Current H200 cloud pricing
Today’s sourced per-provider H200 on-demand rates, source-linked and dated.
← Back to Compute College
Compute College
H200 Price Per Hour Explained
H200 price per hour is the hourly cost of accessing one NVIDIA H200 GPU for AI workloads.
H200 price per hour definition
H200 price per hour is the hourly cost to rent or operate one NVIDIA H200 GPU for AI workloads. H200 rates may carry a premium over H100 capacity because of its larger, faster memory — roughly 141 GB of HBM3e at about 4.8 TB/s, versus 80 GB of HBM3 at about 3.35 TB/s on the H100 SXM — which can help large-model and memory-heavy serving workloads.
Memory trick: H100 is the yardstick; H200 asks whether more memory is worth the premium for this workload.
Live price band
H200 on-demand price band
H200 on-demand capacity ranges roughly $4.29–$4.39 per GPU-hour across 2 sourced providers, as of Jul 7, 2026. Each row below links to the provider's public price page and carries its own observation date.
Public on-demand list prices normalized to a per-GPU-hour rate — not negotiated quotes or reserved pricing. How we label and date evidence: methodology.
Sourced providers
Per-provider H200 on-demand rates
| Provider | Region | $/GPU-hour | Source | Observed |
|---|---|---|---|---|
| Crusoe | Multi-region | $4.29 | price page | Jul 7, 2026 |
| RunPod Secure | Multi-region | $4.39 | price page | Jul 7, 2026 |
Why it matters
Why H200 hourly pricing matters
H200 pricing helps readers distinguish paying for scarce capacity from paying for better workload fit. A higher rate is not automatically a worse economic choice: if memory allows a job to finish faster, avoid bottlenecks, or serve a larger model efficiently, effective cost can improve.
- Memory demand can create pricing pressure separate from raw computing throughput.
- Large context windows and memory-heavy inference may value H200 capacity differently from other work.
- A premium over H100 helps reveal where buyers believe the constraint lies.
Simple example
In an illustrative comparison, an H100 costs $7 per hour while an H200 costs $9 per hour. The H200 headline rate is about 28.6% higher. If the particular workload completes 35% faster on the H200 or avoids a memory bottleneck, the final cost per completed job may still be competitive or lower.
- Compare the same job on each accelerator rather than hourly rates alone.
- Measure runtime, output, and failure or memory limitations under comparable terms.
- Treat performance improvements as workload-specific until measured or sourced.
Example figures are illustrative calculations, not current quoted market prices.
Market signal
How to read the market signal
A widening H200 premium may indicate strong demand for memory-rich capacity or limited available H200 supply. A narrowing premium may point to broader deployment, provider discounting, weaker incremental demand, or buyers moving toward B200 systems.
- Compare premiums across similar contract types and regions.
- Track whether H200 availability improves while rates remain elevated.
- Read memory-linked pricing alongside supply news and new-generation capacity additions.
Market read: an H200 premium becomes informative when it persists across comparable offers and corresponds with demand for memory-heavy jobs, not when it appears in one isolated quote. Figures here are illustrative unless explicitly sourced and dated — see our methodology.
Common mistake
The common mistake is choosing the cheapest GPU-hour without measuring the workload. Hardware with more usable memory may reduce runtime, reduce the number of GPUs required, or make a workload feasible at all. Conversely, a workload that does not benefit from the memory premium may not justify the higher rate.
Practical takeaway
What you can do with this
Ask whether the job is compute-bound, memory-bound, latency-sensitive, or limited by availability. Compare quotes using completed workload cost or serving output rather than assuming an hourly premium should always be avoided.
- Buyers: test representative workloads before committing to premium capacity.
- Product managers: match serving model requirements to hardware memory needs.
- Analysts: use the H200-to-H100 premium as one signal of demand for memory-rich compute.
- Procurement teams: request comparable H100 and H200 configurations instead of comparing unmatched offerings.
- Operators: watch whether memory relief improves throughput enough to reduce total GPUs or elapsed runtime.
- Finance teams: model both the premium hourly rate and the shorter-runtime case before approving capacity or renewing a reservation commitment.
Decision check: pay a memory premium only when a representative workload or sourced evidence shows it improves cost, throughput, capacity access, or feasibility.
Compute College
Turn the lesson into a number
Use the GPU-Hour Cost Calculator, AI Training Cost Calculator, or Model Serving Cost Calculator.
Compute College track
GPU Pricing & Capacity
Step 2 of 8: H200 price per hour