NVIDIA B200 rental price per hour
Series 1 · GPU prices · NVIDIA · snapshot 16 Oct 2026
A Blackwell B200 hour costs $6.25 at the median published rate, 1.8× the H100 median. The floor is $3.69; the ceiling, $16.11, is a Google Cloud list price.
Figure 1B200: median rental price and published range, USD per GPU-hour, 2025-06-01 to 2026-10-14
§1Reading the B200 market
DeepInfra holds the cheapest confirmed on-demand B200 at $3.69, a few cents under Packet.ai's $3.75 Dynamic tier. Verda advertises $3.12, but that is a spot rate; its on-demand price is $6.23, so the lower number is kept out of the floor.
Several clouds publish ranges rather than one figure, depending on cluster size and commitment: Civo $3.79–$4.49, Nebius $3.95–$7.15, CoreWeave $4.26–$8.60. The trackers disagree by more than a dollar — Silicon Data reads ~$5.45, AIMultiple ~$6.52, Vast.ai's marketplace median ~$6.82 — which the tracker treats as normal cross-source noise rather than a trend. Oracle bare metal ($14.00) and Google's a4-highgpu-8g ($16.11) form the expensive end.
The series starts at $8.90 in Jun 2025 and has drifted down to $6.25, with the last move on 28 Sep 2026 ($6.01 → $6.25). No provider in the record publishes a clean 12-month B200 rate, so the reserved column stays empty rather than estimated.
Table 1B200 prices by provider, USD per GPU-hour
| Provider | Offer | USD / GPU-hr | Detail | As of |
|---|---|---|---|---|
| Verda | Spot | $3.12 | spot, excluded from the on-demand floor | 2026-09-28 |
| DeepInfra | On-demand | $3.69 | 2026-09-28 | |
| Packet.ai | On-demand | $3.75 | Dynamic tier | 2026-09-28 |
| Civo | On-demand | $3.79–$4.49 | 2026-09-28 | |
| Nebius | On-demand | $3.95–$7.15 | 2026-09-28 | |
| CoreWeave | On-demand | $4.26–$8.60 | 2026-09-28 | |
| Silicon Data index | Third-party index | ~$5.45 | 2026-09-28 | |
| Verda | On-demand | $6.23 | 2026-09-28 | |
| AIMultiple index | Third-party index | ~$6.52 | 2026-09-28 | |
| Vast.ai | Marketplace | ~$6.82 | marketplace median | 2026-09-28 |
| Oracle Cloud | On-demand | $14.00 | bare metal | 2026-09-28 |
| Google Cloud | On-demand | $16.11 | a4-highgpu-8g | 2026-09-28 |
Rows are the prices recorded in the tracker note for this model; 3 of the 12 come from the provider's own price page, the rest through comparison trackers. Third-party index rows are shown for context and not counted as providers. Every page behind this table: Sources, B200.
§2Specifications
Table 2DGX B200 system (8 GPUs); per-GPU memory = system total / 8
| Memory per GPU (index) | 180 GB |
| GPUs per system | 8x NVIDIA Blackwell GPUs |
| GPU memory, system | 1,440 GB total (180 GB per GPU) |
| Memory bandwidth, system | 64 TB/s HBM3e |
| FP8 Tensor Core, system, with sparsity | 72 PFLOPS |
| NVLink, system | 14.4 TB/s aggregate |
| System power | ~14.3 kW max |
Quoted from NVIDIA — DGX B200, checked 17 Oct 2026. Vendor tensor figures are "with sparsity" unless stated; dense throughput is half.
§3What moves the B200 price
- System-level power: NVIDIA lists the eight-GPU DGX B200 at about 14.3 kW maximum, roughly 1.8 kW per GPU slot including CPUs and networking — hosts need high-density racks, which limits who can offer it.
- Memory per GPU: 180 GB (1,440 GB per eight-GPU system), more than double the H100's 80 GB.
- Range pricing: CoreWeave and Nebius quote bands rather than a single rate; where you land inside the band depends on cluster size and commitment.
- Hyperscaler premium: the Google and Oracle list prices are 2.2–2.6× the median.
§4Rent or buy
A full eight-GPU B200 node rented continuously at the median is $36,500 a month; at DeepInfra's floor it would be $21,550.
B200 is bought as an eight-GPU system (HGX or DGX class) rather than a card. No system price appears in our cited sources, so none is shown. Ownership also implies roughly 14 kW per node of power and cooling capacity, which not every colocation hall can supply per rack — check the AI-ready capacity in the campus census before planning a purchase.
Where B200-class density can actually be hosted: AI data center census.
Other models with their own page: GB200, MI355X, H200, H100, MI300X, A100, L40S, RTX 4090. All models: GPU rental price index. How the figures are built: methodology.