$META'S PROMETHEUS AI SUPERCLUSTER NOW CONNECTS HUNDREDS OF THOUSANDS OF NVIDIA GPUS, WITH 1 MILLION PLANNED Meta has completed the first phase of its gigawatt-scale Prometheus supercluster in Ohio, which was used to train its latest Muse Wonder AI models. The system combines NVIDIA $NVDA GB200 and GB300 NVL72 platforms across multiple interconnected buildings, operating as a single large-scale training system. Meta says it has reduced training interruptions by 99% from their peak, following extensive engineering work with NVIDIA on hardware reliability, networking and firmware. The company reported more than 85% productive training time for Muse Wonder pretraining and over 90% for reinforcement learning. Meta plans to expand the cluster further with NVIDIA's next-generation Vera Rubin systems, continuing its buildout of infrastructure for frontier AI models. Source: Meta Engineering.

Jensen Huang on Nvidia's offering growing from roughly $18B to more than $40B per gigawatt: “We used to be a GPU company … $NVDA became the largest & fastest growing networking company” “We’re now going to be one of the largest CPU companies, surely the fastest growing one” “We’re also growing share because we offer more to the world’s capex”

$NVDA CEO Jensen Huang says classical computing, including fraud detection and recommendation systems, accounts for roughly a third of NVIDIA’s business and is growing 30 to 40% a year.