GPU MODELS / B200

B200 cloud GPU pricing

46 listings across 31 providers. Cheapest published rate: $3.500/GPU/hr at HydraHost. Sorted by per-GPU hourly price. Every row links to the provider's own published price.

WHAT IS THE B200

The B200 is Blackwell's datacenter flagship: roughly 2.3x the H100's dense compute, a new FP4 precision that doubles effective inference throughput, 192 GB at 8 TB/s, and NVLink 5 at 1.8 TB/s. It is the chip frontier training is migrating to. Supply is still ramping, so panel availability and pricing move week to week.

MAKER
NVIDIA
ARCHITECTURE
Blackwell
LAUNCHED
2025
MEMORY
192 GB HBM3e
MEMORY BANDWIDTH
8 TB/s
TENSOR COMPUTE
2.25 PFLOPS FP16/BF16 tensor dense; 4.5 PFLOPS FP8; 9 PFLOPS FP4 dense
POWER DRAW
1,000 W
INTERCONNECT
NVLink 5.0 1.8 TB/s

WHAT IT IS GOOD AT

Frontier pre-training
The current standard for new large training runs; NVL72 racks are the reference shape.
FP4 inference
FP4 with acceptable accuracy loss is Blackwell's party trick - it halves serving cost per token where it works.
Large MoE training
192 GB per GPU and 1.8 TB/s interconnect are sized for trillion-parameter-class mixtures.

WHO SHOULD NOT RENT THE B200

FP16/BF16 workloads that cannot use FP8/FP4 - much of the premium is in the new precisions.
Anyone who needs capacity today at scale - H100/H200 availability is still far broader.
ProviderConfigVRAMLocation$/hr$/GPU/hrSource
SpotGPUsSpotGPUsB200 SXM × 1180 GBNot published$2.490$2.490STALEreceipt
HydraHostHydraHostB200 × 1180 GBNot published$3.500$3.500receipt
VultrVultrB200 × 8192 GBNot published$28.000$3.500STALEreceipt
VultrVultrB200 (bare metal) × 8192 GBNot published$28.000$3.500STALEreceipt
packet.aipacket.aiB200 × 1192 GBUS East · Virginia$3.750$3.750receipt
BeamBeamB200 (from) × 1180 GBNot published$4.110$4.110receipt
LiumLiumB200 × 1192 GBNot published$4.750$4.750receipt
Deep InfraDeep InfraB200 × 1180 GBNot published$4.890$4.890receipt
GMI CloudGMI CloudB200 (from) × 1192 GBNot published$5.000$5.000receipt
Core42Core42B200 (from) × 1192 GBUAE$5.000$5.000receipt
UpCloudUpCloudB200 × 1180 GBNot published$5.001$5.001STALEreceipt
Shakti CloudShakti CloudB200 (bare metal) × 81536 GBIndia (Navi Mumbai)$40.319$5.040receipt
Beyond.plBeyond.plB200 × 8180 GBPoland$40.800$5.100STALEreceipt
Yotta LabsYotta LabsB200 × 1180 GBNot published$5.370$5.370receipt
Massed ComputeMassed ComputeB200 SXM6 × 8192 GBNot published$43.460$5.433receipt
KoyebKoyebB200 × 1180 GBNot published$5.500$5.500receipt
RunPodRunPodB200 × 1180 GBGlobal$5.980$5.980receipt
HyperbolicHyperbolicB200 (from) × 1192 GBNot published$5.990$5.990receipt
HyperstackHyperstackB200 × 1192 GBUS / EU / CA$6.000$6.000receipt
CerebriumCerebriumB200 × 1180 GBNot published$6.012$6.012receipt
DaytonaDaytonaB200 × 1180 GBNot published$6.250$6.250receipt
Lambda LabsLambda LabsB200 SXM6 × 1180 GBUS$7.990$7.990receipt
gpu.aigpu.aiB200 × 1192 GBNot published$8.010$8.010receipt
E2E NetworksE2E NetworksB200 × 1192 GBIndia$8.088$8.088receipt
NebiusNebiusHGX B200 × 1192 GBEU / US$8.500$8.500receipt
CoreWeaveCoreWeaveB200 × 8180 GBNorth America / Europe$68.800$8.600receipt
Hostnot GPUHostnot GPUB200 × 1192 GBUS East$8.820$8.820receipt
Verda (formerly DataCrunch)Verda (formerly DataCrunch)B200 SXM6 × 8180 GBEurope (Finland / Iceland)$73.830$9.229receipt
Verda (formerly DataCrunch)Verda (formerly DataCrunch)B200 SXM6 × 1180 GBEurope (Finland / Iceland)$9.229$9.229receipt
Verda (formerly DataCrunch)Verda (formerly DataCrunch)B200 SXM6 × 2180 GBEurope (Finland / Iceland)$18.460$9.230receipt
Verda (formerly DataCrunch)Verda (formerly DataCrunch)B200 SXM6 × 4180 GBEurope (Finland / Iceland)$36.920$9.230receipt
Verda (formerly DataCrunch)Verda (formerly DataCrunch)B200 CC SXM6 × 4180 GBEurope (Finland / Iceland)$37.650$9.412receipt
Verda (formerly DataCrunch)Verda (formerly DataCrunch)B200 CC SXM6 × 8180 GBEurope (Finland / Iceland)$75.310$9.414receipt
Verda (formerly DataCrunch)Verda (formerly DataCrunch)B200 CC SXM6 × 2180 GBEurope (Finland / Iceland)$18.830$9.415receipt
GPUBrazilGPUBrazilB200 × 1179 GBNorth America$9.527$9.527receipt
GPUBrazilGPUBrazilB200 × 1179 GBAsia-Pacific$9.527$9.527receipt
BasetenBasetenB200 × 1180 GBNot published$9.980$9.980receipt
SpheronSpheronB200 × 1192 GBNot published$10.170$10.170receipt
Hostnot GPUHostnot GPUB200 × 1192 GBEU East$10.550$10.550receipt
Hostnot GPUHostnot GPUB200 × 1192 GBUS West$10.550$10.550receipt
GPUBrazilGPUBrazilB200 × 1180 GBGlobal$10.587$10.587receipt
Oracle OCIOracle OCIB200 × 1180 GBNot published (global list)$14.000$14.000receipt
Hostnot GPUHostnot GPUB200 × 4192 GBUS East$56.120$14.030receipt
Hostnot GPUHostnot GPUB200 × 8192 GBUS East$118.820$14.852receipt
Hostnot GPUHostnot GPUB200 × 4192 GBUS West$59.420$14.855receipt
Hostnot GPUHostnot GPUB200 × 2192 GBUS East$31.370$15.685receipt

THE B200 MARKET, BY THE NUMBERS

PROVIDERS
31
CONFIGURATIONS
46
MEDIAN $/GPU/HR
$8.60
SPREAD
4.5x

31 providers publish 46 B200 configurations on the panel today. Published per-GPU hourly rates run from $3.500 (HydraHost) to $15.69 (Hostnot GPU), a 4x spread between the cheapest and the most expensive published rate for the same chip. The median listing sits at $8.60/GPU/hr. 11 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 B200 BY GEOGRAPHY
united statespacket.ai$3.750/hruaeCore42$5.000/hrindiaShakti Cloud$5.040/hrcanadaHyperstack$6.000/hrfinlandVerda (formerly DataCrunch)$9.229/hricelandVerda (formerly DataCrunch)$9.229/hr
B200 PRICE TREND, DAILY PANEL SNAPSHOTS

Across every B200 row on the panel, the daily floor moved from $4.000 to $3.500/GPU/hr over 53 days of tracking (-12.5%). The median continuously-tracked listing moved from $6.23 to $8.50 (+36.4%). Snapshots run daily; the full history is public in the repo.

B200 PROVIDER NOTES
HydraHost · 1 config
Sets the panel floor for this model.
from $3.500/hr
packet.ai · 1 config · 1 region
Within 15% of the panel floor.
from $3.750/hr
Beam · 1 config
Published catalog rates.
from $4.110/hr
Lium · 1 config
Published catalog rates.
from $4.750/hr
Deep Infra · 1 config
Published catalog rates.
from $4.890/hr
GMI Cloud · 1 config
Published catalog rates.
from $5.000/hr
Core42 · 1 config · 1 region
Published catalog rates.
from $5.000/hr
Shakti Cloud · 1 config · 1 region
Published catalog rates.
from $5.040/hr
BUYING B200: WHAT THE PANEL SAYS

At the current floor, one B200 running around the clock costs about $2,555 a month (730 hours of on-demand arithmetic). Node shape matters: the single-GPU floor is $3.500/GPU/hr against $5.040/GPU/hr on an 8-GPU node (+44.0%). Per-GPU floors by config size: 1x $3.50 · 2x $9.23 · 4x $9.23 · 8x $5.04. Compare the alternatives below before committing - the same budget often buys more than one chip class.

B200 VS THE ALTERNATIVES

B200 vs H200 SXM
B200 more than doubles compute and adds FP4 at 1.4x the memory. If your stack cannot use FP8/FP4 well, the realized gain shrinks - and H200s are far easier to rent today.
B200 vs B300
Blackwell Ultra adds memory (288 GB) and roughly 50% more FP4 throughput, aimed at inference reasoning models. For training, the B200 is the mainstream part.

Compare live rates: B200 vs H100 SXM · B200 vs H200 SXM · B200 vs B300 · B200 vs GB300 · B200 vs GB200 · B200 vs MI350X · B200 vs MI300X

B200 PRICING FAQ

What is the cheapest B200 cloud GPU right now?
The cheapest published B200 rate on the panel is $3.500 per GPU-hour at HydraHost (B200 x 1, Not published). Every price links to the provider's own published rate.
How much does a B200 cost per hour?
Across 31 providers tracked today, published per-GPU hourly rates for B200 run from $3.50 to $15.69, with a median of $8.60. Committed-use and reserved rates are excluded; everything shown is on-demand.
How many providers rent B200 GPUs?
31 providers publish 46 distinct B200 configurations on Compute Cafe. 11 of those rows were reported in stock by the provider's live endpoint at last observation.
Where is B200 cheapest?
By published per-GPU hourly rate, the cheapest geography for B200 today is united states at $3.75/hr (packet.ai). The spread across 6 tracked geographies runs to $9.23/hr in iceland.
Are B200 prices going up or down?
Over the 53 days Compute Cafe has tracked the panel, the B200 floor is falling: $4.000 to $3.500/GPU/hr (-12.5%). The median continuously-tracked listing moved from $6.23 to $8.50 (+36.4%). Daily snapshots are public in the repo, so the trend is auditable.
What does a B200 cost per month?
At the current floor of $3.500/GPU/hr, a B200 running 24/7 costs about $2,555 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 B200 node cheaper than a single card?
On today's panel: the single-GPU floor is $3.500/GPU/hr and the 8-GPU node floor is $5.040/GPU/hr - more expensive per GPU on a full node (+44.0%). Check the table for the exact configs behind each floor.
Is B200 worth the premium over H100?
For FP8/FP4-native training and high-throughput serving, yes - per-token and per-step economics favor Blackwell. For FP16/BF16 workloads that fit in 80 GB, an H100 at half the hourly rate often wins on cost per result. Run the panel's live spread before committing.

See also alternatives to the B200 · cheapest GPU-hour per model · inference chips · training chips · by location