H100 vs B300 rental price per hour
The B300 costs 132% more per hour than the H100 at the median and has 260% more memory (288 GB against 80 GB). Per GB of memory, the B300 is 36% cheaper. Based on 28 H100 and 8 B300 public on-demand listings, last checked .
- H100 median
- $3.99
- 17 providers
- B300 median
- $9.25
- 8 providers
- H100 memory
- 80 GB
- Per GPU
- B300 memory
- 288 GB
- Per GPU
Side by side
H100 and B300 public on-demand prices
US dollars per GPU-hour from public pages. Every row behind these numbers is on the H100 price page and the B300 price page, each with its source link.
| Measure | H100 | B300 |
|---|---|---|
| Memory per GPU | 80 GB | 288 GB |
| Cheapest on-demand | $1.68 (Latitude.sh) | $7.40 (Hyperstack) |
| Median on-demand | $3.99 | $9.25 |
| Priciest on-demand | $12.29 | $17.80 |
| Providers with a price | 17 | 8 |
| Median price per GB of memory, per hour | $0.050 | $0.032 |
Same provider
Providers that list both
Lowest public on-demand price per GPU-hour at each provider that lists both chips.
| Provider | H100 lowest | B300 lowest | B300 vs H100 |
|---|---|---|---|
| Amazon Web Services | $6.88 | $17.80 | +159% |
| Civo | $2.99 | $7.49 | +151% |
| Crusoe | $3.90 | $8.96 | +130% |
| Hyperstack | $2.50 | $7.40 | +196% |
| Nebius | $4.50 | $9.50 | +111% |
| Oracle Cloud (OCI) | $10.00 | $15.00 | +50% |
| RunPod | $2.89 | $8.99 | +211% |
| Verda | $4.99 | $11.81 | +137% |
FAQ
H100 vs B300 questions
Is the B300 more expensive than the H100 to rent?
At the median public on-demand price, yes by 132%: $9.25 against $3.99 per GPU-hour.
Which has more memory, the H100 or the B300?
The H100 has 80 GB and the B300 has 288 GB per GPU, as published by the vendors and shown in provider listings.
Where is the cheapest H100 and B300?
The cheapest public on-demand H100 we track is $1.68 at Latitude.sh. For the B300 it is $7.40 at Hyperstack. Last checked Nov 9, 2026.
Need H100 or B300 capacity?
Tell us the GPU, count, term, region and budget. We reply with the public prices that match and the checks to run before you pay. Free, and we do not sell compute.
More