NVIDIA H200 and B200 Blackwell GPUs Reach Wide Cloud Availability

NVIDIA's Blackwell GPU generation — the H200 and B200 — is now available across AWS, Google Cloud, and Azure simultaneously. This matters because previously, cutting-edge GPU generations were scarce, with months-long waitlists and uneven availability. Now any company can provision Blackwell instances through standard cloud consoles. The numbers are compelling: training large models on Blackwell costs about 40% less than on the previous H100 generation, simply because the hardware does more work per watt and per dollar. For inference — running models rather than training them — clusters switching to B100 see roughly double the throughput, meaning they can serve twice as many requests for the same cost. For AI labs and enterprises building on the cloud, this accelerates timelines and reduces the capital intensity of AI work. It also signals that the NVIDIA hardware upgrade cycle is now fast enough to affect strategic planning: the hardware you pick today meaningfully affects your cost structure for the next two years.