NVIDIA H200 141GB Data Center GPU

The NVIDIA H200 pushes memory capacity to 141GB of HBM3e, nearly double the H100, making it the top choice for enterprises training and serving the largest current-generation AI models. Demand for this GPU consistently outpaces supply across the data center market. The H200 is built on the same Hopper architecture as the H100 but upgrades to HBM3e memory, nearly doubling capacity to 141GB per GPU — a meaningful difference for the largest current language models that no longer comfortably fit in 80GB of memory. Its memory bandwidth of 4.8TB/s is roughly 40% higher than the H100, directly speeding up inference for memory-bandwidth-bound workloads rather than purely compute-bound ones. NVIDIA designed the H200 as a drop-in upgrade for existing H100 server infrastructure, meaning organizations that already built out H100-compatible chassis and cooling can often upgrade without a full infrastructure redesign. As the newest flagship data center GPU, expect the longest lead times and highest demand of any product in this catalog — organizations with time-sensitive AI infrastructure plans should place orders well ahead of their target deployment date.