Nvidia A100 Still in Demand as CoreWeave Extends Leases to 2029; Huang Cites CUDA Longevity

Nvidia's A100 accelerator, released in 2020, is still in short supply as AI compute demand continues to surge. The latest lease agreement signed by cloud computing provider CoreWeave shows that A100 units have been renewed by customers through 2029, extending the chip's lifecycle to nearly a decade. Based on the Ampere architecture, the A100 has been widely used for AI training and inference tasks since its launch, and its popularity has not waned despite subsequent product iterations.

Nvidia CEO Jensen Huang responded that the CUDA ecosystem offers developers and engineers a unified platform, enabling architectures such as Ampere, Hopper, and Blackwell to receive continuous upgrades throughout their lifecycles. This versatility makes Nvidia's computing resources highly substitutable and efficient, thereby extending hardware durability and turning them into assets that can be leased, financed, and produced steadily over the long term.

Public information shows that the A100 features 6,912 CUDA cores and up to 80GB of HBM2E high-bandwidth memory. Its core scale and computational efficiency fall short of later Hopper, Blackwell, and Rubin architectures, but with global demand for large-model training and inference rising and memory resources generally scarce, the single-card 80GB HBM2E capacity still meets some workloads. The rental cost for related products is relatively low, which sustains its application value in the AI infrastructure sector.