Start a request

H200 CLOUD PRICING

H200 rental pricing.
Compare the actual offer.

On demand, 11 providers we track list H200 from $3.99 to $6.31 per GPU-hour, checked September 23 to September 28, 2026. Compare each rate with its configuration, billing unit and purchase terms.

Compare H200 offers ↓Explore H200 providers ↗

PUBLISHED OFFER SNAPSHOTS

H200 price per GPU-hour.
By provider, with terms.

Alphabetical by provider, not a ranking. Dates apply to individual records. Compare currencies, minimum allocations and operating models before comparing the rate.

Dated H200 offers. Billing units and commitments differ.
ProviderPublished price and billing unitConfiguration and termsEvidence
AceCloudGPU cloud₹381.46INR · 1-GPU instance-hourPublished hourly plan

H200 NVL, 16 vCPU, 128 GB RAM, Noida. Monthly alternative ₹222775. Taxes excluded. Other regions require separate pricing.

Published · checked 2026-09-24Official source Capacity needs confirmation
BeamGPU cloudFrom $2.09USD · machine-hour for listed 141 GB H200 configurationOn-demand starting price

H200 SXM5, 24 vCPU, 192 GB RAM, 2 TB NVMe included. Multi-node clusters require quote.

Published · checked 2026-09-24Official source Capacity needs confirmation
CoreWeaveGPU cloud$50.44USD · 8-GPU instance-hour$6.31 / GPU-hour · calculatedOn-demand

North America example: 8 HGX H200 GPUs, 128 vCPUs, 2,048 GB RAM and 61.44 TB local storage. $6.31/GPU-hour is calculated; the instance is $50.44/hour.

Published · checked 2026-09-24Official source Capacity needs confirmation
CrusoeGPU cloud$4.29USD · GPU-hourOn-demand

H200 HGX 141 GB. Multiply by the GPU count in your instance. Managed inference is a different service and rate.

Published · checked 2026-09-24Official source Capacity needs confirmation
DigitalOceanGPU cloud$4.47USD · GPU-hourOn-demand

HGX H200; 1 or 8 GPUs per Droplet. $3.40/GPU-hour requires 12-month reservation.

Published · checked 2026-09-23Official source Capacity needs confirmation
E2E NetworksGPU cloud$3.99USD · GPU-hourOn-demand

H200 141 GB, 30 vCPUs and 375 GB RAM. Published from $3.99/GPU-hour; confirm purchase term and selected configuration. Taxes excluded.

Published · checked 2026-09-24Official source Capacity needs confirmation
EdgevanaMarketplaceFrom $3.86USD · GPU-hourMarketplace starting price

H200 NVL listing shows $3.86/hour total. Other variants, locations and reserved offers differ. Confirm the exact listing, GPU count and term before ordering.

Published · checked 2026-09-24Official source Capacity needs confirmation
GcoreGPU cloud€21.84EUR · 8-GPU cluster-hour€2.73 / GPU-hour · calculatedCalculator example; term to confirm

8x H200 SXM, 1128 GB VRAM; VAT excluded. Confirm commitment/selected pricing model before treating as on-demand.

Published · checked 2026-09-23Official source Capacity needs confirmation
GMI CloudGPU cloudFrom $2.60USD · GPU-hourAdvertised starting price

Page offers on-demand and reserved capacity but does not identify which commitment or region unlocks the starting H200 rate. Confirm before normalizing.

Published · checked 2026-09-23Official source Capacity needs confirmation
HyperstackGPU cloud$3.99USD · GPU-hourOn-demand

H200 SXM. Product page describes an 8-GPU VM. Reserved rate from $2.79/GPU-hour is separate.

Published · checked 2026-09-23Official source Capacity needs confirmation
JarvislabsGPU cloud$3.99USD · GPU-hourOn-demand

H200 141 GB. Single GPU and larger configurations; storage extra. Regional rates and availability vary.

Published · checked 2026-09-23Official source Capacity needs confirmation
ModalServerless$0.001261USD · GPU-second$4.54 / GPU-hour · calculatedServerless GPU tasks

H200 SXM; about $4.54/GPU-hour calculated as rate x3600. CPU and memory charged separately. Distinct from an all-in dedicated VM rental.

Published · checked 2026-09-23Official source Capacity needs confirmation
NebiusGPU cloud$4.50USD · GPU-hour$5.40 from October 1, 2026On-demand

HGX H200 with 16 vCPUs and 200 GB RAM per GPU. Published price increases to $5.40/GPU-hour on October 1, 2026. Confirm the GPU count and applicable rate for your start date.

Published · checked 2026-09-28Official source Capacity needs confirmation
OblivusGPU cloud$3.99USD · GPU-hourOn-demand

H200 SXM5 141 GB. Page says multiply price by GPU count, CPU/RAM/storage included while running. Published availability page showed zero H200 in its displayed regions; no inventory guarantee.

Published · checked 2026-09-23Official source Capacity needs confirmation
OVHcloudGPU cloudFrom $49.56USD · 8-GPU instance-hour$6.20 / GPU-hour · calculatedPublished estimated starting price

Calculated approximately $6.20/GPU-hour, but requires the 8-GPU instance. Region/currency-specific rates; not a purchasable $6.20 single-GPU offer.

Published · checked 2026-09-23Official source Capacity needs confirmation
RunpodGPU cloud$4.59USD · GPU-hourOn-demand Pod

H200 Pod example: 141 GB VRAM, 24 vCPUs, 276 GB RAM. Storage, Serverless and Clusters have separate pricing.

Published · checked 2026-09-24Official source Capacity needs confirmation
SpheronMarketplaceFrom $5.44USD · GPU-hourDedicated/on-demand marketplace starting price

Partner-supply marketplace. Per-minute billing after 20-minute minimum. Spot $2.85 is separate. Current page supersedes indexed $4.80.

Published · checked 2026-09-23Official source Capacity needs confirmation
STNGPU cloudRequest project quoteproject quoteReserved; published three-year minimum

Single-tenant GPU One capacity. Published plans include networking and egress; storage is separate. Confirm node count, location and current terms.

Published · checked 2026-09-21Official source Capacity needs confirmation
Together AIGPU cloud$5.99USD · GPU-hourOn-demand GPU clusters

Published reserved rates $4.99 for 7–30 days, $4.15 for 31–90 days, $3.99 for 91–180 days; GPU cluster pricing, not dedicated endpoint pricing.

Published · checked 2026-09-23Official source Capacity needs confirmation
Vast.aiMarketplaceLive offersoffer dependentDynamic host marketplace

Live H200 offers vary by host, configuration and rental term. Check the marketplace for a current numeric rate.

Published · checked 2026-09-23Official source Capacity needs confirmation
Verda (formerly DataCrunch)GPU cloud$4.55USD · GPU-hourOn-demand

1x H200 SXM5, 141 GB; 44 CPUs, 170 GB RAM. Spot $2.27 is separate. Rendered pricing page checked; search snippets may show older rates.

Published · checked 2026-09-24Official source Capacity needs confirmation

These are published snapshots, not live inventory or guaranteed quotes. Serverless GPU-only rates, marketplace listings and whole-instance prices cover different purchases. Confirm storage, transfer, support, tax and start-date terms.

WORK THROUGH IT WITH AN ADVISOR

Make the quote fit your workload.

Bring your workload, launch date and what could change. We can help you evaluate capacity, compare proposals and identify the deployment support you need.

Discuss my project ↗

02 / BUILD YOUR SHORTLIST

Can the provider grow
with your workload?

Businesses often cannot predict GPU demand precisely. Our advisory priority is to establish how a provider can support changing requirements, not just whether it can supply the first allocation.

PLAN FOR A RANGE, NOT ONE FORECAST

Capacity today is only the first question.

Bring a baseline, a likely growth range and a demand-surge scenario. Ask each provider to explain what is available now, what can be added, how quickly it can arrive and what must be reserved. Treat unconfirmed expansion capacity as an open project risk.

START

Launch capacity

Confirm the exact GPU count, configuration, region and ready date. Make sure the allocation fits the pilot and the initial production workload.

BURST

Unexpected demand

Ask about account limits, maximum concurrent capacity and time to add GPUs. Agree what happens if the preferred hardware or region is full.

EXPAND

Sustained growth

Request an expansion path with notice periods, allocation increments, pricing and any reservation commitment. Include whether growing means moving the workload or data.

The question to put in every proposal: “If our workload grows beyond the initial estimate, what additional capacity can you commit to, by when, in which region, and on what terms?”

Separate the ability to request more resources from a commitment to supply them. For example, Modal’s scaling documentation describes autoscaling controls alongside platform limits. Validate both the software limits and the capacity plan for the service you choose.

01 / PROVE THE WORKLOAD

A pilot with uncertain demand

Start with the smallest suitable allocation and a defined test. Use the results to refine your growth range, then check the path to the next allocation before committing to a longer term.

Evidence to request: the minimum purchase, full pilot cost and a path to the next GPU count.

See the single- vs eight-GPU decision →
02 / COMPLETE A TRAINING RUN

A job spanning several servers

Compare the entire cluster. Ask for the links between GPUs within a server and the network between servers, plus the storage path for datasets and checkpoints.

Evidence to request: a topology diagram and a representative run using your software, data and target GPU count.

Explore cluster evaluation →
03 / SERVE REAL USERS

Inference with changing traffic

Define response-time and throughput targets. Test quiet periods and bursts, not just a warmed-up demo. Compare a persistent machine with a serverless deployment using the same traffic pattern.

Evidence to request: measured response times and the full bill, including the capacity kept ready between requests.

See the serverless decision →
04 / GET INTO PRODUCTION

A team that needs deployment help

Write down who will configure the environment, deploy the model, monitor it and restore service. If nobody on your team owns a task, ask for it to be scoped and priced before choosing a provider.

Evidence to request: a named implementation owner, support boundaries, launch milestones and acceptance criteria.

Discuss the help your project needs →

A RATE IS NOT AN ALLOCATION

Budget for the machine you must buy.

Our CoreWeave snapshot is $50.44 per hour for an eight-GPU H200 node. An illustrative 100-hour allocation costs $5,044 in node charges, even if your application uses only part of that node. Additional services and taxes are excluded. The approximately $6.31 per-GPU figure is a comparison calculation, not a one-GPU order price.

CoreWeave pricing, checked 2026-09-24 ↗

Technical context: NVIDIA’s H200 configurations and Modal’s container startup guidance. This framework is procurement guidance, not a claim that we have benchmarked these providers.

H200 QUESTIONS BY OFFER

Make the next
conversation specific.

These questions apply to the researched H200 offerings. Full company profiles, locations and customer examples now live in the provider directory.

AceCloud

Request the initial Noida allocation and a path to the next GPU count, including whether expansion changes the server, location, connectivity or scope of deployment help.

Full provider profile →

Beam

Ask how to grow from the bundled machine to the next allocation or a cluster, including lead time, region, storage migration and any new reservation requirement.

Full provider profile →

CoreWeave

Request the initial node count and an expansion plan that keeps the required network and storage design intact. Confirm notice periods and what capacity is committed at each stage.

Full provider profile →

Crusoe

Ask how the chosen compute or inference service handles a larger workload, which capacity requires advance notice and who owns changes to the application or cluster software.

Full provider profile →

DigitalOcean

Ask what changes when moving from one GPU to eight or to several machines: region, data migration, downtime, quota approval and reservation terms.

Full provider profile →

E2E Networks

Ask for the initial allocation and the additional GPUs available in the same site, with notice periods, connectivity requirements and complete costs including applicable taxes.

Full provider profile →

Edgevana

Ask whether additional matching machines can be secured from the same listing or operator, how long the offer is held and who handles support if expansion requires a different host.

Full provider profile →

Gcore

Request the terms behind the eight-GPU calculator example and the next capacity increment, including region, lead time, contract duration, currency and VAT.

Full provider profile →

GMI Cloud

Ask which initial and expanded H200 allocations qualify for the starting rate, in which facilities, on what dates and with what minimum commitment.

Full provider profile →

Hyperstack

Request the H200 flavor and region for launch and growth. Ask whether added VMs can use the required network and what notice or reservation is needed to secure them.

Full provider profile →

Jarvislabs

Ask for a pilot-to-production plan showing the region, next GPU counts, persistent storage and how capacity would be secured if demand rises faster than forecast.

Full provider profile →

Modal

Test a realistic demand surge. Confirm account and concurrency limits, expected time to add GPU containers, regional requirements and a bill including warm containers, CPU and memory.

Full provider profile →

Nebius

Request a dated launch and expansion quote, including the applicable October rate, GPU counts, region, required notice and explicit capacity commitments.

Full provider profile →

Oblivus

Confirm the actual launch allocation first, then request a realistic expansion plan and notice period. Document what happens to data and charges if compute must be stopped or moved.

Full provider profile →

OVHcloud

Ask how to add capacity beyond the first eight-GPU instance in the selected region, including account limits, availability, networking and changes to the full project cost.

Full provider profile →

Runpod

Ask whether extra Pods can access the required persistent data in the selected location, how capacity is secured and how checkpoints survive a change of machine.

Full provider profile →

Spheron

Ask whether growth can stay with the same partner and region, who owns incidents across partners and what capacity is committed versus dependent on future marketplace offers.

Full provider profile →

STN

Request a staged allocation plan, growth rights, deployment responsibilities and the full commitment cost before choosing a term.

Full provider profile →

Together AI

Request a plan for increasing the cluster during the project: GPU increments, network design, lead time and whether the expansion changes the reservation term or rate.

Full provider profile →

Vast.ai

Ask whether the exact host can supply your next allocation. If not, establish how to move data, reproduce the environment and arrange support on a replacement host.

Full provider profile →

Verda (formerly DataCrunch)

Request a growth plan within the selected service and Nordic region. Clarify which changes require a new cluster, data movement, reservation or revised support scope.

Full provider profile →

COMPARING GPU GENERATIONS

Considering B200?

Compare the B200 allocation and total workload cost against your H200 proposal. Our dedicated B200 guide covers published rates, minimum configurations and a representative acceptance test.

H100 pricing →H200 pricing →B200 pricing →B300 pricing →All providers →

AWS, AZURE, AND GOOGLE CLOUD

The big three clouds.
Eight GPUs at a time.

The largest clouds sell H200 and B200 as eight-GPU instances. These are published Linux rates in each cloud’s main US region. Per GPU-hour figures are our calculation: the instance rate divided by eight.

USD · Linux · 8-GPU instancesChecked

Scroll the table horizontally to see configurations and source links.

Published prices for eight-GPU H200 and B200 instances on AWS, Google Cloud, and Microsoft Azure, checked September 22, 2026.
Cloud and GPUInstance and regionPer instance-hourPer GPU-hourOfficial source
AWS · H200p5en.48xlarge: 8 × H200, 192 vCPUs, 2,048 GiB memory. US East (N. Virginia). On-demand.$63.30$7.91On-Demand pricing
Azure · H200ND96isr H200 v5: 8 × H200. East US 2. Pay as you go.$84.80$10.60Linux VM pricing
Google Cloud · H200a3-ultragpu-8g: 8 × H200, 224 vCPUs, 2,952 GB memory, 12,000 GiB SSD. Iowa (us-central1). On-demand.$84.81$10.60Accelerator-optimized pricing
AWS · B200p6-b200.48xlarge: 8 × B200, 192 vCPUs, 2,048 GiB memory. US East (N. Virginia). On-demand.$113.93$14.24On-Demand pricing
Google Cloud · B200a4-highgpu-8g: 8 × B200, 224 vCPUs, 3,968 GB memory, 12,000 GiB SSD. Iowa (us-central1). No on-demand rate listed; Flex-start shown.$64.44$8.06Accelerator-optimized pricing

The AWS figures were read from AWS’s published on-demand price data, and the Azure figure from Microsoft’s Retail Prices API. Azure did not list the H200 instance in East US, and its price list showed GB200-based instances rather than an HGX B200 instance. Google’s Flex-start runs your job when capacity frees up, so it is not the same as on-demand.

Google Cloud discounts

For the H200 instance in Iowa, a 3-year committed use discount lists at $37.21 per hour, about $4.65 per GPU-hour. Flex-start lists at $42.40, about $5.30. For B200, a 3-year commitment lists at $56.71, about $7.09 per GPU-hour, and Spot at $39.63, about $4.95.

Azure reservations

For ND96isr H200 v5 in East US 2, a 1-year reservation lists at $407,526 for the term, about $5.82 per GPU-hour. A 3-year reservation lists at $1,109,592, about $5.28 per GPU-hour. Both are our calculations over the full term.

How to read the gap

The USD on-demand specialist examples in this comparison are generally below these hyperscaler snapshots. Hyperscaler discounts close much of the gap, but only with a multi-year commitment, a queued start, or interruptible capacity. See our AWS vs Azure vs Google Cloud comparison for H100.

BEFORE YOU COMMIT

Ask for evidence.
Compare the same project.

Send each shortlisted provider the same requirements. Resolve the items that could stop the project before negotiating a lower hourly rate.

01

A launch and expansion commitment

Request the physical region, GPU variant, initial count and ready date, plus a path to your growth range. Confirm notice periods and reservation terms for added capacity. Ask what happens if the required allocation is unavailable.

02

A complete deployment design

Request the machine or cluster specification, links between GPUs and servers, storage and connectivity to your data. Confirm the exact service and facility covered by any security or residency requirements.

03

A bill for the full project window

Include setup, testing, productive runtime and any paid idle time. Itemize compute, persistent storage, transfer, support, software and taxes. State currency, billing increments, minimum term and renewal rules.

04

A clear owner for every operating task

Assign environment setup, model deployment, monitoring, backups and recovery. Ask what support covers, when it is available and what response commitment applies. Infrastructure access alone does not define application support.

05

A representative acceptance test

Agree on the model or application, data, software versions, output quality, runtime or latency target and maximum cost. Include a restart or recovery exercise if interruption would put delivery at risk.

06

An exit and change plan

Document how to resize, extend or cancel, export data and model files, and preserve checkpoints. Establish which charges continue after stopping compute and when stored data is deleted.

MAKE THE NEXT CONVERSATION USEFUL

Unsure how much capacity you will need?

Bring the workload, launch window and what could change. We can help turn that uncertainty into questions about initial capacity, growth and deployment support. Use the same brief with every provider.

Download the provider quote brief
Plan my capacity

H200 PRICING QUESTIONS

H200 cost,
answered.

Answers use the published rates on this page.
Get a quote for your project

How much does an H200 cost per hour?

On demand, 11 GPU clouds we track list H200 from $3.99 to $6.31 per GPU-hour. Most list $3.99 to $4.59. Rates were checked September 23 to September 28, 2026. Nebius lists $5.40 from October 1.

Can I rent a single H200?

Some providers sell one GPU at a time. Others sell whole servers. CoreWeave lists an 8-GPU instance at $50.44 per hour, about $6.31 per GPU. Confirm the smallest allocation before you compare hourly rates.

What is the difference between H200 and H100?

H200 has 141 GB of GPU memory and 4.8 TB/s of memory bandwidth, according to NVIDIA. H100 SXM has 80 GB and 3.35 TB/s. More memory helps large models and long inputs fit on fewer GPUs. On demand H100 lists from $3.49 to $6.16 per GPU-hour. See H100 cloud pricing.

Should I rent H200 or B200?

B200 is the newer Blackwell GPU. On demand B200 lists from $6.69 to $8.60 per GPU-hour, against $3.99 to $6.31 for H200. Test your own workload on both. A faster GPU can cost less per finished job even at a higher hourly rate. See B200 cloud pricing.

Can I pay less than the on demand rate?

Often, yes. Providers offer lower rates for reserved terms, and some sell interruptible spot capacity. Ask whether a commitment reserves the GPUs you need or only lowers the rate. Ask what happens if you need more GPUs mid-term.

HOW TO READ THE EVIDENCE

Published facts.
Project-specific questions.

Sources and check dates travel with each record. This directory covers our researched examples, not every provider or GPU they sell.

Published

A linked provider source supports the stated offer or profile fact. It is a dated snapshot. It does not confirm stock, your start date or your access to capacity.

Provider-confirmed

This label is reserved for a dated, direct confirmation covering a specific allocation and scope. No offer in this edition has that status.

Needs confirmation

Current capacity, growth rights and the final commercial and support terms must be resolved for your project. An unverified item is not a negative provider rating.

Fit and buying questions are editorial assessments, not firsthand benchmarks or financial risk ratings. Customer examples are provider-published relationships, not endorsements or proof of a particular GPU deployment unless stated. Our advisory services are free to you; we are compensated by the provider you choose through us. Listing a provider does not establish a reseller relationship.

WHY WORK WITH US

Powered by
Bridgepointe Technologies.

GPU Cloud Advisors is powered by Bridgepointe Technologies, a technology advisory firm since 2002. You get one advisor, the research on this site, and the provider relationships behind it.

Bridgepointe clients include

  • NVIDIA
  • Salesforce
  • Adobe
  • Uber
  • Netflix
  • Airbnb
  • Roblox
  • LinkedIn
  • CrowdStrike
  • Sony
  • PayPal
  • Mastercard

75+ Fortune 500 clients

Including more than 20 of the Fortune 100, with 97% client satisfaction.

1,200+ data center and cloud projects

Requirements assessed, capacity sourced and providers selected.

Terms, not just the rate

We negotiate ramp schedules, early billing and expansion terms, so you aren’t paying for capacity before your workload can use it.

One requirement, every fit

Bridgepointe partners with 380+ technology providers. We take one requirement to the providers that fit it and bring back a side-by-side comparison you can actually decide from.

Case study: Bridgepointe helped Hive, an AI company, choose data center and connectivity providers across 15 projects, saving its team at least 180 hours. Ask us for the case study

LET’S PLAN YOUR GPU PROJECT

Tell us what
you need.

Two short steps. Tell us what you’re building and an advisor will help you evaluate providers, confirm capacity requirements, and plan the support you need to get into production.

Not sure which GPU or how many you need? Start with your workload and what could change.

Our advisory services are free to you. We’re compensated by whichever provider you choose through us.

No obligationSizing help welcome
What we’ll work through with you
  • A provider shortlist suited to your workload and location.
  • Initial capacity, expansion needs and delivery timing to confirm.
  • Total cost, commitment terms and deployment support to compare.
Step 1 of 2What you need

Include GPU type, quantity and likely growth if known. A rough outline is enough.

Step 2 of 2Who you are

We’ll use these details to respond to your project request. Please leave out sensitive data and credentials. Privacy notice.

Prefer to talk it through? Book a free 30-minute consultation or call 844-506-2299.