Top 70+ Cloud GPU Providers
We track 71 cloud GPU providers and 14 curated GPU model families spanning more than 2,500 distinct instance configurations.
Cloud GPU providers fall into three groups. The chart groups broad cloud platforms under Hyperscaler, GPU and AI specialists under Neocloud, and distributed marketplaces under Community. The first group also includes smaller general-purpose clouds; these labels describe service models, not company size.
Provider comparison table
Cloud | Models | Combinations | Billing tiers | Leading-edge |
|---|---|---|---|---|
4 | 4 | 1 (On-demand) | No | |
Sesterce | 23 | 83 | 1 (On-demand) | Yes |
RunPod | 35 | 78 | 2 (On-demand/Reserved) | Yes |
EmpirioLabs AI | 22 | 52 | 2 (On-demand/Spot) | Yes |
Massed Compute | 13 | 45 | 2 (On-demand/Spot) | Yes |
Seeweb | 11 | 44 | 2 (On-demand/Reserved) | Yes |
Verda | 11 | 43 | 3 (On-demand/Spot/Reserved) | Yes |
GPU.ai | 18 | 42 | 1 (On-demand) | Yes |
Lium | 21 | 42 | 2 (On-demand/Spot) | Yes |
AWS | 18 | 42 | 2 (On-demand/Reserved) | Yes |
Column definitions:
- Models: distinct GPU model families offered across all vendors (NVIDIA + AMD + Intel). H100 and H100 NVL count as one family.
- Combinations: distinct (GPU model, GPU count) instance SKUs across the public catalog.
- Billing tiers: how many of On-demand, Spot, and Reserved appear in the captured catalog (max 3). Reserved records can have different commitment lengths; Custom indicates quote-based listings.
- Leading-edge: Yes if the provider lists any of B200, B300, MI300X, or RTX 5090.
Ranking: Subscribers are linked and highlighted at the top of the table. The remaining providers are ranked by catalog depth (the Combinations column) in descending order. The table covers 92 providers: 91 in the November 5, 2026 catalog capture plus the separately maintained IONOS catalog. Sesterce leads the captured non-sponsor listings with 83 combinations, followed by RunPod with 78.
IONOS
IONOS is a European cloud platform headquartered in Germany. The public GPU catalog covers four single-GPU SKUs:
The IONOS reference price for RTX PRO 6000 Blackwell is $1.79 per GPU-hour.
IONOS also runs a dedicated GPU server tier with NVIDIA H100/H200 at $3,990/month and NVIDIA L40s or Intel Gaudi 2 & 3 on request. The four hourly SKUs above are on-demand only, with a posted monthly maximum cap on each. Hosting and data centers are in the EU/EEA, which matters for buyers with GDPR data-residency requirements.
Hyperscaler providers
Hyperscalers offer GPU rental alongside compute, storage, networking, identity and managed services. Price differences vary by model, provider and month.
Amazon Web Services
The captured AWS catalog spans 18 GPU model families, including H100 in the P5 family. It contains on-demand and reserved records. The absence of spot records in this capture does not establish that AWS lacks spot GPU services.123
AWS also offers its own AI accelerators (Trainium for training, Inferentia for inference), which sit outside the GPU rental scope of this comparison. SageMaker, Redshift, and the broader managed-services catalog are common reasons enterprises pick AWS despite the GPU rate premium.
Quota approval is required for most GPU instance types. We received a quota for all H100 and A100 types within a day of applying in our trial.
Microsoft Azure
Azure posts H100 through the ND H100 v5 series (H100 SXM); smaller H100 PCIe configurations are available through the NC-series. The catalog spans 10 GPU model families and includes B200 (ND B200) and AMD MI300X (ND MI300X v5).45
All three billing tiers are publicly listed. Azure has also been building its own AI accelerator program (Maia) for in-house training workloads; those chips are not rentable through the standard GPU instance API.6
Google Cloud Platform
GCP spans 10 GPU model families in the captured catalog. H100 on-demand listings range from $4.75 to $11.68 per GPU-hour, with a $6.64 listing median across variants. B200 is offered in the A4 family; A3 Ultra uses H200.78
All three billing tiers are publicly listed. GCP also offers TPU accelerators (v5p, v6e, Trillium) as a separate product line outside the GPU rental scope of this comparison.
Oracle Cloud Infrastructure
OCI lists 15 GPU model families, including AMD MI300X and MI355X. Its GPU services include bare-metal servers and cluster networking for multi-node workloads.9
Other hyperscaler-tier general clouds
OVHcloud, Scaleway, DigitalOcean, Vultr and Linode/Akamai are additional general-purpose clouds with GPU services. Region, instance availability and the bundled services differ across providers.101112131415
Alibaba Cloud lists four GPU model families in the captured catalog.16
Neocloud providers
Neoclouds focus on GPU and AI infrastructure. Some sell virtual machines or dedicated clusters, while others provide managed training and inference. Compare the hardware bundle and support terms alongside the hourly rate.
Lambda Labs
Lambda lists eight GPU model families and 18 model/count combinations in the capture. The records use on-demand billing. Commitment offers requiring a quote are outside those listed rates.17
CoreWeave
CoreWeave lists 11 GPU model families in the captured catalog, including B200 and B300. On-demand and spot records are present.18
CoreWeave’s ARENA program lets customers test workloads on its infrastructure before committing to capacity.
RunPod
RunPod offers Secure Cloud and Community Cloud capacity. The captured catalog spans 35 GPU model families and 78 model/count combinations, including AMD MI300X. On-demand and reserved records are present; no spot record appears in this capture.19
RunPod also offers Serverless endpoints and deployment templates.
Crusoe
Crusoe provides GPU cloud infrastructure. Its captured catalog includes five GPU model families, with custom pricing records that require a quote.20
FluidStack, Hyperstack, Nebius
FluidStack aggregates GPU capacity from multiple data center operators. Hyperstack has nine GPU model families and all three billing tiers in the capture. Nebius lists eight GPU model families, including B200 and B300.212223
Paperspace by DigitalOcean
Paperspace is part of DigitalOcean. The captured catalog spans 12 GPU model families. Its Gradient notebooks and GPU machines provide different interfaces for running code.24
Other specialist neoclouds
TensorDock, CUDO Compute, Hot Aisle, Sesterce, Lyceum, Cirrascale, Together and Replicate provide additional GPU infrastructure or hosted compute services. Billing and deployment models differ, so the comparison table reports the tiers present in the capture.2526272829303132
Verda and Seeweb (European specialists)
Verda and Seeweb each list 11 GPU model families in the capture. Verda has 43 model/count combinations across on-demand, spot and reserved tiers. Seeweb has 44 combinations across on-demand and reserved tiers.3334
Community marketplaces
Community marketplaces aggregate GPU capacity from participating hosts. Hardware, service terms and availability can vary by host and listing.
Vast.ai
Vast.ai is a GPU marketplace, but it has no rows in the November 5 current-offerings capture. Its historical observations remain in the price charts. Current catalog depth, lowest prices and billing coverage therefore cannot be ranked from this snapshot.35
Salad
Salad supplies distributed compute from participating hosts. Its captured catalog contains 35 GPU model families, with on-demand and spot listings. Hardware, available capacity and host requirements vary.36
Theta EdgeCloud
Theta EdgeCloud spans 20 GPU model families and 25 model/count combinations in the capture. Its records use on-demand billing.37
Provider demand: adoption, spend, and search
Adoption, share of spend, and search share measure demand for each provider.
- Adoption is the share of GPU-cloud buyers using each provider.
- Share of spend is each provider’s share of total GPU-cloud dollars.
- Search share is each provider’s portion of the seven providers’ combined monthly search volume for their brand names.
Adoption and share of spend come from Ramp’s aggregated GPU-cloud card-spend data.38
CoreWeave leads all three signals in 2026 Q1, at 35.4% of buyers, 50.8% of dollars, and 31.3% of search. The widest divergence is between Crusoe and RunPod. Crusoe holds 27.7% of dollars on 3.0% of buyers and 0.9% of searches. RunPod accounts for 27.7% of buyers and 25.7% of searches, with 0.5% of spend.
Vultr accounts for 18.5% of brand search and 3.0% of spend in the 2026 Q1 comparison. Brand searches can include its other cloud products. The search data does not identify which searches concern GPU rentals.
How the search signals are measured
Search share sums to 100% within the seven shown providers over the 2025 Q3 to 2026 Q1 window where all three signals overlap. Each provider’s search is counted on its brand term:
For the price trajectories behind these providers, see the Cloud GPU rental price index.
Add as preferred sourceDeployment models
GPU rental services arrive in three deployment shapes. Each shape trades off control for convenience.
Serverless GPU
Serverless GPU services manage provisioning and scaling. Billing may include active execution, idle workers, CPU, memory or other resources, depending on the service. The pricing model can suit intermittent jobs, but total cost depends on the workload and configuration.
Serverless GPU providers include Replicate, RunPod Serverless, Modal, Fal.ai and Together. Compare the full job cost with a provisioned instance at the expected utilization; the catalog does not establish a universal utilization break-even point.3239404131
Virtual GPU (vGPU)
GPU virtual machines provide a virtualized host environment. GPU access may use a whole device or a partition, depending on the instance. Check the allocation method, memory and interconnect before comparing it with a dedicated server.
Bare-metal GPU
Bare-metal GPU services provide a dedicated physical server with direct hardware access. They can support workloads needing control over the host configuration and network. A GPU instance family name alone does not establish that the service is bare metal.
FAQs
Hyperscaler here groups broad cloud platforms, including smaller general-purpose clouds. Neocloud groups GPU and AI specialists. The chart compares observed rates by GPU, billing tier and month; no fixed price multiple applies across all listings.
Serverless can suit intermittent jobs and applications that benefit from managed scaling. Compare active and idle billing, cold starts, CPU and memory charges, and expected utilization against a provisioned instance.
Check the selected data-center region, data-processing terms, support access and any cross-border transfers. A provider’s headquarters alone does not establish where a workload or its data will run. IONOS, OVHcloud, Scaleway, Nebius, Verda and Seeweb are providers to evaluate.
Further reading
- Multi-GPU Benchmark: B200 vs H200 vs H100 vs MI300X
- GPU Concurrency Benchmark
- Top 25+ AI Chip Makers: NVIDIA & Its Competitors
- Cloud GPU Rental Price Index
Cite this research
Pick the format that matches where you're publishing. Pasting the link version into your CMS preserves the backlink.
@misc{dogan2026,
author = {Dogan, Sedat and Sarı, Ekrem},
title = {{Top 70+ Cloud GPU Providers}},
year = {2026},
month = sep,
howpublished = {\url{https://aimultiple.com/cloud-gpu-providers}},
note = {AIMultiple. Retrieved September 16, 2026}
}Results and timestamps of 3.8 thousand data points. Download the summary data shown in this article's charts and tables as a ZIP file containing 5 CSV files.
Want the granular data behind it? Join Premium
Changelog
22 updatesRemoved the Pricing by Provider Tier section from the cloud GPU providers comparison.
Added a Provider demand section on adoption, spend, and search share.
Replaced the provider list with a tiered comparison table covering 64 providers' models, combinations and billing tiers.
Added AWS EC2 G7e with NVIDIA Blackwell GPUs to Recent Updates.
Reference Links
- Has 20 years of experience as a white-hat hacker and development guru, with extensive expertise in programming languages and server architectures.
- Is a board advisor at a VC investing in early-stage technology firms and at Ödeal, a regional digital payment platform serving 125,000 merchants.
- Has led the technology infrastructure and cybersecurity of seven national elections, and has been recognized in the cybersecurity Hall of Fame by global technology leaders including Twitter.
Comments 4
Share Your Thoughts
Your email address will not be published. All fields are required. Comments are left in their original language.
Nice article, Cem! Could you add Koyeb and a few other serverless GPU providers?
Sure, thank you for the suggestion, we will consider it in the next edit.
Hi Cem, please also check out Dataoorts at https://dataoorts.com. We'd greatly appreciate being listed here.
Sure, we'll review to see if we can include Dataoorts in the next edit.
Hi Cem, we just launched Atlascloud.ai with the lowest H100 pricing on internet 2.48 on demand. Would love to get on your list.
Sure, we'll be reaching out to understand what Atlascloud.ai is offering.
Where is Nebius.ai ???
Thank you! It is added now.