Nvidia H100 Rental Prices Jump 22% as Older AI Hardware Retains Value

The H100 was widely used to train early generative AI models. Rental prices have recovered from previous lows but remain below the peak of $7–$8 per hour. Huang said Nvidia’s computing technology is durable and continues to generate revenue.

The rise in rental prices comes as newer Blackwell systems remain largely reserved for large customers keeping demand for older H100 clusters strong. However, critics note that prices are still well below historical highs, raising questions about how long the trend can continue.

Nvidia reported second-quarter revenue of $96.2 billion in August, highlighting the strong demand for its AI chips and computing systems.

Read More:Oracle stock surges 12.6% ahead of earnings on OpenAI deal buzz

The rebound in H100 rental prices differs from the way companies normally account for their GPUs. Hyperscalers typically depreciate GPUs over about five to six years. However, investor Michael Burry believes the actual useful life of these chips is shorter. He increased his short position in Nvidia in late August.

Another concern is who is actually paying for the demand. CoreWeave, a cloud company that buys Nvidia chips and rents them to customers, had $35 billion in debt at the end of June.

Nvidia has also agreed to rent unused capacity from some of its partners. Critics argue that such arrangements can make AI demand look stronger than it actually is because some deals are not fully reflected on Nvidia’s balance sheet.

Read More:Alchelyst Appoints Ben Middleton as Managing Director