It's referenced as a baseline for what frontier-scale open-weight models can run on — for example, Cohere cited a single Nvidia Blackwell B200, or two H100s, as sufficient hardware for one of its Apache 2.0-licensed model releases.
Blackwell is Nvidia's GPU architecture generation used for large-scale AI training and inference, including chips like the B200.
It's referenced as a baseline for what frontier-scale open-weight models can run on — for example, Cohere cited a single Nvidia Blackwell B200, or two H100s, as sufficient hardware for one of its Apache 2.0-licensed model releases.