HPC providers, side by side
One H100 lists from $1.99 per GPU-hour at Voltage Park to $12.29 on Azure across 35 HPC providers (checked 9–11 Nov 2026). Fourteen run InfiniBand, seventeen offer Slurm, thirteen host in the EU, and three public programmes allocate supercomputer time. Every cell links to the page it came from.
Last verified
| Schedulers | GPUs / accelerators | Regions | ||||||
|---|---|---|---|---|---|---|---|---|
| Voltage Park | gpu-cloud | InfiniBand | Kubernetes | H100, H200, B200, GB300 | $1.992 | Yes | US; six Tier 3+ data centers across four states (company about page) | 2026-11-09 |
| Denvr Dataworks | gpu-cloud | InfiniBand | — | GB200, B200, H200, GH200, H100, A100 | $2.451 | Yes | Canada (Calgary) | 2026-11-09 |
| Core42 | gpu-cloud | InfiniBand | Slurm, Kubernetes | H100, H200, B200, GB300, MI300X, MI355X | $2.501 | Yes | UAE; US (New York deployment); Europe (Italy, France; European HQ in Dublin). Company cites 9 global data centers. | 2026-11-09 |
| GMI Cloud | gpu-cloud | InfiniBand | Slurm, Kubernetes | H100, H200, B200, GB200, GB300 | $2.601 | Yes | North America, Europe, Asia-Pacific (company site; specific sites not named); Taiwan AI factory (announced Nov 2025); Kagoshima, Japan site planned (announced Mar 2026) | 2026-11-09 |
| OVHcloud | sovereign | — | Kubernetes | H200, H100, A100, L40S, L4 | $2.991 | — | GPU instances in Gravelines (France) and Beauharnois/Montreal (Canada) per regions page; 46 data centers overall | 2026-11-09 |
| Vultr | gpu-cloud | — | Slurm, Kubernetes | B200, H100, GH200, A100, L40S, MI355X, MI325X, MI300X | $2.991 | Yes | 34 cloud data center regions worldwide, latest Chennai, India (Oct 2026, company blog) | 2026-11-09 |
| Hyperstack | gpu-cloud | — | Kubernetes | B300, B200, GB300, H200, H100, A100, RTX Pro 6000, L40 | $3.201 | — | Europe and North America (sites not named) | 2026-11-09 |
| Crusoe | gpu-cloud | InfiniBand | Slurm, Kubernetes | Vera Rubin NVL72, GB300, B300, GB200, B200, H200, H100, A100, L40S, MI355X, MI300X | $3.901 | — | Data center sites named in company releases: Abilene and Armstrong County (Texas), Wyoming, Norway (with Polar); Paraguay deployment announced Oct 2026 | 2026-11-09 |
| Nebius | gpu-cloud | InfiniBand | Kubernetes | GB300, B300, GB200, B200, H200, H100, RTX PRO 6000, L40S | $4.501 | — | Finland, France (2), Spain (Madrid), UK (2), Israel, US (Kansas City MO; Woodbury MN); private region in Iceland (company docs, Nov 2026) | 2026-11-09 |
| Lambda | gpu-cloud | InfiniBand | Slurm, Kubernetes | B200, H100, GH200, A100 | $4.891 | — | On-demand regions in the US (Virginia, Washington DC, Illinois, Texas x3, California, Arizona, Utah), Japan (Tokyo, Osaka), India, Germany, Israel (company docs, Nov 2026); new data centers announced in Mayes County, Oklahoma (Sep 2026) and Lowndes County, Alabama (Nov 2026) | 2026-11-09 |
| Verda | gpu-cloud | InfiniBand | Slurm, Kubernetes | GB300, B300, B200, H200, H100, A100, L40S, RTX PRO 6000 | $5.0871 | — | EU (company site states EU data centers; sites not named on site). Offices in Helsinki, London, San Francisco. | 2026-11-11 |
| CoreWeave | gpu-cloud | InfiniBand | Slurm, Kubernetes | GB300, GB200, B300, B200, RTX PRO 6000, GH200, H200, H100, L40S, L40, A100 | $6.161 | Yes | US and Europe; 50+ active data centers (company site, Nov 2026) | 2026-11-09 |
| AWS HPC | hyperscaler | EFA | AWS Parallel Computing Service (managed Slurm), AWS ParallelCluster (open source), AWS Batch | NVIDIA H100 (P5), NVIDIA H200 (P5en), 8 per instance | $6.888 | — | — | 2026-11-11 |
| Oracle Cloud HPC | hyperscaler | RDMA | — | 8x NVIDIA H100 80 GB (BM.GPU.H100.8) | $10.003 | Yes | — | 2026-11-11 |
| IBM Cloud | hyperscaler | RoCE | IBM Spectrum LSF, IBM Spectrum Symphony | NVIDIA B300, H200, H100 SXM5, A100 PCIe, L40S, L4; AMD Instinct MI300X; Intel Gaudi 3 | $10.733 | Yes | H100 profile: Dallas, Washington DC, Toronto, Sao Paulo, Frankfurt, Madrid, Paris, London, Sydney, Tokyo; H200 adds Chennai | 2026-11-11 |
| Azure HPC | hyperscaler | InfiniBand | Azure CycleCloud (Slurm, PBS Pro, LSF, Grid Engine, HTCondor), Azure Batch | NVIDIA H100 80 GB, 8 per ND H100 v5 VM | $12.295 | — | — | 2026-11-11 |
| Advania (Gompute) | managed-hpc | — | — | — | — | Yes | Nordic data centres (cities not named); Iceland on-premises HPC deployments | 2026-11-11 |
| Alibaba Cloud E-HPC | hyperscaler | RDMA | E-HPC NEXT (Slurm cluster type) | NVIDIA L20, A10, V100; Blackwell-based families (model not named on the page) | — | Yes | — | 2026-11-11 |
| Bull HPC as a Service (JARVICE, formerly Nimbix) | simulation-cloud | — | JARVICE platform with unified HPC/AI workload scheduling; runs on Kubernetes and dispatches jobs to downstream Slurm clusters | — | — | — | — | 2026-11-11 |
| Cirrascale Cloud Services | gpu-cloud | — | — | GB300, B300, B200, RTX PRO 6000, MI355X | — | — | US (US-West, US-East, US-South per company about page) | 2026-11-09 |
| CUDO Compute | gpu-cloud | InfiniBand | — | H100, H200, B100, B200, B300, GB200, GB300, VR200 | — | Yes | North America, Europe, UK, Middle East/MENA (company site; cities not named) | 2026-11-09 |
| EuroHPC JU access | public-supercomputer | — | — | — | — | — | LUMI, Leonardo, MareNostrum 5, JUPITER, MeluXina, Discoverer, Deucalion, Arrhenius | 2026-11-11 |
| Google Cloud HPC | hyperscaler | Ethernet | Cluster Toolkit (open source) with Slurm | A3 Mega and High: H100; A3 Ultra: H200; A4: B200; A4X: GB200 and GB300 | — | — | — | 2026-11-11 |
| Hot Aisle | gpu-cloud | — | — | MI300X, MI355X | — | Yes | US (Michigan) | 2026-11-09 |
| LUMI (CSC) | public-supercomputer | Slingshot | Slurm (partitions and quickstart in the LUMI docs) | LUMI-G: AMD Instinct MI250X (a GPU-hour is a full MI250X module, 2 GCDs) | — | — | — | 2026-11-11 |
| Nscale | gpu-cloud | — | Slurm, Kubernetes | GB300 | — | Yes | Norway (Glomfjord, Narvik, Stavanger, Oslo), UK (Loughton, Hayes), Portugal (Sines), Iceland (Keflavik, Blonduos), US (Texas, North Carolina, West Virginia); mix of Nscale-operated and partner sites | 2026-11-09 |
| NSF ACCESS | public-supercomputer | Slingshot | — | Delta: NVIDIA A100 (40 GB), A40, H200 (141 GB); AMD MI100 | — | — | Delta at NCSA, Illinois | 2026-11-11 |
| Penguin On Demand (POD) | managed-hpc | — | ClusterWare cluster management software (no job scheduler named on the page) | — | — | Yes | One US Tier III data centre (city not named) | 2026-11-11 |
| Rescale | simulation-cloud | — | — | — | — | — | — | 2026-11-11 |
| Sabalcore | managed-hpc | InfiniBand | — | — | — | Yes | — | 2026-11-11 |
| Scaleway | sovereign | — | — | H100, B300, L40S, L4 | — | Yes | EU (Paris fr-par-1/fr-par-2, Warsaw pl-waw-2) | 2026-11-09 |
| STACKIT | sovereign | — | — | NVIDIA H100 HGX 80 GB (up to 8 per VM), H100 NVL 94 GB, A100 PCIe 80 GB, L40S 48 GB | — | — | European data centres (EU01 server naming) | 2026-11-11 |
| T Cloud Public (Open Telekom Cloud) | sovereign | InfiniBand | Slurm, Moab/NODUS cloud bursting, UNIVA, SGE | NVIDIA P100, V100, T4 (older generations; no H100 listed) | — | Yes | Germany (Biere, Magdeburg), Netherlands, Switzerland | 2026-11-11 |
| Taiga Cloud (Northern Data) | gpu-cloud | InfiniBand | Slurm, Kubernetes | NVIDIA H100 SXM and H200 SXM | — | Yes | UK (Chester, London), Portugal (Sines), Sweden (Boden), Norway (Kristiansand, Lefdal) | 2026-11-09 |
| TensorWave | gpu-cloud | RoCE | Slurm, Kubernetes | MI300X, MI325X, MI355X, MI455X | — | Yes | North America (sites not listed on official site) | 2026-11-09 |
Shortlist by what you need
How this table is built
Full methodologyEvery value comes from a public source: the provider’s pricing page, docs, filings or a named report. The small marker next to a figure opens that source. Empty cells mean the provider does not publish the figure; we never estimate.
Each row carries the date we last checked it, and the newest check was 11 Nov 2026. Spot a stale number? Email hello@hpcproviders.com with the source and we update the row.
Free to reuse under CC BY 4.0 with a link to this page: CSV, JSON, llms.txt.
Questions buyers ask
What is an HPC provider?
An HPC provider rents tightly coupled compute: nodes joined by a low-latency fabric such as InfiniBand, AWS EFA or Slingshot, scheduled with Slurm or a similar job scheduler, with CPU or GPU nodes you can book by the hour, by the core-hour or by the month. This table covers hyperscaler HPC services, GPU clouds that sell clusters, managed HPC services, EU sovereign clouds and publicly funded supercomputers.
Which HPC providers offer InfiniBand?
Fourteen providers state InfiniBand as their cluster fabric: Azure HPC, T Cloud Public, CoreWeave, Lambda, Crusoe, Nebius, Voltage Park, Verda, Taiga Cloud, Denvr Dataworks, GMI Cloud, Core42, Cudo Compute and Sabalcore. AWS uses its own Elastic Fabric Adapter, TensorWave and IBM Cloud use RoCE, Oracle and Alibaba name RDMA cluster networking, and LUMI and NSF ACCESS Delta use HPE Slingshot.
How much does an H100 cost per hour for HPC?
Published list prices for one H100 range from $1.99 per GPU-hour at Voltage Park (a starting price) through $2.45 at Denvr Dataworks, $2.99 at OVHcloud and Vultr, $3.90 at Crusoe and $4.50 at Nebius, to $6.16 at CoreWeave and $6.88 on AWS. Oracle lists $10.00, IBM Cloud $10.73 and Azure $12.29. The basis differs by provider (on-demand, a starting price, or an 8-GPU price divided by 8), and it sits next to every price on the provider page. AWS Capacity Blocks bring an H100 down to $5.97.
Which providers offer Slurm?
Seventeen providers offer or document Slurm: AWS (Parallel Computing Service), Azure (CycleCloud), Google Cloud (Cluster Toolkit), Alibaba Cloud E-HPC, T Cloud Public, LUMI, CoreWeave (SUNK), Lambda, Crusoe, TensorWave, Verda, Vultr, Nscale, Taiga Cloud, GMI Cloud, Core42 and Bull JARVICE, which dispatches jobs to Slurm clusters. IBM Cloud builds its HPC clusters on IBM Spectrum LSF and Symphony instead.
Can I get HPC time for free?
Yes, on public systems. EuroHPC JU allocates time on LUMI, Leonardo, MareNostrum 5 and JUPITER through competitive calls, with an industry track and AI Factories calls for startups. NSF ACCESS gives US researchers and educators free allocations on systems such as NCSA Delta. LUMI also sells company time through CSC at EUR 0.535 per GPU-hour.
Which HPC providers host in the EU?
Thirteen providers name EU capacity: T Cloud Public (Germany), STACKIT, OVHcloud (France), Scaleway (France, Poland), Nebius (Finland, France, Spain), Verda, Nscale (Portugal), Taiga Cloud (Portugal, Sweden), Lambda (Germany), Core42 (Italy, France), IBM Cloud (Frankfurt, Madrid, Paris), LUMI (Finland) and EuroHPC JU.