# How much does a B200 cost per hour?

> How much does a B200 cost per hour? $5.43 to $14.24 on demand at 9 providers on 29 Oct 2026, and $4.61/hour here in crypto, no KYC.

- URL: https://gpuwithcrypto.com/gpu-prices/b200/
- Last updated: 2026-11-03
- Publisher: GPUwithCrypto (https://gpuwithcrypto.com)

Every on-demand B200 price we found, the spot prices that beat ours, what fits in 180 GB, and who takes crypto or asks for ID.

No ID and no card: top up in crypto, start a B200 as soon as the top-up is credited.

**Quick answer — How much does a B200 cost per hour?**

As of 29 October 2026, an NVIDIA B200 with 180 GB HBM3e costs from $5.43 to $14.24 per GPU-hour on demand across 9 providers, with a median of $8.50. On **GPUwithCrypto** it costs $4.61/hour, billed per minute and paid in crypto with no KYC. Interruptible B200s go lower: Nebius lists preemptible B200s from $0.99, and CoreWeave and Verda sell spot B200s at $4.26 and $4.39.

## How much does a B200 cost per hour at each provider?

One row per provider at its cheapest on-demand B200 configuration, sorted by price. The most expensive on-demand B200 we found (AWS, $14.24 per GPU in an 8-GPU instance) costs 2.6 times the cheapest.

| Provider | Configuration | On-demand $/GPU-hour | Spot / marketplace $/GPU-hour | Billing increment | Accepts crypto | ID / KYC |
| --- | --- | --- | --- | --- | --- | --- |
| **GPUwithCrypto** | B200 180GB, 1 to 8 GPUs; GPU count shown per host | **$4.61** | — (on-demand only) | Per minute | Yes | No |
| Massed Compute | B200 SXM6, 8-GPU node ($43.46/hour) ÷ 8 | $5.43 | — | Hourly rate (increment not stated) | Yes (BTC) | None documented |
| Hyperbolic | B200 (form factor and GPU count not stated) | $5.99 | — | — | Unknown | None documented |
| Lambda | B200 SXM6, 8-GPU instance (1 GPU: $6.99; 2: $6.89; 4: $6.79) | $6.69 | — | Per minute | No (cards only) | None documented |
| RunPod | B200 (form factor not stated), 1 GPU | $7.99 | $5.98 (Community Cloud) | Per second | Yes | KYC before first crypto payment |
| Nebius | B200 SXM (HGX) | $8.50 | From $0.99 (preemptible) | — | No | None documented |
| CoreWeave | B200 SXM (HGX), 8-GPU instance ÷ 8 | $8.60 | $4.26 (spot) | — | Unknown | Unknown |
| Verda (formerly DataCrunch) | B200 SXM6, 1 GPU | $8.78 | $4.39 (spot) | — | No | None documented |
| Spheron | B200 (form factor not stated) | $11.12 | $5.95 (spot) | Per minute, 20-minute minimum | Yes (USDC) | None documented |
| AWS | B200, 8 GPUs (p6-b200.48xlarge) ÷ 8 | $14.24 | — | — | No | None documented |
| Google Cloud | B200, 8-GPU VM ÷ 8 (a4-highgpu-8g) | Shown as N/A | $4.95 (spot) | Per second, 1-minute minimum | No | None documented |
| Crusoe | B200 180GB (HGX) | Contact sales | — | Per hour or per second | No | None documented |
| Vast.ai | B200, marketplace | — | $7.50 (Vast.ai's public price feed, cheapest of 50 rentable offers) | Per second | Yes (BitPay, Crypto.com) | None documented |
| Shadeform | B200, 8 GPUs, resold from other clouds | — | None available (11 listings, lowest listed $3.74) | — | Yes (USDC, per Stripe) | None documented |

*Prices checked on October 29, 2026.*

USD per GPU-hour, taxes excluded, rounded to the cent. Multi-GPU prices are divided by the GPU count. Hyperscaler prices cover the whole VM (AWS US East, Google Cloud Iowa). 'None documented' means no ID step in the official sign-up, billing or terms pages we read, not a promise that a provider never asks. RunPod's Community Cloud is community-hosted. Google Cloud, Crusoe, Vast.ai and Shadeform are not in the median.

## Is the B200 worth the extra cost over an H200?

Yes when your model needs more than 141 GB per GPU, or when your job runs in FP8 or FP4. According to GPUwithCrypto's price check of 29 October 2026, a median on-demand B200 costs 61% more per hour than a median H200 ($8.50 against $5.29). For that it has 28% more memory (180 GB against 141 GB), 67% more bandwidth (8,000 against 4,800 GB/s) and 2.3 times the dense FP8 rating (4,500 against 1,979 TFLOPS).

On paper, B200 compute is cheaper: at median prices, a dense FP8 TFLOPS costs 29% less per hour on a B200 than on an H200. These are NVIDIA's peak ratings, not measured speeds. The B200 also has FP4 Tensor Cores, rated 9,000 TFLOPS dense; NVIDIA gives the H200 no FP4 rating.

Memory is the clearest gain. Llama 3.1 70B in BF16 needs about 168 GB: one B200, or two H200s. The B200 is the only GPU we rent that holds it on one card. When your model already fits in 141 GB, the H200 costs less per hour: $3.07/hour here. The [H200 price page](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?") has every H200 row, and the [H100 vs H200 comparison](https://gpuwithcrypto.com/compare/h100-vs-h200/ "H100 vs H200: is the H200 worth the extra cost?") covers the step below.

## Why are some spot and preemptible B200s cheaper than ours?

Because the provider can take them back. On 29 October 2026, three interruptible B200 prices were below our $4.61:

- **Preemptible:** Nebius listed B200s from $0.99 per GPU-hour. The rate floats, capped one cent below its on-demand price, and Nebius can reclaim the GPU.
- **Spot:** CoreWeave's 8-GPU B200 instance cost $34.11 per hour on spot, or $4.26 per GPU. Verda's single-GPU spot B200 cost $4.39.
- **Above ours:** Google Cloud's spot B200 cost $4.95 per GPU, Spheron's spot $5.95 and RunPod's Community Cloud $5.98. The cheapest rentable B200 in Vast.ai's public price feed was $7.50, among 50 offers with a median of $8.75.

Shadeform's API listed B200s from $3.74 per GPU-hour, but none of its 11 on-demand listings was available when we read it. Spot suits jobs that save checkpoints and can restart. A long training run or a live inference server needs a GPU that stays put. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. The [spot GPU guide](https://gpuwithcrypto.com/what-is/spot-gpu/ "What is a spot GPU, and is it worth it?") explains when an interruption is worth the saving.

## Which B200 sets the reference price, and can you rent a single GPU?

The reference is Massed Compute's 8-GPU B200 SXM6 node at $43.46 per hour, or $5.43 per GPU-hour. Massed Compute lists no smaller B200 machine, and the AWS price is for an 8-GPU instance.

Single B200s exist, at a higher rate per GPU. Lambda charges $6.99 for one GPU and $6.69 per GPU in an 8-GPU instance. RunPod lists one B200 at $7.99 and Verda at $8.78.

Three clouds show no self-serve on-demand B200 price. Crusoe says 'Contact sales', Google Cloud lists the on-demand price of its 8-GPU B200 VM as N/A, and Vultr's API marks its 8-GPU B200 plan as not deployable on demand. None of them counts in the median.

GPUwithCrypto has one B200 price for every host. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect.

## How much does a B200 cost per day, week or month?

Our price for one GPU next to the lowest and the median self-serve on-demand price of 29 October 2026, for the same number of hours. Our hourly rate is 46% under the median.

| Period | GPU-hours | GPUwithCrypto | Lowest elsewhere (Massed Compute) | Median elsewhere |
| --- | --- | --- | --- | --- |
| 1 hour | 1 | **$4.61** | $5.43 | $8.50 |
| 1 day | 24 | **$110.64** | $130.38 | $204.00 |
| 1 week | 168 | **$774.48** | $912.66 | $1,428.00 |
| 30 days | 720 | **$3,319.20** | $3,911.40 | $6,120.00 |

*Prices checked on October 29, 2026.*

Massed Compute's rate is $5.4325 per GPU-hour before rounding ($43.46 ÷ 8). Our day, week and month match the 24, 168 and 720 GPU-hour blocks.

## What are the B200's specs?

NVIDIA publishes B200 figures only for its 8-GPU HGX B200 and DGX B200 systems: 1,440 GB of memory, 64 TB/s and 72 PFLOPS of sparse FP8. The B200 column divides those totals by 8; dense is half the sparse figure, per NVIDIA's own footnote. The H200 SXM is shown for comparison.

| Spec | B200 (per GPU, HGX B200) | H200 SXM |
| --- | --- | --- |
| Architecture, launch | Blackwell, 2024 | Hopper, 2023 |
| Memory | 180 GB HBM3e | 141 GB HBM3e |
| Memory bandwidth | 8,000 GB/s | 4,800 GB/s |
| FP4 Tensor, dense | 9,000 TFLOPS (18,000 sparse) | — |
| FP8 Tensor, dense | 4,500 TFLOPS (9,000 sparse) | 1,979 TFLOPS (3,958 sparse) |
| FP16 / BF16 Tensor, dense | 2,250 TFLOPS (4,500 sparse) | 989.5 TFLOPS (1,979 sparse) |
| TF32 Tensor, dense | 1,125 TFLOPS (2,250 sparse) | 494.5 TFLOPS (989 sparse) |
| FP32 | 75 TFLOPS | 67 TFLOPS |
| GPU-to-GPU link | NVLink 1,800 GB/s | NVLink 900 GB/s |
| Max power | 1,000 W | 700 W |

NVIDIA states the B200's NVLink speed per GPU; the other B200 figures are 8-GPU totals ÷ 8. The B200 power figure is the per-GPU maximum that nvidia-smi reports in a DGX B200.

## What fits on one B200?

Memory needed for common open models, using weights × 1.2 for inference (EleutherAI's rule of thumb) and weights plus gradients and optimizer for fine-tuning. The [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") has the formulas and more models.

| Model and job | Memory needed | One B200 180GB? |
| --- | --- | --- |
| Llama 3.1 70B inference, BF16 | ≈ 168 GB | Yes |
| Llama 3.1 70B, FP8 weights plus two full 128K-token contexts (BF16 cache) | ≈ 156 GB | Yes |
| Qwen2.5 72B inference, FP8 | ≈ 87.2 GB | Yes |
| Llama 3.1 70B LoRA fine-tune | ≈ 140 to 168 GB | Yes |
| Llama 3.1 8B full fine-tune | ≈ 128 to 144 GB before activations | Yes, with 36 to 52 GB left for activations |
| Llama 3.1 405B inference, 4-bit | ≈ 243 GB | No: 2× B200 |
| Llama 3.1 405B inference, FP8 | ≈ 486 GB | No: 3× B200 |
| DeepSeek-V3 or R1 inference, FP8 | ≈ 805 GB | No: one 8× B200 node (1,440 GB) |

Two 128K-token contexts for Llama 3.1 70B add 2 × 42.9 GB of BF16 KV cache to its 70 GB of FP8 weights. Only 37B of DeepSeek-V3's 671B parameters are active per token, but all of them must sit in memory.

## What does a B200 cost on GPUwithCrypto?

$4.61/hour, at least 15% under a named on-demand reference read on 29 October 2026: Massed Compute at $5.43. We do not claim to be the cheapest for every GPU; spot and marketplace offers can be lower. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire.

| Plan | Quantity | Price | Price per unit |
| --- | --- | --- | --- |
| Try | 10 GPU-hours | $46.10 | $4.61/hour |
| Day | 24 GPU-hours | $110.64 | $4.61/hour |
| Week | 168 GPU-hours | $774.48 | $4.61/hour |
| Month | 720 GPU-hours | $3,319.20 | $4.61/hour |

In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire

**Before you pay**

- Minimum top-up $25. Top up once in crypto, then buy any plan from your balance.
- Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it.
- Credited once your payment is checked on the blockchain. Typical network confirmation time: BTC ≈ 20 min · USDT TRC-20 ≈ 1 min · USDT ERC-20 ≈ 5 min · USDC ERC-20 ≈ 5 min · ETH ≈ 5 min · SOL ≈ 1 min · LTC ≈ 15 min

## B200 price questions

### What is the cheapest B200 cloud right now?

On 29 October 2026 the lowest self-serve on-demand B200 price we found was $5.43, for Massed Compute's 8-GPU node, and GPUwithCrypto charges $4.61. Interruptible B200s cost less: Nebius preemptible from $0.99, CoreWeave spot at $4.26 and Verda spot at $4.39. They can be taken back mid-job.

### How much does a B200 cost per month?

720 GPU-hours cost $3,319.20 on GPUwithCrypto. At the median on-demand price of 29 October 2026 the same month cost $6,120.00, and $3,911.40 at Massed Compute's rate.

### Can a 70B model run on one B200?

Yes, even in BF16. Llama 3.1 70B needs about 168 GB in BF16, within the B200's 180 GB. In FP8 it needs about 84 GB, which leaves room for two full 128K-token contexts.

### Can I rent a B200 with crypto?

Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC with no ID, no selfie and no card. Of the 9 self-serve B200 providers we checked, 3 accept crypto: Massed Compute (BTC), Spheron (USDC) and RunPod, which asks for KYC before the first crypto payment. The [crypto GPU cloud comparison](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?") lists coins and rules.

### Can I rent a single B200, or only 8-GPU nodes?

Both exist. Lambda rents 1, 2, 4 or 8 B200s, RunPod and Verda list single GPUs, and Massed Compute lists only an 8-GPU node. On GPUwithCrypto the GPU count depends on the host, and the dashboard shows it before you start.

## Sources

1. [Massed Compute — VM pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29)
2. [Hyperbolic — GPU marketplace](https://www.hyperbolic.ai/marketplace) — Hyperbolic (read 2026-10-29)
3. [Lambda — Pricing](https://lambda.ai/pricing) — Lambda (read 2026-10-29)
4. [Runpod — Pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29)
5. [Nebius — Prices](https://nebius.com/prices) — Nebius (read 2026-10-29)
6. [CoreWeave — Pricing](https://www.coreweave.com/pricing) — CoreWeave (read 2026-10-29)
7. [Verda — Pricing](https://verda.com/pricing) — Verda (read 2026-10-29)
8. [Spheron — Pricing](https://www.spheron.network/pricing/) — Spheron (read 2026-10-29)
9. [AWS — EC2 on-demand price list, US East (N. Virginia)](https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/US%20East%20(N.%20Virginia)/Linux/index.json) — Amazon Web Services (read 2026-10-29)
10. [Google Cloud — Accelerator-optimized VM pricing](https://cloud.google.com/products/compute/pricing/accelerator-optimized) — Google Cloud (read 2026-10-29)
11. [Crusoe — Cloud pricing](https://www.crusoe.ai/cloud/pricing) — Crusoe (read 2026-10-29)
12. [Vultr — Bare-metal plans API](https://api.vultr.com/v2/plans-metal?per_page=500) — Vultr (read 2026-10-29)
13. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29)
14. [Shadeform — Instance types API](https://api.shadeform.ai/v1/instances/types) — Shadeform (read 2026-10-29)
15. [Runpod — Billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29)
16. [NVIDIA HGX platform](https://www.nvidia.com/en-us/data-center/hgx/) — NVIDIA (read 2026-10-29)
17. [NVIDIA DGX B200](https://www.nvidia.com/en-us/data-center/dgx-b200/) — NVIDIA (read 2026-10-29)
18. [Introduction to the NVIDIA DGX B200 system](https://docs.nvidia.com/dgx/dgxb200-user-guide/introduction-to-dgxb200.html) — NVIDIA (read 2026-10-29)
19. [NVIDIA H200 Tensor Core GPU](https://www.nvidia.com/en-us/data-center/h200/) — NVIDIA (read 2026-10-29)
20. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29)

---

GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026.