The price tag of Nvidia's artificial intelligence (AI) accelerator H100, released four years ago, is instead climbing. As supplies of the latest AI accelerators fail to keep up with demand, rental demand is spilling over even to older products.
AI computing trading platform Ornn said on the 7th (local time) that the cloud lending price for the Nvidia H100, released in September 2022, rose 21.89% over the past month.
According to Ornn, the hourly lending fee for the H100 is $3.28, up $0.59 from $2.69 a month ago. That is about 72% of the hourly lending fee for the higher-end H200 at $4.56. Even compared with the next-generation B200 ($6.75), it is close to half.
Typically, information technology (IT) products, including semiconductors, lose value as new products are released and time passes. But the H100 is showing an unusual trend of prices rising in reverse even four years after its release.
Ornn said, "Under a typical depreciation model, the value should fall, but the market is paying an even higher premium."
Behind the renewed rise in the H100's price is a sharp increase in demand for AI computing. As the surge in demand becomes difficult to meet with supplies of the latest AI accelerators such as Blackwell and Rubin alone, analysts say tasks such as real-time inference, training of small and mid-sized AI models, and fine-tuning are being distributed to the H100.
The fact that the H100 is already widely deployed in data centers and that the related software ecosystem is in place is also cited as a factor supporting demand. Even if performance lags behind the latest chips, not every AI task needs the highest-spec accelerator, so demand continues to choose the H100 considering expense and availability.
Jensen Huang, Nvidia chief executive officer (CEO), mentioned the H100's price increase on X (formerly Twitter) and emphasized that "Nvidia's compute resources are fungible, durable, and in very high lending demand," calling them "productive assets that generate revenue."
However, the H100's price is not on a continuous upward trajectory. Ornn said the hourly lending fee has shown high volatility in the past 30 days, fluctuating from the mid-$2 range to the $3 range.
The H100's price trend also touches on the debate over depreciation of AI accelerators. Michael Burry, a short seller known as the real-life model for the film "The Big Short," argued last year that major big tech corporations are inflating profits by setting the depreciation period of AI chips longer than in reality.
On the other hand, a recent market view holds that as demand for AI computing grows rapidly, the economic value of Nvidia accelerators may last longer than expected. The rise in H100 rental fees is seen as a case showing that demand for previous-generation AI accelerators can continue for a considerable period even after the launch of the latest products.