NVIDIA
NVIDIA B300
Blackwell Ultra accelerator designed for reasoning-era AI workloads, with 288 GB of HBM3e per GPU in NVIDIA HGX B300 systems and higher inference performance than B200.
- Architecture
- Blackwell Ultra
- Memory
- 288 GB HBM3e
- Memory bandwidth
- Up to 8 TB/s
- Release
- Shipping in NVIDIA DGX B300 systems in 2026
Main use cases
AI reasoning, long-context inference, large-scale training
Cloud pricing
Full tracker →| Provider | Price / hour | Region | Availability | Updated |
|---|---|---|---|---|
| Nebius | USD 9.5 | — | Published on-demand USD rate per GPU-hour, effective 1 October 2026; taxes extra; capacity not checked | 7 Oct 2026, 06:04 |
| Runpod | USD 7.89 | — | Secure Cloud Pod; published on-demand rate | 29 Sept 2026, 07:24 |
| Verda | USD 7.5 | — | On-demand GPU instance; published rate | 29 Sept 2026, 07:28 |
New-generation cloud capacity remains limited and provider pricing changes quickly; compare current verified cloud rates.
NVIDIA B300 in the news
GPU news →No coverage yet.