Alternatives
GPU cloud providers, compared on the prices they publish
30 providers, their own published rates, turned into the only number that matters for a machine you keep busy: what it costs for a month. No affiliate links, no cherry-picking — every figure below is dated and linked to its source.
The short version
An hourly GPU is priced for a machine you switch off. Ours is priced for a machine you keep. Below 730 hours a month, most of these providers beat us — above it, the gap opens fast. Here is the same H100 SXM, everywhere, for one month.
| Provider | Published rate, H100 SXM | One month, 24/7 | vs our $1,249/mo |
|---|---|---|---|
 GPU Cloud HQDedicated, billed monthly
|
$1,249 per month, flat | $1,249 | Our reference |
Microsoft AzureHyperscaler
|
$98.32/h for ND96isr H100 v5 (8× H100, East US) — $12.29 per GPU-hour · Spot $2.27 | $8,972 | −86% |
Google CloudHyperscaler
|
$88.49/h for a3-highgpu-8g (8× H100) — $11.06 per GPU-hour · Spot $6.60 | $8,074 | −85% |
AWS (Amazon EC2)Hyperscaler
|
$6.88/h for p5.4xlarge (1× H100) · $55.04/h for eight | $5,022 | −75% |
CoreWeaveEnterprise AI cloud
|
$49.24/h per 8-GPU HGX H100 node — $6.16 per GPU-hour | $4,493 | −72% |
NebiusFull-stack AI cloud
|
$4.50/h on demand · spot from $0.79 | $3,285 | −62% |
DigitalOcean (Paperspace)General cloud with GPU Droplets
|
$4.41/GPU/h on demand · $3.26 with a 12-month commitment | $3,219 | −61% |
Verda (DataCrunch)European GPU cloud
|
$4.17/h on demand · $2.08 spot | $3,044 | −59% |
RunPodOn-demand pods, two trust tiers
|
$3.99/h Secure Cloud (SXM) · $2.69 Community | $2,913 | −57% |
LambdaOn-demand cloud from a GPU builder
|
$3.99–$4.29/h (SXM) depending on instance size | $2,913 | −57% |
Together AIInference platform with GPU clusters
|
$3.99/h on demand · $3.69 reserved 3–30 days · $1.99 preemptible | $2,913 | −57% |
Crusoe CloudAI cloud with its own data centers
|
$3.90/GPU-h on demand (HGX) | $2,847 | −56% |
ScalewayEuropean cloud, billed per minute
|
€6.62/h for H100-SXM-2-80G (2× H100 SXM) — €3.31 ($3.71) per GPU-hour · H100 PCIe €2.87/h | $2,708 | −54% |
JarvislabsPer-minute GPU cloud, India
|
$3.49/h on demand (SXM) · $1.54 spot | $2,548 | −51% |
Novita AIGPU containers and model APIs
|
$3.39/h (SXM) · $1.70 spot | $2,475 | −50% |
HyperstackEuropean GPU cloud
|
$3.20/h SXM on demand · from $2.72 reserved | $2,336 | −47% |
Thunder ComputeLow-cost GPU virtualization
|
$3.20/h (SXM), 1–8 GPUs | $2,336 | −47% |
Massed ComputeOn-demand GPU virtual machines
|
$3.14/h (SXM5) · $25.12/h for eight | $2,292 | −46% |
OVHcloudEuropean cloud, hourly or monthly billing
|
$2,070/mo for h100-380 (1× H100 PCIe) · $2.99/h by the hour | $2,070 | −40% |
HOSTKEYDedicated GPU servers, hourly or monthly
|
€1,590/mo ($1,774) — 1× H100 80 GB virtual server, pre-order | $1,774 | −30% |
VultrDeveloper cloud, hourly
|
$2.30/GPU/h for an 8× HGX H100 instance | $1,679 | −26% |
TensorDockMarketplace of independent hosts
|
from $2.25/h (SXM5) | $1,643 | −24% |
Vast.aiOpen GPU marketplace
|
median $2.20/h (14 Oct) · offers from $1.73 | $1,606 | −22% |
LeasewebDedicated servers, monthly contracts
|
CA$2,078.74/mo ($1,463) — 1× H100 NVL 94 GB (PCIe), Montreal | $1,463 | −15% |
Voltage ParkH100 fleet at scale
|
from $1.99/h for a single H100, no contract | $1,453 | −14% |
Cherry ServersBare-metal cloud, Lithuania
|
A100: from €1,416.82/mo ($1,658) — A100 80 GB in a custom bare-metal server, waiting list | $1,658 | −61%vs our A100 at $649/mo |
HetznerDedicated servers, Germany and Finland
|
RTX PRO 6000: €999/mo ($1,199) — GEX63, RTX PRO 6000 Blackwell Max-Q, plus a $1,199 setup fee | $1,199 | −38%vs our RTX PRO 6000 at $749/mo |
Salad CloudDistributed consumer GPUs
|
RTX 5090: $0.50/h High priority · $0.25 batch | $365 | −24%vs our RTX 5090 at $279/mo |
Cudo ComputeQuote-based capacity
|
Quote on request | — | Not published |
FluidstackFrontier cluster builder
|
Quote on request | — | Not published |
BitLaunchCrypto-paid reseller of cloud VPS
|
Quote on request | — | Not published |
Rate shown is each provider's standard on-demand tier for the GPU in the row (not spot, not preemptible, not community hardware), read on their own pricing page on 15 October 2026. Monthly equivalent assumes the machine stays up for 730 hours; for providers that publish a monthly price, it is that price. Providers without a published H100 price are listed after the others, with the GPU compared named in the row. Where a provider is cheaper than us, the row says so. GPU by GPU rather than provider by provider: the GPU cloud pricing index.
Every provider, audited
What each one is good at, where it hurts, and who should pick it instead of us.
-
Microsoft Azureazure.microsoft.com
−86%
Microsoft's cloud sells GPUs as ND- and NC-series virtual machines, from fractional RTX PRO 6000 sizes to 8-GPU H100 and H200 nodes, with reservations, savings plans and Spot.
$8,972 a month for one H100 at their published rateCompare
-
Google Cloudcloud.google.com
−85%
Google's general-purpose cloud sells GPUs bundled into accelerator-optimized virtual machines, with per-second billing, Spot and Flex-start capacity, and quotas to request first.
$8,074 a month for one H100 at their published rateCompare
-
AWS (Amazon EC2)aws.amazon.com
−75%
Amazon EC2's GPU instances — P5 for H100 and H200, P6 for Blackwell, G6 and G6e for L4 and L40S — billed per second, behind a GPU quota that starts at zero.
$5,022 a month for one H100 at their published rateCompare
-
CoreWeavecoreweave.com
−72%
Large-scale, Kubernetes-native infrastructure sold by whole 8-GPU nodes, aimed at funded AI companies with a procurement process.
$4,493 a month for one H100 at their published rateCompare
-
Nebiusnebius.com
−62%
A large European AI cloud with its own data centers, published on-demand and spot rates, and multi-month cluster reservations.
$3,285 a month for one H100 at their published rateCompare
-
DigitalOcean (Paperspace)digitalocean.com
−61%
Paperspace folded into DigitalOcean: GPU Droplets next to the databases, load balancers and object storage of a general-purpose cloud.
$3,219 a month for one H100 at their published rateCompare
-
Verda (DataCrunch)verda.com
−59%
The provider formerly known as DataCrunch — Finnish data centers, published spot and on-demand rates, and reservations from one month to two years.
$3,044 a month for one H100 at their published rateCompare
-
RunPodrunpod.io
−57%
Per-second pods on a Secure Cloud of vetted data centers and a cheaper Community Cloud of third-party hosts.
$2,913 a month for one H100 at their published rateCompare
-
Lambdalambda.ai
−57%
The cloud arm of a company that has built GPU workstations and servers for a decade — formerly Lambda Labs — with 1–8 GPU instances and a familiar ML image.
$2,913 a month for one H100 at their published rateCompare
-
Together AItogether.ai
−57%
Best known for serverless model APIs, with dedicated GPU clusters sold alongside them by the hour or by reservation.
$2,913 a month for one H100 at their published rateCompare
-
Crusoe Cloudcrusoe.ai
−56%
A vertically integrated AI cloud running its own data centers, with published on-demand rates for H100, H200, A100 and L40S and reserved capacity through sales.
$2,847 a month for one H100 at their published rateCompare
-
Scalewayscaleway.com
−54%
A Paris-based cloud renting L4, L40S, H100 and B300 GPU instances by the minute in France and Poland, and dual-GPU Dedibox servers by the month.
$2,708 a month for one H100 at their published rateCompare
-
Jarvislabsjarvislabs.ai
−51%
A per-minute GPU cloud with instances you can pause and resume, spot and 1–12-month reserved tiers, and every server located in India.
$2,548 a month for one H100 at their published rateCompare
-
Novita AInovita.ai
−50%
An AI platform selling model APIs alongside per-second GPU container instances in many regions, with spot capacity and 8-GPU bare metal on contract.
$2,475 a month for one H100 at their published rateCompare
-
Hyperstackhyperstack.cloud
−47%
NexGen Cloud's self-serve platform, billed by the minute, with published discounts for reserved capacity.
$2,336 a month for one H100 at their published rateCompare
-
Thunder Computethundercompute.com
−47%
A low-cost GPU cloud built on its own GPU virtualization stack, with H100, A100 and L40 instances in North America billed by the minute.
$2,336 a month for one H100 at their published rateCompare
-
Massed Computemassedcompute.com
−46%
An on-demand GPU cloud with a broad catalogue — from the L40S to 8-GPU B300 instances — billed from a prepaid balance, with Bitcoin accepted.
$2,292 a month for one H100 at their published rateCompare
-
OVHcloudovhcloud.com
−40%
The French cloud provider rents H200, H100, A100, L40S and L4 GPUs as Public Cloud instances, billed by the second or at a monthly price, and L4 bare-metal servers by the month.
$2,070 a month for one H100 at their published rateCompare
-
HOSTKEYhostkey.com
−30%
A European dedicated-server host renting GPU machines by the hour or the month — RTX 4090 and 5090 bare metal, RTX PRO 6000, H100 and A100 virtual servers — paid by card or crypto, behind a mandatory KYC check.
$1,774 a month for one H100 at their published rateCompare
-
Vultrvultr.com
−26%
A developer cloud renting NVIDIA and AMD GPUs as virtual machines and bare metal, billed by the hour — and one of the few large clouds that takes crypto, through BitPay.
$1,679 a month for one H100 at their published rateCompare
-
TensorDocktensordock.com
−24%
A broker between people with idle GPUs and people who need them, with KVM virtual machines, root access and a prepaid balance.
$1,643 a month for one H100 at their published rateCompare
-
Vast.aivast.ai
−22%
A bidding marketplace where independent hosts and data centers set their own rates, with on-demand, interruptible and 1–6 month reserved options.
$1,606 a month for one H100 at their published rateCompare
-
Leasewebleaseweb.com
−15%
A Dutch hosting company renting dedicated GPU servers on monthly contracts — L4, L40S, RTX PRO 6000 and the PCIe NVL versions of the H100 and H200 — in Europe, North America and Asia.
$1,463 a month for one H100 at their published rateCompare
-
Voltage Parkvoltagepark.com
−14%
A large H100 estate — now merged with Lightning AI — rented by the hour on Ethernet, by the node on InfiniBand, or on long-term contracts.
$1,453 a month for one H100 at their published rateCompare
-
Cherry Serverscherryservers.com
−61%
A Lithuanian bare-metal provider that adds GPUs to custom dedicated servers, billed hourly, monthly or yearly, with crypto accepted through CoinGate and BVNK.
$1,658 a month for one A100 at their published rateCompare
-
Hetznerhetzner.com
−38%
The German hosting company rents dedicated servers with one workstation-class GPU — up to an RTX PRO 6000 Blackwell Max-Q — by the hour up to a monthly cap, with a setup fee of about a month.
$1,199 a month for one RTX PRO 6000 at their published rateCompare
-
Salad Cloudsalad.com
−24%
A distributed cloud running containers on gaming GPUs in homes and small businesses, with four priority tiers and some of the lowest RTX prices anywhere.
$365 a month for one RTX 5090 at their published rateCompare
-
Cudo Computecudocompute.com
An NVIDIA Cloud Partner aggregating capacity across partner data centers. Its public on-demand platform closed on 31 March 2026; every GPU is now quote on request.
No public price listCompare
-
Fluidstackfluidstack.io
Once a self-serve GPU cloud, now a builder of large dedicated AI data centers for frontier labs — with no public price list and no self-serve signup on its site.
No public price listCompare
-
BitLaunchbitlaunch.io
A reseller of DigitalOcean, Vultr and Linode servers paid in crypto with no KYC — whose only GPU offer is fractions of an NVIDIA A40 on Vultr, without a public price.
No public price listCompare
How we compare — and when to ignore us
Every comparison on this site follows three rules, because a comparison you cannot check is an advertisement.
- Their number, not oursWe quote the rate printed on the provider's own pricing page, link to it, and date the reading. Nothing is estimated. When a provider publishes no price, the table says « quote on request » rather than guessing.
- The same workloadA monthly rental is one machine, always on. So the hourly rate is multiplied by 730 hours — the industry convention for a full month — and compared with our flat monthly price for the same GPU.
- The honest caveatIf your GPUs run a few hours a week, per-second billing wins and you should use it. Spot and interruptible tiers are cheaper still. We are the better deal when the machine stays busy — and we say so instead of hiding it.
Run the numbers on your own workload
Pick a GPU, a term and a region — the price you see is the price you pay, every month.