Recently, Huang referred to Musk's efforts as 'superhuman' for his team's accomplishment in assembling and deploying NVIDIA equipment in just 19 days. Musk is not stopping there and plans to increase the size of the Colossus supercomputer from 100,000 to 200,000 NVIDIA H100 graphics processing units, used for training large language models from xAI.
The swift deployment played a crucial role in the success, as xAI and NVIDIA completed Colossus in just 122 days, which would have taken months or even years otherwise. NVIDIA emphasized that the Colossus system provides data throughput at 95% without delays or packet loss due to the Spectrum-X congestion control system.
Morgan Stanley forecasts that NVIDIA will sell approximately 1.5 million Hopper graphics processing units in the fourth quarter of 2024, followed by a decrease to 1 million in the first quarter of 2025.