The NVIDIA B200 is one of the most advanced GPUs you can rent in 2026, built on the Blackwell architecture with 180 GB of memory per GPU. Demand is extremely high, which keeps pricing volatile and access limited.
With the Blackwell line, NVIDIA dropped PCIe from its major data center GPUs. Whereas you could plug an H100 into a standard server, the B200 can only be housed in top-tier systems.
This guide covers NVIDIA B200 pricing for October 2026, including public rates across hyperscalers and specialist GPU clouds. It also breaks down the full B200 spec sheet, compares the chip against the H100, H200, and B300, and shows which LLMs fit on a single card.
Key Takeaways
- Specialist clouds are the value tier. Hyperbolic, Hyperstack, Modal, Lambda, Runpod, Nebius, and Vast.ai price the B200 in the $5.99-$8.50 range, versus $14-$16 on hyperscalers.
- Choose B200 only when you must. Very large model training that overflows H100 or H200 memory.
- Thunder Compute roadmap. We don't offer B200 nodes. You can launch an H100 80GB at $3.20/hr, or 2x H100 for $6.40, with one-click VS Code, per-minute billing, and persistent storage.
The pricing information in this guide is reviewed weekly.
NVIDIA B200 Pricing
B200 pricing is not standardized: the same GPU starts at $5.99/GPU-hr on specialist clouds and over $16 on hyperscalers. The B200 is now broadly available on the cloud, but a tiered provider market and enterprise-committed capacity keep public rates spread out.
| Provider | GPU / Instance | On-Demand $/GPU-hr* | Notes |
|---|---|---|---|
| Hyperbolic | B200 | $5.99 | |
| Hyperstack | B200 | $6.00 | |
| Modal | B200 | $6.25 | |
| Vast.ai | B200 | $6.29 | Median of 3 distinct verified US/CA hosts; ranges $6.28-$7.84/GPU-hr. Priced as compute plus 100GB storage, with bandwidth billed separately. |
| Lambda | B200 | $6.69 | |
| Runpod | B200 | $6.79 | |
| Nebius | B200 | $8.50 | |
| Vultr | B200 | $8.50 | |
| CoreWeave | 8 x B200 | $8.60 | Normalized from an 8-GPU node. |
| Oracle Cloud | BM.GPU.B200.8 | $14.00 | |
| AWS | p6-b200.48xlarge | $14.24 | Normalized from an 8-GPU node. |
| Google Cloud | a4-highgpu-8g | $16.11 | On-demand-equivalent. Only has reservation, Spot, or Flex-start capacity. Normalized from an 8-GPU node. |
| Last reviewed on October 1, 2026. | |||
Methodology: Why You Can Trust These Numbers
- Public rates only. The table uses on-demand rates where available. Google Cloud B200 is shown as an on-demand-equivalent rate for reservation, Spot, or Flex-start capacity.
- Same silicon. Every row is a 180 GB NVIDIA B200 (SXM).
- Public price lists only. Figures come straight from each provider's pricing page or API in October 2026.
- US regions, USD. Node prices are normalized to a per-GPU figure where a provider only sells full 8-GPU nodes.
B200 Cost Benchmark
The 10-hour column shows what a single B200 costs for a typical training or inference session, which makes the gap between specialist clouds and hyperscalers concrete.
| Provider | On-Demand $/GPU-hr | 10-Hour Cost |
|---|---|---|
| Hyperbolic | $5.99 | $59.90 |
| Hyperstack | $6.00 | $60.00 |
| Modal | $6.25 | $62.50 |
| Vast.ai | $6.29 | $62.92 |
| Lambda | $6.69 | $66.90 |
| Runpod | $6.79 | $67.90 |
| Nebius | $8.50 | $85.00 |
| Vultr | $8.50 | $85.00 |
| CoreWeave | $8.60 | $86.00 |
| Oracle Cloud | $14.00 | $140.00 |
| AWS | $14.24 | $142.42 |
| Google Cloud | $16.11 | $161.10 |
Bottom line: 10 hours on the cheapest B200 costs more than 22 hours of an H100 80GB on Thunder Compute at $3.20/hr. For workloads that fit in 80 GB, on-demand H100 availability is difficult to match with Blackwell GPUs.
B200 Hardware Price: Buy vs Rent
After comparing cloud rates, it helps to understand what a B200 actually costs to own. NVIDIA does not publish list prices, and the B200 does not ship as a standalone PCIe card. The figures below come from OEM reseller quotes and secondary-market listings tracked through 2026.
| Configuration | Price Range1 | Notes |
|---|---|---|
| Single B200 (SXM), effective street price | $45,000-$55,000 | Street quotes under 2026 supply constraints; list price in 8-GPU volumes runs lower (~$30,000-$40,000). |
| 8-GPU HGX B200 system | $450,000-$550,0001 | Integrated platform with NVLink fabric, NICs, and CPUs. |
| 1 An integrated 8-GPU system is an estimate: 8 GPUs plus $120,000-$180,000 of CPU, networking, chassis, and integration. NVIDIA does not publish list prices; ranges reflect reseller and system-implied estimates. | ||
At $45,000-$55,000 per GPU before the server around it, a B200 rented at $6/hr would take roughly 7,500-9,200 GPU-hours to match in equivalent hardware cost. Each card also draws about 1,000 W, making power distribution and cooling a real engineering challenge.
Cloud rental is the right choice for most teams: no capital tied up, no cooling or power engineering, no depreciation, and the flexibility to switch hardware as workloads change.
NVIDIA B200 Specifications
The B200 pairs two Blackwell dies into a single SXM module with 180 GB of HBM3e and native FP4 support, a full generational step beyond the Hopper-based H100 and H200.
| Spec | NVIDIA B200 |
|---|---|
| Architecture | Blackwell (dual-die) |
| Tensor Core Generation | 5th Gen (native FP4) |
| VRAM | 180 GB HBM3e |
| Memory Bandwidth | Up to 8,000 GB/s1 |
| NVLink | 5th Gen, 1.8 TB/s per GPU (18 ports x 100 GB/s) |
| Form Factor | SXM6 |
| TDP | 1,000 W |
| Cooling | Liquid cooling required |
| 1 Peak theoretical figure. Real-world bandwidth varies with hardware and software bottlenecks. | |
B200 vs H100 and H200
The B200's advantage is memory capacity, bandwidth, and native FP4, not a free performance win across every workload. For any compute-bound job that fits inside 80 GB, the older Hopper cards deliver the work at a fraction of the price.
| Feature | NVIDIA H100 | NVIDIA H200 | NVIDIA B200 |
|---|---|---|---|
| Architecture | Hopper | Hopper | Blackwell |
| GPU Memory | 80 GB HBM3 | 141 GB HBM3e | 180 GB HBM3e |
| Memory Bandwidth | 3.35 TB/s | 4.8 TB/s | 8 TB/s |
| Native FP4 Support | No | No | Yes |
| NVLink | 900 GB/s | 900 GB/s | 1.8 TB/s |
| On-Demand Cloud Price | $2.99 - $10.98/hr | $3.99 - $10.85/hr | $5.99 - $16.11/hr |
Choose the B200 for new large-model training, or where FP4 serving and 180 GB per card meaningfully cut GPU count.
- For workloads under 141 GB, the H200 is cheaper and more available
- For anything under 80 GB, the H100 delivers strong FP8 compute; compare provider rates before choosing it over B200.
- Two H100 GPUs can match a B200 for workloads that don't need Blackwell, with a comparable combined VRAM pool and strong NVLink scaling.
Compare current H200 rates in our NVIDIA H200 pricing guide.
B200 vs B300: When to Step Up
The B300 (Blackwell Ultra) is the memory-heavy sibling of the B200, carrying 288 GB of VRAM versus 180 GB and drawing 1,400 W versus roughly 1,000 W. It rents for about a dollar more per GPU-hour at every provider that lists both.
Step up to the B300 only when a single model needs more than 180 GB, for example a dense 130B-class model in FP16 that would otherwise span two B200s. For everything that fits in 180 GB, the B200 is the better value.
See the full breakdown in our NVIDIA B300 pricing guide.
What LLMs Fit on a B200?
The B200's 180 GB of VRAM determines which models run on a single card versus requiring multi-GPU parallelism. The estimates below use published weight sizes and assume 20-30% overhead for KV cache, activations, and framework buffers at a moderate context window.
| Model | Parameters | Precision | VRAM (weights only)1 | Single B200 (180 GB)? |
|---|---|---|---|---|
| Llama 3 70B | 70B | FP16 | ~140 GB | Yes, tight with overhead |
| Qwen 3 72B | 72B | FP16 | ~144 GB | Yes, tight with overhead |
| Dense 130B-class model | 130B | FP16 | ~260 GB | No |
| DeepSeek R1 | 671B | FP8 | ~670 GB | No |
| 1 Weights only. Add 20-30% for KV cache, activations, and framework overhead at moderate context. | ||||
An 8-GPU B200 node provides 1,440 GB of pooled VRAM (8 x 180 GB), enough to serve DeepSeek R1 at FP8 and other large open-weight models. Where a single model overflows 180 GB but you would rather not shard, the B300's 288 GB is the more direct answer.
AWS P6-B200 Instances
AWS offers the B200 through its flagship EC2 P6 instance family in an 8-GPU configuration. As usual for AWS, egress fees are $0.09/GB after the first free 100 GB, which makes the effective cost even higher than the listed price.
| SKU | GPUs | vCPUs | RAM | Hourly Price | Price Per-GPU |
|---|---|---|---|---|---|
| p6-b200.48xlarge | 8 x B200 | 192 | 2,048GB | $113.93 | $14.24 |
Google Cloud A4 Instances
Google Cloud lists the NVIDIA B200 through the A4 high-GPU family. The public on-demand-equivalent price for a4-highgpu-8g is $16.11 per GPU-hour. This instance is only available through reservation, Spot, or Flex-start.
| SKU | GPUs | Region | Hourly Price | Price Per-GPU |
|---|---|---|---|---|
| a4-highgpu-8g | 8 x B200 | us-central1, us-east1, us-west1 | $128.88 | $16.11 |
Why NVIDIA B200 Pricing Is So High
Three factors keep B200 rates well above older GPUs:
- Supply constraints. Blackwell GPUs are in extremely high demand for large-scale AI training.
- Enterprise-only access. Most B200 deployments are reserved for large enterprise customers, so on-demand supply for smaller teams is thin.
- Infrastructure requirements. The B200 only runs in liquid-cooled, high-power systems with fast interconnects, raising the cost of every GPU-hour a provider sells.
Last Thoughts on NVIDIA B200 Pricing
The B200 is a remarkable piece of hardware. If you are running very large model training and have the budget, its raw performance gains may well justify the premium. Pricing stays volatile and availability stays constrained while the Blackwell rollout matures, so most teams will find better ROI elsewhere.
Explore the full cloud GPU market landscape in our AI GPU rental market trends analysis.
FAQ
What Is the NVIDIA B200 Cloud Price in 2026?
B200 pricing ranges from about $5.99-$16.11 per GPU-hour, with specialist clouds at the low end and hyperscalers at the high end. Availability is limited and often restricted to enterprise customers.
What Is the Cheapest NVIDIA B200 Cloud Provider?
Hyperbolic lists the lowest tracked fixed B200 rate at $5.99/GPU-hr as of October 2026. Vast.ai now has a $6.29/GPU-hr marketplace median, while AWS starts at $14.24/GPU-hr on the p6-b200.48xlarge.
How Much VRAM Does the NVIDIA B200 Have?
The B200 offers up to 180 GB of HBM3e VRAM per GPU.
Can I Rent a Single B200 GPU, or Do I Need a Full 8-GPU Node?
Hyperscalers (AWS, Google Cloud, Oracle) require full 8-GPU nodes. Hyperbolic, Hyperstack, Modal, Lambda, Runpod, Nebius, and Vast.ai offer single-GPU B200 access for smaller workloads.
What Is the NVLink Bandwidth of the B200?
The B200 uses 5th Generation NVLink, giving each GPU 1.8 TB/s of bidirectional bandwidth across 18 ports of 100 GB/s each.
Is the B200 Better Than the H100?
The B200 is more powerful, but for most workloads the H100 offers better availability, lower cost, and sufficient performance. The B200 earns its premium mainly on very large model training.