GPU MODELS / H200 NVL

H200 NVL cloud GPU pricing

14 listings across 9 providers. Cheapest published rate: $0.500/GPU/hr at RunPod. Sorted by per-GPU hourly price. Every row links to the provider's own published price.

WHAT IS THE H200 NVL

The H200 NVL pairs two 141 GB HBM3e cards over an NVLink bridge - 282 GB per pair, the most inference-friendly memory footprint in the Hopper line. It serves 70B models at BF16 on a single card and large MoE models on the pair, in standard PCIe servers.

MAKER
NVIDIA
ARCHITECTURE
Hopper
LAUNCHED
2024
MEMORY
141 GB HBM3e per GPU
MEMORY BANDWIDTH
4.8 TB/s
TENSOR COMPUTE
835 TFLOPS FP16/BF16 tensor dense; FP8 1,671 dense
POWER DRAW
up to 600 W
INTERCONNECT
NVLink bridge 600 GB/s

WHAT IT IS GOOD AT

70B+ serving
141 GB per card covers BF16 70B with context headroom.
MoE and long-context inference
The pair's 282 GB handles footprints 80 GB cards cannot.

WHO SHOULD NOT RENT THE H200 NVL

Training clusters - inference-first design.
Small models - the 141 GB goes unused.
ProviderConfigVRAMLocation$/hr$/GPU/hrSource
RunPodRunPodH200 NVL × 1143 GBGlobal$0.500$0.500receipt
SpotGPUsSpotGPUsH200 NVL × 1141 GBNot published$0.990$0.990receipt
gpu.aigpu.aiH200 NVL × 1141 GBNot published$3.290$3.290receipt
Massed ComputeMassed ComputeH200 NVL × 1141 GBNot published$3.620$3.620receipt
Hostnot GPUHostnot GPUH200 NVL × 1141 GBUS East$3.620$3.620receipt
Hostnot GPUHostnot GPUH200 NVL × 2141 GBUS East$7.240$3.620receipt
Massed ComputeMassed ComputeH200 NVL × 8141 GBNot published$28.960$3.620receipt
Massed ComputeMassed ComputeH200 NVL NVLink × 8141 GBNot published$30.560$3.820receipt
TeamCloudTeamCloudH200 NVL (from) × 1141 GBMalaysia$4.684$4.684receipt
GPUBrazilGPUBrazilH200 NVL × 1140 GBEurope$4.759$4.759receipt
GPUBrazilGPUBrazilH200 NVL × 1140 GBNorth America$5.005$5.005receipt
NovaGPU (IP ServerOne)NovaGPU (IP ServerOne)H200 NVL × 1141 GBMalaysia$5.022$5.022receipt
NovaGPU (IP ServerOne)NovaGPU (IP ServerOne)H200 NVL × 2282 GBMalaysia$10.047$5.024receipt
SelectelSelectelH200 NVL × 1141 GBRussia$7.877$7.877receipt

THE H200 NVL MARKET, BY THE NUMBERS

PROVIDERS
9
CONFIGURATIONS
14
MEDIAN $/GPU/HR
$3.82
SPREAD
15.8x

9 providers publish 14 H200 NVL configurations on the panel today. Published per-GPU hourly rates run from $0.500 (RunPod) to $7.88 (Selectel), a 16x spread between the cheapest and the most expensive published rate for the same chip. The median listing sits at $3.82/GPU/hr. 10 rows were reported in stock by the provider at last observation; the rest are published catalog rates. All rates are on-demand - committed-use and reserved pricing is excluded.

CHEAPEST H200 NVL BY GEOGRAPHY
united statesHostnot GPU$3.620/hrmalaysiaTeamCloud$4.684/hr
H200 NVL PRICE TREND, DAILY PANEL SNAPSHOTS

Across every H200 NVL row on the panel, the daily floor held flat at $0.500/GPU/hr over 11 days of tracking. The median continuously-tracked listing held flat at $3.62. Snapshots run daily; the full history is public in the repo.

H200 NVL PROVIDER NOTES
RunPod · 1 config · 1 region · 1 in stock
Sets the panel floor for this model.
from $0.500/hr
SpotGPUs · 1 config · 1 in stock
1 of 1 configs reported in stock at last check.
from $0.990/hr
gpu.ai · 1 config · 1 in stock
1 of 1 configs reported in stock at last check.
from $3.290/hr
Massed Compute · 3 configs
Published catalog rates.
from $3.620/hr
Hostnot GPU · 2 configs · 1 region · 2 in stock
2 of 2 configs reported in stock at last check.
from $3.620/hr
TeamCloud · 1 config · 1 region · 1 in stock
1 of 1 configs reported in stock at last check.
from $4.684/hr
GPUBrazil · 2 configs · 2 regions · 2 in stock
2 of 2 configs reported in stock at last check.
from $4.759/hr
NovaGPU (IP ServerOne) · 2 configs · 1 region · 2 in stock
2 of 2 configs reported in stock at last check.
from $5.022/hr
BUYING H200 NVL: WHAT THE PANEL SAYS

At the current floor, one H200 NVL running around the clock costs about $365 a month (730 hours of on-demand arithmetic). Node shape matters: the single-GPU floor is $0.500/GPU/hr against $3.620/GPU/hr on an 8-GPU node (+624.0%). Per-GPU floors by config size: 1x $0.50 · 2x $3.62 · 8x $3.62. Compare the alternatives below before committing - the same budget often buys more than one chip class.

H200 NVL VS THE ALTERNATIVES

H200 NVL vs H100 NVL
Same shape, 50% more memory and 23% more bandwidth per card. For serving, the H200 NVL is the better part; the H100 NVL is often the better price.

Compare live rates: H200 NVL vs H100 NVL

H200 NVL PRICING FAQ

What is the cheapest H200 NVL cloud GPU right now?
The cheapest published H200 NVL rate on the panel is $0.500 per GPU-hour at RunPod (H200 NVL x 1, Global). Every price links to the provider's own published rate.
How much does a H200 NVL cost per hour?
Across 9 providers tracked today, published per-GPU hourly rates for H200 NVL run from $0.50 to $7.88, with a median of $3.82. Committed-use and reserved rates are excluded; everything shown is on-demand.
How many providers rent H200 NVL GPUs?
9 providers publish 14 distinct H200 NVL configurations on Compute Cafe. 10 of those rows were reported in stock by the provider's live endpoint at last observation.
Where is H200 NVL cheapest?
By published per-GPU hourly rate, the cheapest geography for H200 NVL today is united states at $3.62/hr (Hostnot GPU). The spread across 2 tracked geographies runs to $4.68/hr in malaysia.
Are H200 NVL prices going up or down?
Over the 11 days Compute Cafe has tracked the panel, the H200 NVL floor is holding flat: $0.500 to $0.500/GPU/hr (+0.0%). The median continuously-tracked listing moved from $3.62 to $3.62 (+0.0%). Daily snapshots are public in the repo, so the trend is auditable.
What does a H200 NVL cost per month?
At the current floor of $0.500/GPU/hr, a H200 NVL running 24/7 costs about $365 per month (730 hours of on-demand arithmetic on the cheapest published rate). Committed-use contracts can price lower; the panel tracks on-demand rates only.
Is a full 8-GPU H200 NVL node cheaper than a single card?
On today's panel: the single-GPU floor is $0.500/GPU/hr and the 8-GPU node floor is $3.620/GPU/hr - more expensive per GPU on a full node (+624.0%). Check the table for the exact configs behind each floor.
What fits on an H200 NVL pair?
A BF16 70B model with long-context KV cache sits on one card; the pair's 282 GB covers quantized 180B-class models or heavy multi-tenant serving.

See also alternatives to the H200 NVL · cheapest GPU-hour per model · inference chips · training chips · by location