GPU Platforms
The GPUs We Source, Today and Next-Gen
Backplane isn't just a site broker — we put buyers directly on GPU capacity as a service, or match and finance a dedicated deployment on a powered site. Here's what we're sourcing right now, and what's coming next.
Available Platforms
Three platforms, one conversation.
Current-generation Blackwell Ultra capacity available now, next-generation Vera Rubin by early access.
NVIDIA B300
The B300 is NVIDIA's Blackwell Ultra GPU built for teams that need Blackwell-class training and inference performance without the liquid-cooling infrastructure a full rack-scale system demands. It's the platform to reach for when air-cooled data halls are what you have, or what you can stand up fastest.
NVIDIA GB300
GB300 pairs the Grace CPU with the Blackwell Ultra GPU in NVIDIA's NVL72 rack-scale architecture — 72 GPUs and 36 Grace CPUs acting as one NVLink-coherent domain per rack. It's the platform for teams pushing the largest training runs and highest-throughput inference, where rack-scale memory and bandwidth are the constraint.
NVIDIA Vera Rubin
Vera Rubin is NVIDIA's next-generation platform after Blackwell — the Vera CPU paired with the Rubin GPU. We're tracking it now so buyers who join early aren't starting from zero once capacity opens up.