New GPU Servers: NVIDIA B300

NVIDIA B300 GPUs are now available on GPU Servers! Built on NVIDIA's Blackwell Ultra architecture, B300 is the most powerful GPU in our lineup. It builds on B200 with more GPU memory and faster AI performance, so you can run larger models and serve more requests from a single server.

With B300, you can:

• Run larger models with longer context windows on fewer GPUs
• Train and fine-tune foundation models across multiple GPUs
• Serve inference at scale to many users at once

Each B300 GPU comes with 288 GB of HBM3e memory, and multi-GPU plans are connected over fifth-generation NVLink with 1.8 TB/s of GPU-to-GPU bandwidth.

As with all GPU Servers, you only pay while the server is powered on. See the GPU Server configurations page for the full list of plans and specifications.