Blackwell and Hopper, one axis

The Blackwell price ladder: B300, B200, H200 and H100

Four GPUs on one price axis. Each dot is one provider's lowest provider-set on-demand price, the accent tick is the median of those, read 12 November 2026. The B300 median is 1.14 times the B200 median, and the B300 has the fewest providers of the four.

B300 lowest: $6.60
B200 lowest: $3.50
B300 median over B200 median: 1.14x
Lowest per GB, B300: $0.023

§ 01

Price per GPU-hour

Lowest, median and highest on one axis

Lowest provider-set on-demand price per GPU-hour by provider for four GPUs, read 12 November 2026

§ 02

Price against memory and bandwidth

Catalog memory as sold, maker figures where stated

Four GPUs: providers, price statistics, price per GB of memory and the maker's stated bandwidth and FP8 rate
GPU	Memory as sold	Providers	Lowest	Median	Highest	Lowest per GB	Median per GB	Bandwidth, maker	FP8 dense, maker
B300	288 GB	11	$6.60	$9.50	$17.80	$0.023	$0.033	8 TB/s	5,000 TFLOPS
B200	180 GB	10	$3.50	$8.34	$14.24	$0.019	$0.046	8 TB/s	5,000 TFLOPS
H200	141 GB	15	$3.49	$5.40	$10.60	$0.025	$0.038	4.8 TB/s	not stated
H100	80 GB	19	$1.99	$3.90	$11.06	$0.025	$0.049	3.35 TB/s	not stated

Price per GB is the price per GPU-hour divided by the catalog memory in GB: B300 288, B200 180, H200 141, H100 80. A listing may name a different size, see below. Maker figures link to the page they were read from: B300, B200, H200, H100.

§ 03

Memory, as sold and as stated

Why per-GB figures carry a caveat

Memory per GPU: catalog figure, maker maximum, datasheet figure where the specs note quotes one, and sizes named in provider listings
GPU	Catalog	Maker maximum	Datasheet figure quoted	Named in listings
B300	288 GB	288 GB	279 GB	268 GB (Verda)
B200	180 GB	192 GB	not stated	180 GB (Lambda and Verda)

The maker's comparison table gives the maximum memory of the chip. Parts that ship in instances can carry less, and the sizes that listings name are the ones that count for fitting a model. The per-GB column uses the catalog size, so read it as a rough guide.

§ 04

What this ladder does not price

Only what the data holds

The dataset holds 0 GB200 rows and 0 GB300 rows. We make no price claim about either, and we do not infer one from the B200 or the B300. Hopper GPUs other than the H200 and H100, and AMD and Intel parts, are outside this page. Spot and reserved rows are on their own pages and are not in these figures.

§ 05

Questions

Computed from the table

Q1How much more does a B300 cost than a B200 per hour?

By each GPU's median provider price, the B300 is $9.50 and the B200 is $8.34: 1.14 times, or $1.16 (14%) above the B200 median. By lowest price, $6.60 at Massed Compute against $3.50 at Vultr. Those are different providers, so for a like-for-like premium see the provider-by-provider comparison.

Q2Which of the four has the lowest price per GB of memory?

B200, at $0.019 per GB-hour on its lowest provider-set price ($3.50 over 180 GB). The order from lowest is B200 $0.019, B300 $0.023, H200 $0.025, H100 $0.025. Using each GPU's median price instead, the figures are B300 $0.033, B200 $0.046, H200 $0.038, H100 $0.049.

Q3How does the B300 compare with the H200 and the H100?

The lowest B300 price is 1.89 times the lowest H200 price ($6.60 against $3.49) and 3.32 times the lowest H100 price ($1.99). The H100 has 19 providers on this basis, the H200 15, the B200 10 and the B300 11.

Q4Is there GB200 or GB300 pricing here?

No. The dataset holds 0 GB200 rows and 0 GB300 rows, so this page claims nothing about either.

Q5Where do the memory and bandwidth figures come from?

From the maker's own pages, linked in the table, read 2026-11-12. Where a figure is not stated there, the cell says so and we do not estimate it. The B300 and B200 figures are the maker's maximums from one comparison table; shipping instances can carry less memory.

Also on this desk

Disclosure

B300 Watch is published by the Runcrate team, who also run Runcrate, an AI cloud. Runcrate is listed in the table for B300, marked "ours, live". It is sorted by price like every other row, with no pinned position. Other providers' prices are read from their own public pricing pages. We have no paid placement.