NVIDIA launches new products, will old GPUs become obsolete soon?

As NVIDIA launches new-generation products, expensive GPUs could quickly become obsolete. If that were the case, companies spending billions of dollars on GPUs would end up holding rapidly depreciating assets. However, the data shows just the opposite. According to Silicon Data estimates, the six-year-old A100 still retains about 25% of its original purchase value, while the H100 retains around 58% after approximately four years. Both outperform the levels implied by the traditional three-year or five-year straight-line depreciation methods by a significant margin.The B200's performance is even more impressive: after being widely available for about a year, its estimated value stands at 158% of its assumed original purchase price, meaning it has actually appreciated by 58% rather than depreciated. This strong performance reflects a limited supply and strong demand for high-end AI computing power.Even when a particular GPU is no longer the state-of-the-art choice for training frontier models, it can still hold significant value in inference, fine-tuning, research, and less demanding workloads.This is extremely beneficial for NVIDIA, as it shows that new-generation GPUs do not immediately destroy the economic value of older products; instead, the market appears large enough to absorb multiple generations of NVIDIA hardware simultaneously.This also benefits NBIS, CRWV, and other GPU cloud service providers. If these companies’ GPUs can retain more value and generate rental income over a longer period, they may achieve higher returns over the lifecycle and reduce the risk that the GPU fleet loses economic value before it recoups its cost.Higher residual values could also make it easier to finance GPU infrastructure, as lenders may be more willing to treat GPUs as revenue-generating collateral if the equipment maintains a considerable market value and offers predictable rental income.