public IPs and taxes can add to the bill, and some GPUs are only sold as 8-GPU nodes. Check the provider's full pricing page before you commit.

H100H200A100RTX A6000L40SMI300XB200B300LatitudeThunder ComputeHyperstackSaladCloudCloudRiftDigitalOceanMassed ComputeDaytona
Compare live GPU prices
On-demand and spot rates for H100, H200, B200 and more, verified against each provider's own pricing page.
See GPU prices

More from the blog

H100 floor · on-demand
$1.68
no change vs 7 days ago
Live instances
205
154 on-demand · 51 spot
Providers
24
with live prices
Next price refresh
23:23:16
daily at 06:00 UTC
GPULEADERBOARD

On-demand and spot GPU prices for running inference, read from each provider's own pricing page.

Prices updated 14 Nov 2026 · sources: provider pricing pages

Prices are collected automatically from providers' public pages and can be outdated or wrong. Confirm with the provider before you buy.

© 2026 Marktechpost AI Media Inc. All rights reserved.