on demand observed floor
Civo · $3.49/GPU hr
240 GPU-hours: $837.60
- Region sample
- Unspecified
- Observed
GPU cost intelligence
H200 cloud cost calculator
Compare H200 GPU prices from 6 providers. Observed range: $2.45–$4.50 per GPU-hour, with source evidence, availability signals, and cost estimates.
- Observed hourly range
- $2.45–$4.50 / GPU hr
- 730 GPU-hour range
- $1,788.50–$3,285.00
- Providers
- 6
- Official-source providers
- 6
- Fresh comparable offers
- 13
- Pricing observed
- Aug 13, 2026–Aug 13, 2026
- Regions observed
- 4
Decision snapshot by purchase model
Each card shows the lowest observed representative rate within its own purchase model. Spot and on-demand are intentionally not placed in one cheapest-to-expensive ranking because interruption risk, capacity behavior, and commercial terms are not interchangeable.
These are price floors, not recommendations: availability, region coverage, support, commitments, and workload fit can change the better purchasing choice.
spot observed floor
Nebius AI Cloud · $2.45/GPU hr
240 GPU-hours: $588.00
- Region sample
- eu-north1
- Observed
H200 price by purchase model
Purchase models are kept separate because interruption risk, commitment terms, and capacity behavior are not interchangeable. Each range uses only fresh representative offers from the same market type.
Spot vs on-demand low end
29.8% lower observed spot floor
The lowest representative spot rate is $2.45/GPU hr versus $3.49/GPU hr for on-demand. This compares observed price floors only; spot interruption risk, capacity, regions and workload suitability remain separate decisions.
Decision read: Spot has the lower observed floor in this snapshot. Treat that gap as a candidate saving only for workloads that can tolerate interruption and after checking capacity, regions, and workload fit.
Compute-only break-even: at these observed floors, spot could consume up to 42.4% more billed GPU-hours before its compute cost matches the on-demand floor. This threshold does not price engineering work, checkpointing, restart latency, SLA exposure, lost progress, or opportunity cost.
160 GPU-hours · observed floors
$166.40 lower spot cost
Spot floor: $392.00 · On-demand floor: $558.40.
Compute retry budget: spot could add up to 67.9 billed GPU-hours at this baseline before matching the on-demand floor's compute spend.
Price-floor scenario only; interruption risk, capacity, regions, commitments and workload suitability are not priced into this comparison.
240 GPU-hours · observed floors
$249.60 lower spot cost
Spot floor: $588.00 · On-demand floor: $837.60.
Compute retry budget: spot could add up to 101.9 billed GPU-hours at this baseline before matching the on-demand floor's compute spend.
Price-floor scenario only; interruption risk, capacity, regions, commitments and workload suitability are not priced into this comparison.
730 GPU-hours · observed floors
$759.20 lower spot cost
Spot floor: $1,788.50 · On-demand floor: $2,547.70.
Compute retry budget: spot could add up to 309.9 billed GPU-hours at this baseline before matching the on-demand floor's compute spend.
Price-floor scenario only; interruption risk, capacity, regions, commitments and workload suitability are not priced into this comparison.
Observed H200 price dispersion
Across the representative offers in this snapshot, the normalized rate runs from $2.45 to $4.50 per GPU-hour, a gap of $2.05. At 730 GPU-hours, that rate gap corresponds to $1,496.50.
160 GPU-hours
$328.00 price-range gap
Observed compute budget range: $392.00–$720.00 at this usage level.
If the observed floor rate were otherwise suitable for the same workload, that ceiling budget would mathematically cover 293.9 GPU-hours — 133.9 more than this scenario.
Difference between the lowest and highest representative H200 rates at this usage level.
240 GPU-hours
$492.00 price-range gap
Observed compute budget range: $588.00–$1,080.00 at this usage level.
If the observed floor rate were otherwise suitable for the same workload, that ceiling budget would mathematically cover 440.8 GPU-hours — 200.8 more than this scenario.
Difference between the lowest and highest representative H200 rates at this usage level.
730 GPU-hours
$1,496.50 price-range gap
Observed compute budget range: $1,788.50–$3,285.00 at this usage level.
If the observed floor rate were otherwise suitable for the same workload, that ceiling budget would mathematically cover 1340.8 GPU-hours — 610.8 more than this scenario.
Difference between the lowest and highest representative H200 rates at this usage level.
Read from the budget floor, the highest observed representative rate is 83.7% above the lowest; read from the ceiling, the lowest is 45.6% below the highest. This is price dispersion, not a guaranteed saving: purchase model, availability, region coverage, support, commitments, and workload fit can differ between offers.
Current H200 cloud pricing evidence
Representative comparable offers are normalized to a per-GPU-hour rate and ranked from lowest to highest.
| Provider | Variant | Market | Price | 730 GPU hr | Freshness | Observed | Evidence |
|---|---|---|---|---|---|---|---|
| Nebius AI Cloud | NVLink 141GB | spot | $2.45/GPU hr | $1,788.50 | recent | Source | |
| Civo | H200 | on demand | $3.49/GPU hr | $2,547.70 | recent | Source | |
| Hyperstack | H200 | on demand | $3.99/GPU hr | $2,912.70 | recent | Source | |
| DataCrunch (now Verda) | H200 | on demand | $4.00/GPU hr | $2,920.00 | recent | Source | |
| Crusoe Cloud | H200 | on demand | $4.29/GPU hr | $3,131.70 | recent | Source | |
| DigitalOcean | H200 | on demand | $4.47/GPU hr | $3,263.10 | recent | Source | |
| Nebius AI Cloud | NVLink 141GB | on demand | $4.50/GPU hr | $3,285.00 | recent | Source |
H200 regional coverage
These are explicitly named regions reported by fresh comparable offers. Counts and provider identities describe observed catalog coverage, not live capacity or guaranteed availability; unspecified placements are not assigned to a region.
Regional provider concentration
0 of 4 named regions have evidence from 2+ providers
0 observed regions include fresh comparable H200 catalog evidence from at least two providers, while 4 are represented by one provider only.
Cross-provider catalog coverage: 0% of explicitly named regions.
3+ provider catalog depth: 0 named regions (0%) include a third observed provider option. This is a deeper catalog-diversity signal only, not proof that those providers are operationally substitutable or simultaneously available.
Decision read: Every explicitly named region in this fresh comparable snapshot is represented by one provider only, so regional fallback planning has no cross-provider catalog evidence here.
Single-provider region watchlist: eu-north1 — Nebius AI Cloud, eu-north2 — Nebius AI Cloud, eu-west1 — Nebius AI Cloud, plus 1 more in the region table below. Each provider shown is only the sole provider represented by fresh comparable catalog rows for that named region in this snapshot; it is not a live availability or exclusivity claim.
This is a catalog-breadth signal for fallback planning, not proof of simultaneous capacity, failover readiness, quota, latency, workload fit, or contractual substitutability.
| Region | Fresh comparable offers | Providers represented | Observed providers |
|---|---|---|---|
| eu-north1 | 2 | 1 | Nebius AI Cloud |
| eu-north2 | 2 | 1 | Nebius AI Cloud |
| eu-west1 | 2 | 1 | Nebius AI Cloud |
| us-central1 | 2 | 1 | Nebius AI Cloud |
Deployability evidence coverage
2 of 7 representative offers (29%) have a non-unknown deployment status backed by a separate HTTPS evidence source and a valid observation time.
Deployability evidence gap
29% separately evidenced
5 representative offers currently lack qualifying deployment-eligibility evidence and remain explicitly unevidenced.
Decision read: 5 representative offers should still be treated as unknown for deployability comparisons; do not use a current price row as a substitute for capacity evidence.
Deployment eligibility is separate from price freshness. It does not prove live stock or allocatable capacity; offers without qualifying evidence remain explicitly unknown.
Typed availability evidence
Availability is tracked separately from pricing. The evidence kind distinguishes deployment eligibility from live stock or allocatable capacity; none of those claims is inferred from a listed price.
| Provider | Market | Regions | Status | Evidence kind | Observed | Evidence |
|---|---|---|---|---|---|---|
| Nebius AI Cloud | spot | eu-north1 | available | deployment eligibility | Availability source | |
| Civo | on demand | Unspecified | unknown | unknown | Not separately observed | No separate source |
| Hyperstack | on demand | Unspecified | unknown | unknown | Not separately observed | No separate source |
| DataCrunch (now Verda) | on demand | Unspecified | unknown | unknown | Not separately observed | No separate source |
| Crusoe Cloud | on demand | Unspecified | unknown | unknown | Not separately observed | No separate source |
| DigitalOcean | on demand | Unspecified | unknown | unknown | Not separately observed | No separate source |
| Nebius AI Cloud | on demand | eu-north1 | available | deployment eligibility | Availability source |
Calculate H200 GPU cost
Enter the GPU-hours you expect to consume. The estimate uses the same lowest and highest representative hourly rates shown above and excludes storage, bandwidth, support, taxes, commitments, and negotiated discounts.
H200 cost scenarios
These estimates multiply GPU-hours by the lowest and highest representative hourly rates above. They exclude storage, bandwidth, support, taxes, commitments, and negotiated discounts.
| GPU-hours | Published price range |
|---|---|
| 40 | $98.00–$180.00 |
| 160 | $392.00–$720.00 |
| 240 | $588.00–$1,080.00 |
| 730 | $1,788.50–$3,285.00 |
Compare H200 by provider
Open the exact provider × H200 decision sheets that currently pass the autonomous publication gates. Each page keeps provider-specific regions, configurations, evidence, and normalized rates separate from the broader market view above.
How this comparison is built
- Only comparable normalized records for the exact H200 model are eligible.
- Fresh records are limited to the existing live, recent, or verified freshness states; this snapshot is 100% fresh.
- Representative rows keep the lowest normalized rate for each provider and market type, so spot and on-demand offers are not blended together.
- Pricing evidence does not guarantee live capacity. Check the linked source and provider availability before purchasing.
Verification and policies
Review the broader GPU cloud pricing hub, the data sources, the editorial policy, and the corrections policy to inspect how SaaS Sentinel sources, reviews, and corrects published evidence.