NEWER GPUS. PUBLISHED RATES.
Compare published NVIDIA H200 and B200 cloud prices from specialist GPU clouds and from AWS, Azure, and Google Cloud. See the rate, the billing unit, and what changes on October 1.
01 / H200 PRICING
The H200 keeps the H100’s Hopper design and adds more memory: 141 GB per GPU against 80 GB. These are published on-demand rates from specialist GPU clouds, lowest first.
Scroll the table horizontally to see billing details and source links.
| Provider | Example configuration | Published price | How to read the rate | Official source |
|---|---|---|---|---|
| Crusoe | H200 HGX · 141 GBOn-demand GPU instance | $4.29per GPU / hour | Multiply by the instance’s GPU count. Crusoe lists spot and reserved H200 pricing through sales. | View pricing |
| Nebius | HGX H200On-demand GPU instance | $4.50per GPU / hour | Published with 16 vCPUs and 200 GB RAM per GPU. Nebius lists $5.40 effective October 1, 2026. Preemptible capacity lists from $0.79, with interruption terms. | View pricing |
| Runpod | H200 · 141 GBGPU Pod listing | $4.59per GPU / hour | The listed configuration has 24 vCPUs and 276 GB RAM. Storage, Serverless endpoints, and Clusters are priced separately. | View pricing |
| Edgevana | H200 SXM · 141 GBGPU marketplace listing | $5.47per GPU / hour | The listing shows $43.80/hour total for the server, with 128 CPU cores and 1,600 GB RAM. Marketplace listings change with inventory. | View pricing |
| CoreWeave | 8 × HGX H200 · 141 GB eachOn-demand instance | $50.44per 8-GPU instance / hour | About $6.31 per GPU-hour, our calculation from the instance rate. Includes 128 vCPUs, 2,048 GB RAM, and 61.44 TB local storage. | View pricing |
Lambda’s instance pricing page listed B200, H100, and other GPUs but no on-demand H200 instance on our check date. Configurations and billing units differ between providers, so the hourly rate alone is not a like-for-like comparison.
02 / B200 PRICING
The B200 is NVIDIA’s Blackwell-generation data center GPU. Most providers list it with 180 GB of memory per GPU. These are published on-demand rates, lowest first.
Scroll the table horizontally to see billing details and source links.
| Provider | Example configuration | Published price | How to read the rate | Official source |
|---|---|---|---|---|
| Lambda | 8 × B200 SXM6 · 180 GB eachOn-demand instance | $6.69per GPU / hour | The 8-GPU configuration includes 208 vCPUs, 2,900 GiB RAM, and 22 TiB SSD. Smaller instances list higher, up to $6.99 for one GPU. Taxes are additional. | View pricing |
| Runpod | B200 · 180 GBGPU Pod listing | $6.79per GPU / hour | The listed configuration has 28 vCPUs and 283 GB RAM. Storage, Serverless endpoints, and Clusters are priced separately. | View pricing |
| Nebius | HGX B200On-demand GPU instance | $7.15per GPU / hour | Published with 20 vCPUs and 224 GB RAM per GPU. Nebius lists $8.50 effective October 1, 2026. Preemptible capacity lists from $0.99. | View pricing |
| CoreWeave | 8 × HGX B200 · 180 GB eachOn-demand instance | $68.80per 8-GPU instance / hour | About $8.60 per GPU-hour, our calculation from the instance rate. Includes 128 vCPUs, 2,048 GB RAM, and 61.44 TB local storage. | View pricing |
| Edgevana | B200 SXMGPU marketplace listing | $8.68per GPU / hour | The listing shows $69.43/hour total for the server, with 160 CPU cores and 1,792 GB RAM. Marketplace listings change with inventory. | View pricing |
Crusoe listed B200 on-demand and spot pricing as “Contact sales” on our check date. Newer Blackwell systems, including B300 and GB200 NVL72, are often quoted by sales rather than listed. Ask for them by name if your workload needs them.
04 / H100, H200, OR B200
Newer is not automatically better value. The right choice depends on how much memory your model needs and what a finished job costs, not the hourly rate.
H100 rates are from our GPU cloud pricing comparison, checked .
A B200 that finishes a training run faster can cost less in total than a cheaper GPU that runs longer. It can also cost more if your workload cannot use its extra capability. Run a representative job on each candidate, then compare the total bill for the finished work.
LET’S FIND YOUR WAY FORWARD
Bring an H200 or B200 quote, or the workload you are sizing. An advisor can help you compare providers, terms, and availability, and agree on the next step.
Our advisory services are free to you. We’re compensated by whichever provider you choose through us.
A few details help us make the most of your time:
Book securely on Calendly. Times appear in your time zone.
H200 AND B200 PRICING QUESTIONS
Bring your workload or an existing quote.
Meet with an advisor
In our September 22, 2026 check, specialist GPU clouds published on-demand H200 rates of $4.29 to about $6.31 per GPU-hour. AWS listed about $7.91, and Azure and Google Cloud about $10.60, for eight-GPU instances. Rates vary by region and change over time.
Specialist clouds published on-demand B200 rates of $6.69 to $8.68 per GPU-hour in our check. AWS listed about $14.24 per GPU-hour for its eight-GPU instance. Google Cloud listed no on-demand B200 rate, only Flex-start, Spot, and commitment rates.
At least one provider has announced increases. Nebius lists H200 rising from $4.50 to $5.40 and B200 from $7.15 to $8.50 per GPU-hour, effective October 1, 2026. Ask any provider whether a quote is protected from future list-price changes.
Consider H200 when your model or context length needs more than 80 GB per GPU, or when it lets you use fewer GPUs. The same-provider premium in our check ranged from about 2 percent at CoreWeave to about 32 percent at Runpod. Benchmark your own workload before deciding.
Our advisory services cost you nothing. We’re compensated by whichever provider you choose through us. You pay the provider for your infrastructure and services, with no obligation to choose a provider.