B200 cloud pricing per GPU-hour: AWS, Google, Oracle and neoclouds, November 2026

Several providers raised their B200 list prices in 2026, and a reservation saves most over three years. Below: the list prices of the large clouds and the main GPU clouds on 10 November 2026, what a reservation saves against the same provider's on-demand price, the few public prices by term, and what one 8-GPU server costs per month.

B200 list prices, 10 November 2026

Each price as the provider's own page or price API showed it on 10 November 2026, before tax, per GPU-hour; where the provider prices a server of 8 GPUs we divide by 8. Lambda's page shows instance prices that take effect on 16 November 2026 at 00:00 UTC and no longer shows the current instance price; $6.69 for the 8-GPU instance comes from its archived copy of 2 November. Its 1-Click Cluster prices apply now. B300, GB200 and GB300 next to these: cloud GPU prices.

Across all providers: the median and the range

getdeploying compares 44 providers on its B200 page. On 10 November 2026 it counted 74 priced on-demand configs from 22 providers with a median of $8.04 per GPU-hour. The lowest listed was $4.11, not verified in stock; the lowest it verified in stock was $6.10. Spot (interruptible) B200 had a median of $4.46. Eight providers also quote B200 on a custom contract, priced per deal (getdeploying, B200).

What a reservation saves

The fair comparison is a provider's reserved price against its own on-demand price for the same GPU. Measured that way across all GPUs, getdeploying puts the median discount at about 14% for 1 month, 16% for 6 months, 26% for 1 year and 48% for 3 years (getdeploying GPU price trends, 9 November 2026). For B200, AWS's 3-year EC2 Instance Savings Plans are 57% to 62% below its on-demand price (cloud GPU prices). Google's A4 costs $7.09 per GPU-hour on a 3-year commitment against $8.06 on Flex-start, while one year, $11.12, costs more than Flex-start. Together AI lists $6.79 for 91 to 180 days against $8.19 by the hour, and an AWS Capacity Block costs about the same as AWS on demand.

getdeploying's B200 page pools every reserved config, whatever its term, into one median: $6.02 against $8.04 on demand on 10 November 2026. The two medians come from different providers and configs, so their gap is not a discount anyone gets.

Public B200 prices by term

Week of 2 November 2026, the last complete week, from the getdeploying GPU price dataset (GetDeploying, CC BY 4.0); the week of 9 November is still partial. getdeploying captures prices once a day, and each figure of a weekly row, the lowest and highest too, is the median of that week's daily values, column by column. Listings count priced offers not marked sold out. Reservations without a stated term are left out.

Only two to five providers publish a B200 price for each term, so each median here is the median of those few providers, and medians of different terms, or a term against on demand, compare different providers. Inside a term the spread is wide: 36-month listings run from $3.79 to $8.74.

Are B200 prices going up?

getdeploying's B200 page says the median on-demand price has risen about 24% over the past 90 days and about 25% over the past 12 months, to $8.04 per GPU-hour (getdeploying, B200, 10 November 2026). Its weekly dataset now counts more than twice as many providers with an on-demand B200 price as a year ago, 24 in the week of 2 November 2026 against 11 in the week of 10 November 2025, and getdeploying notes that a median can move without any price changing when a provider starts or stops listing a GPU (GPU price trends). Named providers did raise their own list prices: Lambda's 8-GPU instance goes to $7.69 per GPU-hour on 16 November 2026, against $6.69 in its archived copy of 2 November, and the earlier steps at AWS, Nebius and Together AI are on cloud GPU prices.

What one 8-GPU server costs per month

At list prices, for 8 GPUs and 730 hours (an average month): about $83,200 at AWS on demand, $81,800 at Oracle, $50,200 at CoreWeave, $47,000 at the median on-demand price of 10 November and $41,400 on Google's 3-year commitment. Buying the same server outright costs about $360,000 to $550,000; see B200 price.

Beyond public price lists

In the week of 2 November no B200 price list that getdeploying tracks went beyond three years, and they are priced per server or per small cluster. Blocks of hundreds or thousands of GPUs on longer terms are quoted per deal. In our pipeline 7 operators have sent us B200 offers since September 2026, operators in Europe (4) and the US (2) among them; in current offers some can start within 30 days of signing, typically on 2-year terms. We publish rates only from operators' public price lists or with their written consent, so we show none for them here.

We ask operators whether they can cover your request without naming you, and introduce you to every operator that can cover it. Price and terms come from the operator, you sign with the operator directly, and the operator pays our fee. See how it works and B200 rental.

Send a request

Requirements are handled confidentially. Your name stays with us until the operator signs our NDA and referral agreement. We reply within one business day from dima@ancapex.io.

FAQ