Nvidia and major cloud operators are signaling that older AI accelerators can remain economically valuable for years despite rapid chip releases. CEO Jensen Huang posted on X on August 13, 2026, that Nvidia’s A100 GPUs remain effective from 2020 through 2029, attributing their longevity to regular updates to CUDA, Nvidia’s proprietary parallel-computing platform. Alibaba CFO Toby Xu said an A100 purchased in 2020 and a V100 purchased in 2018 were still running at full capacity, while the company expects its AI assets to generate very positive and robust cash flow after a roughly three-year payback period. CoreWeave also has an A100 rental contract running through 2029 at pricing it considers attractive. The disclosures come as Nvidia ships its Blackwell series, has released the Hopper line including the H100 and H200, and prepares its Rubin architecture for the 2026 to 2027 window. Strong secondary-market pricing for A100 and H100 chips, supported particularly by inference workloads, suggests AI demand may be deep enough to sustain several hardware generations simultaneously, although Alibaba did not disclose what share of its workloads uses legacy GPUs.