CoreWeave Extends Nvidia A100 GPU Deployment Until 2029: Legacy Chips Remain Profitable

U.S.-based AI cloud provider CoreWeave, specializing in AI infrastructure, has signed a contract to deploy Nvidia A100 GPUs until 2029. These chips, launched in 2020 on the Ampere architecture, will continue generating revenue nearly a decade after their debut. The company reported record financial results: Q2 revenue reached $2.58 billion (up 112% year-over-year), with an order backlog exceeding $104 billion.

CoreWeave CEO Mike Intrator noted that rental costs for legacy GPUs remain stable or even higher than in previous years. This contradicts analyst concerns, including those from Michael Berry, about rapid chip obsolescence. Nvidia also confirmed that A100 GPUs maintain full utilization even six years after launch, with new contracts extending their lifecycle to nine years.

The primary driver behind the demand for older GPUs is data center infrastructure limitations. Modern solutions like Nvidia GB200 and GB300 require up to 140 kW of power per rack and liquid cooling, while existing facilities support only 20 kW with air cooling. Retrofitting such sites is economically unfeasible, leading providers to continue using proven solutions.

CoreWeave already holds contracts for 4.2 GW of capacity—nearly triple its current operational capacity. CFO Nitin Agrawal emphasized that the share of legacy GPUs in its portfolio is minimal, yet their value remains consistently high. This underscores sustained demand for AI compute power despite the emergence of new architectures.