📊 LIVE B200 rental prices

NVIDIA B200 SXM Rental Price

All B200 SXM rental rates, live: $/hr, ₹/hr, $/day, $/month — on-demand and spot. Not the hardware price. The actual cost to run the flagship Blackwell GPU.

Updated from live feed

B200 SXM Rental — live providers

B200 SXM · 192 GB HBM3e · Blackwell · 2250 FP8 TFLOPS
ProviderOn-demand $/hrSpot $/hr₹/hr$/day$/monthSpot saving
Loading live rental prices…

*Live gpuprice.in feed (pricing-data.json, source Parallel.ai), updated daily. Spot saving = % cheaper than that provider's on-demand. $/day = 24h, $/month = 730h, ₹/hr ≈ ₹83/USD.

B200 specs that matter for rental

192 GB
HBM3e VRAM
2250
TFLOPS FP8
Blackwell
Architecture
8.0 TB/s
Memory bandwidth

The B200 is the flagship Blackwell card: 192GB of HBM3e — the most VRAM of any single GPU on the market — and 2250 FP8 TFLOPS, roughly twice the throughput of the H100/H200. It is the only single GPU that fits 400B-class MoE models (DeepSeek R1/V3, Qwen-MoE, Grok-class) without multi-GPU sharding.

That power is scarce and premium: only 2 providers (Spheron, Nebius) list B200 on the live feed, and on-demand starts at $4.50/hr — roughly twice an H100. Spot from $2.12/hr (53% off) at Spheron is the value entry point if your workload is interruptible. Expect allocation lead times — high-end capacity is often quoted to reserve weeks ahead.

How much will your B200 workload cost?

Enter hours — get cost for on-demand or spot, and what you'd pay on a comparable setup.

Need $/1M-token cost? Use the Workload Cost Calculator → to see whether a B200 or a cheaper card is the per-token bargain for your model.

On-demand B200

Pick a provider + enter hours

Spot B200 (interruptible)

Pick a provider + enter hours

Example: 1,000 hours of 400B-class MoE inference on the B200 = ~$4,500 on-demand, or ~$2,120 on spot. Because only 2 providers list B200, compare against an H200 or multi-H100 setup via the table below — sometimes two H100s beat one B200 on price per token.

B200 vs alternatives: rent by what fits

Usually the choice is VRAM, not brand. Cheapest live on-demand rate per GPU.

GPUArchVRAMCheapest $/hrCheapest ₹/hr

B200 vs H200

The H200 (141GB, $2.80/hr) is ~37% cheaper and fits everything under ~150B. The B200 (192GB) is the only single-GPU answer for 150-400B models and is ~2× the FP8 throughput. If your model fits 141GB, the H200 wins on price; above that, only the B200 fits.

B200 vs multi-H100

Two H100s (2×80GB=160GB sharded, ~$5/hr) can match the B200's VRAM for big models. But sharding adds serve/comms overhead and the B200's Blackwell FP8 is ~2× H100's effective throughput. The B200 wins on simplicity and throughput; two H100s can win on raw $/hr flexibility.

B200 vs A100

The A100 (80GB, $1.67/hr) is one-third the B200's price and deep in supply, but can't come close to 192GB or Blackwell speed. Different jobs entirely: you rent a B200 to run a specific giant model; you rent an A100 for cheap batch/fine-tuning on smaller models.

B200 rental vs buying

A B200 SXM is roughly $30,000-40,000 to buy (≈₹25 lakh+ in India), and is scarce — many buyers can't get allocation at all. Renting from $4.50/hr (~$3,285/month at 730h) is the fastest way to run 400B-class models without a months-long hardware allocation wait.

Why rent a B200: it is next-gen hardware with fast obsolescence and tight supply — classic rent-not-buy. Lease compute by the hour to run the biggest models today, and let the provider absorb depreciation. See full ₹-based TCO in GPU rental vs buy in India.

Cheapest B200 rental (right now, live): $4.50/hr on-demand at Spheron with spot from $2.12/hr (53% off); Nebius lists $5.00/hr on-demand, $2.90/hr spot. Only these 2 providers — expect allocation lead times on sustained capacity.

B200 Rental — FAQ

Honest answers for teams renting NVIDIA B200 SXM compute.

How much does it cost to rent an NVIDIA B200 SXM?

Live on this page: B200 on-demand from $4.50/hr (≈₹374/hr) at Spheron; Nebius at $5.00/hr. Spot from $2.12/hr (Spheron, 53% off). A day is ~$108; a 730-hour month ~$3,285.

Why is the B200 so expensive to rent?

It is the flagship: 192GB VRAM (most of any single GPU) and ~2× H100 throughput. Only 2 providers list it and high-end capacity is scarce. For models that fit 141GB or less, a cheaper H200 or multi-H100 setup is often a better price per token.

Is the B200 the only option for 400B-class models?

As a single GPU, yes — 192GB is the only card that fits 400B-class MoE (DeepSeek R1/V3, Qwen-MoE) without sharding. You could shard across multiple H100/H200s, but that adds overlap and serve overhead. The B200 is the simplest path to the biggest open models.

B200 on-demand vs spot?

On-demand is guaranteed — right for production serving. Spot at Spheron is 53% cheaper ($2.12/hr) but interruptible — fine for training and checkpointing batch runs. With only 2 providers, verify capacity/lead time before planning.

How quickly can I get a B200?

On marketplaces like Spheron, capacity can be available in minutes to hours when listed; for sustained high-end capacity expect allocation lead times (some clouds quote 2-4 weeks for bulk high-end H100/B200).