Huang Challenges Conventional Wisdom on AI Processor Obsolescence

Nvidia CEO Jensen Huang is challenging the conventional wisdom that costly AI processors quickly lose economic value as newer chips arrive. He argues that older GPUs remain valuable revenue-producing assets, citing data showing that three-year-old H100 rental prices rose 22% last month to $3.28 an hour.

Huang's claim matters for cloud providers and companies spending billions on Nvidia systems because it suggests that customers can recover their investments over longer periods, improving lifetime returns and supporting continued purchases of newer products. This also strengthens Nvidia's ecosystem by creating a tiered computing market where older chips serve less demanding workloads while premium models migrate to newer architectures.

Huang highlighted how quickly top-end demand is scaling, noting that OpenAI's GPT-6 Astra was trained on more than 100,000 Nvidia Grace Blackwell NVLink72 chips and adding '400K GPUs coming online next.' He also declared that 'AGI has arrived,' underscoring his confidence that increasingly capable models will require vastly more compute.