NVIDIA B300
- Hyperstack ↗$7.40/hr
B300 on-demand; 288 GB listed.
- Runpod ↗$7.89/hr
B300 Pod; 288 GB listed. Excludes serverless pricing.
- Nebius ↗$9.50/hr
HGX B300; rate effective October 1, 2026.
Racklion / GPU price watchSame ambition.
Same ambition.
Less GPU spend.
Compare published GPU-hour rates. Reserve the capacity you need. Put more of your budget into the work that matters.
Public on-demand rates / USD per GPU-hourChecked Oct 5, 2026
| NVIDIA GPU | Average on-demand | At 30% lower | Listed range | Explore |
|---|---|---|---|---|
| B300Blackwell Ultra | $8.263 providers | $5.78Illustrative / GPU-hr | $7.40–$9.50 | ↗ |
| B200Blackwell | $7.103 providers | $4.97Illustrative / GPU-hr | $6.00–$8.50 | ↗ |
| H200Hopper · 141 GB | $4.663 providers | $3.26Illustrative / GPU-hr | $3.99–$5.40 | ↗ |
| H100SXM · 80 GB | $3.733 providers | $2.61Illustrative / GPU-hr | $3.20–$4.50 | ↗ |
| A100SXM · 80 GB | $1.993 providers | $1.40Illustrative / GPU-hr | $1.59–$2.79 | ↗ |
| L40SAda · 48 GB | $1.322 providers | $0.92Illustrative / GPU-hr | $1.09–$1.55 | ↗ |
Averages cover the selected providers below, not the entire market. The 30% column shows a savings scenario, not a bookable offer.
Put a number on it
What could you save?
Start with the GPU, the fleet size, and how long you need it.
Need a larger cluster, a specific region, or a custom term? Include it in your request.
Your capacity, at a better rate
8 × B30012-month reservation
- Average on-demand / month
- $48,258
- At 30% lower / month
- $33,781
Potential savings over 12 months$173,728
Find my reserved rate
Illustrates a 30% reduction with continuous use at 730 hours/month. Reserved capacity is billed for the committed term. Actual rates depend on availability, configuration, and term; excludes additional storage, network fees, and taxes.
Where these prices come from
Manually checked Oct 5, 2026. Each average is the arithmetic mean of one published on-demand rate per selected provider, normalized per physical GPU-hour. This is a dated reference snapshot, not a live availability feed. Spot, reserved, serverless, fractional GPUs, and contact-sales listings are excluded.
H100 and A100 rows use SXM models with 80 GB memory. B200 providers list 180–192 GB configurations. Included CPU, RAM, networking, region, node size, and minimum GPU count vary; these are price references, not identical offers. Nebius L40S uses its published starting price. Unused reserved hours still cost money.
NVIDIA B200
- Hyperstack ↗$6.00/hr
B200 on-demand; 192 GB listed.
- Runpod ↗$6.79/hr
B200 Pod; 180 GB listed. Excludes serverless pricing.
- Nebius ↗$8.50/hr
HGX B200; rate effective October 1, 2026.
NVIDIA H200
- Hyperstack ↗$3.99/hr
H200 SXM on-demand.
- Runpod ↗$4.59/hr
H200 Pod; not the separate cluster or serverless rate.
- Nebius ↗$5.40/hr
HGX H200; rate effective October 1, 2026.
NVIDIA H100
- Hyperstack ↗$3.20/hr
H100 SXM; excludes PCIe and NVLink listings.
- Runpod ↗$3.49/hr
H100 SXM Pod; excludes PCIe, NVL, and serverless listings.
- Nebius ↗$4.50/hr
HGX H100; rate effective October 1, 2026.
NVIDIA A100
- Hyperstack ↗$1.60/hr
A100 SXM 80 GB; excludes PCIe and 40 GB models.
- Runpod ↗$1.59/hr
A100 SXM 80 GB Pod; excludes serverless pricing.
- Lambda ↗$2.79/hr
A100 SXM 80 GB; per-GPU rate on an 8-GPU instance ($22.32/node-hour).