Grok 4.5 Trained on Over 20,000 Nvidia H100 GPUs, Boosting Chip Demand
NVDA•NVDA•NVDA•NVDA•X’s Grok 4.5 model was trained on over 20,000 Nvidia H100 GPUs in one of the largest known AI training runs, showcasing unprecedented demand for Nvidia’s data-center accelerators. This massive deployment is expected to generate hundreds of millions in new revenue and reinforce Nvidia’s leadership in the AI hardware market.
X’s AI division assembled a cluster of over 20,000 Nvidia H100 GPUs running continuously for an estimated two weeks to train the Grok 4.5 large-language model, leveraging Nvidia’s SXM5 architecture for high-throughput tensor processing.
This training run likely accounted for over $600 million in GPU billings, underscoring the profitability and pricing power of Nvidia’s data-center segment and signaling further enterprise spending on high-end accelerators.
Finance
