As of October 8th, the B200 was trading at $5.88 per GPU-hour, continuing its approach to a two-month high; the H100 slightly declined to $2.80, but remained near the upper limit of its two-year fluctuation range; the A100 rose slightly to $1.53, temporarily recovering from its July low. The B200's continued high level, the H100's previous upward trend not yet completely broken, and the older A100's continued weakness indicate that current AI computing power demand is not experiencing a comprehensive surge, but rather is further concentrating on higher-performance high-end GPUs. The H100's sustained high level, in particular, suggests that demand for the previous generation of mainstream high-end GPUs, in addition to the latest generation B200, remains resilient.
-------- 1. The GPU rental price index primarily tracks the actual rental quotes for NVIDIA A100, H100, and B200 GPUs from cloud service providers, data centers, and the private rental market each day.
2. A100, H100, and B200 are NVIDIA's three generations of GPU chips, from oldest to newest. Generally, newer GPUs offer better performance and are better suited for large-scale, complex models, and their rental prices are typically higher.
3. Due to the high price and rapid updates of GPUs, many companies choose to lease chips from cloud service providers and data centers to save on investment costs. A rise in related indices usually indicates a tighter GPU supply and stronger demand; a decline suggests increased supply or cooling demand. By observing price trends and comparing the price difference between new and old GPUs, investors can understand the current state of AI computing power.
