--- title: "The Era of Eternal Chips: CoreWeave Extends Lease of Legacy Nvidia A100s Until 2029" description: "🤖 Shock Content: CoreWeave will lease Nvidia A100 chips until 2029! By 2026, these accelerators are already 6 years old, and by the end of the lease term, they will be 9. Despite lagging behind modern models, the shortage of power and the \"zero cost basis\" of old chips make them a goldmine. 📉💰 #Nvidia #CoreWeave #AI #Technology #Crypto" date: 2026-08-16T20:57:38.000Z lang: en url: https://xab.info/en/posts/era-of-eternal-chips-coreweave-extends-lease-of-legacy-nvidia-a100s-until-2029 tags: [nvidia, coreweave, artificial-intelligence, gpu, data-centers, tech-news] publisher: "XAB.info" --- # The Era of Eternal Chips: CoreWeave Extends Lease of Legacy Nvidia A100s Until 2029 ![High-performance Nvidia A100 chip on a printed circuit board, symbolizing the extension of rental until 2029](https://xab.info/media/2026/08/17/coreweave-prodlila-arendu-nvidia-a100-do-2029-goda/coreweave-prodlila-arendu-nvidia-a100-do-2029-goda-1.webp) ## 🎯 Key Points - CoreWeave has committed to leasing Nvidia A100 accelerators until 2029. - By the end of the lease term, these chips will be 9 years old. - The A100 is technically obsolete but remains profitable due to full depreciation. - The shortage of computing power in the AI sector forces the use of legacy equipment. In a world where artificial intelligence technologies evolve at a dizzying pace, a paradoxical shift has occurred that challenges conventional equipment lifecycle cycles. In August 2026, it was revealed that CoreWeave has committed to leasing Nvidia A100 accelerators until 2029. This decision means that six-year-old chips, originally released in 2020, will remain in service for at least three more years, reaching the age of nine by the end of the term. ### The Phenomenon of "Eternal" Leasing in the AI Era The monstrous hype surrounding AI has led to a situation where the shortage of computing power outweighs the demand for its currency. CoreWeave, one of the largest cloud providers specializing in AI, confirmed that it will continue to use the Ampere architecture. For a market dominated by the latest solutions with kilowatt-level performance, this looks like a step backward. However, given that demand for GPUs continues to exceed supply even in 2026, the availability of any working hardware becomes a priority for startups and researchers. ### Technical Specifications vs. Economic Viability Recall that the Nvidia A100 in its top-tier version offers 80 GB of HBM2e memory. Its performance in INT8 and FP16 modes is 624 TOPS and 312 TFLOPS, respectively. The accelerator's power consumption ranges from 300 to 400 watts. This represents a colossal technical lag behind current Nvidia products, which consume several kilowatts and offer model training speeds orders of magnitude higher. Nevertheless, for inference (output) tasks and less demanding training stages, the A100 remains a sought-after tool, especially given its availability. ### The Economics of "Zero Cost Basis" The key factor allowing CoreWeave to keep these chips in the rental pool is their financial model. Many of these accelerators were purchased during the 2021-2022 boom and are already fully depreciated. This creates a "zero-cost-basis engine" effect. Renting such chips at market price ("full freight") generates pure profit, as the capital investment in them has already been made. In a climate where new chips cost a fortune, renting "used" equipment becomes a profitable strategy for both sides of the market. ### Contradictory Data There is a divergence in assessments regarding the prospects of this move. On one hand, analysts note that CoreWeave proves the ability to generate profit from 2020-era chips even nine years after their launch. This testifies to the incredible margin of the business. On the other hand, experts warn that reliance on legacy equipment could slow down innovation. There is a risk that extending lease terms to 2029 and beyond could create an illusion that infrastructure updates are unnecessary, which in the long term could reduce the efficiency of global AI projects requiring maximum speed. ### The Future of AI Infrastructure The story of CoreWeave and the Nvidia A100 demonstrates that the AI market has not yet reached a saturation point. Even in 2026, when technology has advanced significantly, the demand for "hardware" remains so high that even nine-year-old accelerators become a strategic asset. It is likely we will see further extensions of the operational life of this equipment, forcing the industry to reconsider approaches to data center disposal and upgrades. ## 🔍 Fact-Check Verification - [CoreWeave CEO: We’re Booking 2020-Era NVIDIA GPUs Through 2029 at “Full Freight”](https://247wallst.com/investing/2026/08/12/coreweave-ceo-were-booking-2020-era-nvidia-gpus-through-2029-at-full-freight/) - CEO CoreWeave подтвердил аренду чипов 2020 года до 2029 года. - [I think you missed CoreWeave's zero-cost-basis engine](https://www.msn.com/en-us/money/general/i-think-you-missed-coreweave-s-zero-cost-basis-engine/ar-AA2a7sGw) - Подтверждение прибыльности использования чипов через 9 лет после развертывания. - [CoreWeave proves Nvidia's aging AI GPUs from 2020 can generate profit nine years after deployment](https://www.msn.com/en-gb/money/other/coreweave-proves-nvidia-s-aging-ai-gpus-from-2020-can-generate-profit-nine-years-after-deployment/ar-AA29XP3H) - Подтверждение прибыльности использования чипов через 9 лет после развертывания. - [CoreWeave: 'Buy' The Dip](https://seekingalpha.com/article/4923363-coreweave-buy-the-dip) - Подтверждено по источнику seekingalpha.com ## ❓ FAQ ### Q: Why does CoreWeave continue to use chips from 2020? **A:** Due to the shortage of computing power and the fact that these chips are already fully depreciated, allowing for pure profit from their rental. ### Q: How old will Nvidia A100 chips be by 2029? **A:** By 2029, these accelerators will be 9 years old. ### Q: What is the performance of the Nvidia A100? **A:** In its top-tier version, the A100 offers 80 GB of HBM2e memory, 624 TOPS in INT8 mode, and 312 TFLOPS in FP16 mode.