Paid links may earn Compute Screener a referral fee. They never affect prices, weights or the order.
About NVIDIA B200
The NVIDIA B200 rental price today is $7.83 per GPU-hour, up 0.6% since yesterday, up 0.6% over the past week and up 23.9% over the past month. At that rate one B200 costs about $5,710 for a month of 730 hours. Over the past year it has ranged from $5.10 to $7.83. The index is the price of one benchmark contract: B200 SXM on HGX, on-demand, 8-GPU node, US hardware, CPU-neutral. It is standardized across 10 cloud providers and marketplaces, from Massed Compute at $5.43 to Vast.ai at $9.23, and is updated every 2 to 3 hours.
The B200 is NVIDIA's first Blackwell data-center GPU, announced on March 18, 2024: two dies working as one GPU, with 208 billion transistors on a custom TSMC 4NP process. Its launch slipped after NVIDIA changed a mask to improve production yields, disclosed in August 2024, and the first large clouds offered it in 2025. It roughly doubles the H100's memory and its FP8 compute, and adds FP4 for inference. With the H100, it is one of the two GPUs with rental futures filed by CME or set for listing by ICE.
Specifications
| Spec | B200 in HGX |
|---|
| Architecture | Blackwell, two dies, TSMC 4NP, 208 billion transistors |
| Memory | 192 GB HBM3e, 180 GB usable in HGX servers; 8 TB/s |
| FP4 tensor | about 9 PFLOPS dense per GPU, 18 with sparsity |
| FP8 tensor | about 4.5 PFLOPS dense per GPU, 9 with sparsity |
| BF16 / FP16 tensor | about 2.25 PFLOPS dense per GPU |
| NVLink | 1.8 TB/s, NVLink 5 |
| Power | about 14.3 kW for an 8-GPU DGX B200 |
| Other versions | GB200: one Grace CPU with two B200s. GB200 NVL72: 72 GPUs in a liquid-cooled rack |
Per-GPU figures are NVIDIA's 8-GPU HGX totals divided by eight, from its HGX specifications and DGX B200 page, accessed October 24, 2026.
Where the B200 is rented
Google Cloud's A4 machines, generally available on March 18, 2025, were among the first B200s from a large cloud, followed by AWS P6-B200 on May 15, 2025, at first only as reserved Capacity Blocks. Oracle, CoreWeave, Lambda, Nebius and Together AI also rent it; GetDeploying tracked 38 B200 providers on October 23, 2026.
What this price measures
The index is the posted on-demand price of one benchmark contract, a B200 in an 8-GPU HGX node of US hardware, rented by the hour from GPU clouds and marketplaces; prices for other node sizes are converted to that contract, and prices from other regions are counted without a regional adjustment for now. It is not a hyperscaler, reserved or spot price, and it does not cover GB200 rack-scale systems.
Listed B200 prices span almost four times. On October 23, 2026 they ran from $3.75 for a single GPU at the low end to $8.60 for an 8-GPU node at CoreWeave, $14.00 at Oracle and $14.24 at AWS per GPU-hour; GetDeploying's median was $7.99 on demand, $5.62 reserved and $4.26 spot.
What moves B200 rental prices
- Supply and power. Cloud GPUs were "sold out", NVIDIA said in November 2025, and in April 2026 all Blackwell capacity coming online through September 2026 was reported booked. An 8-GPU node draws about 14 kW, so data-center power and cooling limit how fast capacity grows.
- Successors. GB300 systems, built on the B300, brought in about two-thirds of NVIDIA's Blackwell revenue by late 2025, and NVIDIA said its next platform, Vera Rubin, was in full production in August 2026.
- Inference economics. FP4 lets the B200 serve models more cheaply per token than Hopper GPUs; NVIDIA's own figures for one open model are about $0.02 per million tokens on Blackwell against $0.09 on Hopper.
- Export controls. NVIDIA's August 2026 outlook assumes no data-center sales in China, so US and allied demand sets B200 rental prices.
Price history
This index's history before October 23, 2026 is GetDeploying's weekly B200 series, rescaled to join the daily index.
Futures and benchmarks
CME has filed B200 rental futures and ICE has set one for listing; neither was trading on October 24, 2026. CME Group's NYMEX contract GPU2 covers 730 GPU-hours and settles on Silicon Data's B200 index; its October 5 launch slipped when the CFTC extended its review to November 9, 2026. ICE Futures U.S. has set its B200 contract, BKL, settled on the monthly average of Ornn's B200 index, for listing with no date. The Bloomberg tickers beside the name are those two firms' B200 indices: Ornn's ORNNB200, from executed rentals, and Silicon Data's SDB200RT. CME's product page shows that code as SDRB200RT.
FAQ
How does the B200 compare with the H100 on price? A B200 rented for about twice an H100's hourly price in October 2026, with about 2.25 times its dense FP8 compute and memory and 2.4 times its memory bandwidth, according to NVIDIA's HGX specifications and Blackwell Ultra article. GetDeploying's medians were $7.99 against $3.85.
What is the difference between the B200 and GB200? The GB200 pairs two B200 GPUs with an NVIDIA Grace CPU, and the GB200 NVL72 joins 72 GPUs in one liquid-cooled rack. This index measures B200s in standard 8-GPU HGX servers.
How often is this index updated? Every 2 to 3 hours, with one value per day in US Central time that is final when the day closes.
Sources
Text reviewed Oct 24, 2026; the figures in it are as of that date. The figures at the top of this page update every 2 to 3 hours. Not investment advice.
