Nvidia AI bank to ensure minimum residual value of GPUs for lenders
NVIDIA
Nvidia will use its own funds to guarantee a minimum residual value of its GPUs, which can be counted as collateral in AI infrastructure projects funded by a consortium of institutional investors. The company aims to support longer depreciation periods and reduce its own debt exposure.
According to the Financial Times, Nvidia is establishing a consortium of institutional investors to fund AI projects up to $500 billion. Traditionally, lenders provide funds for three or five years based on the near-full depreciation of collateral assets. For AI infrastructure, this includes data center buildings and compute accelerators, but residual value of GPUs is problematic due to rapid obsolescence. Nvidia now guarantees a 25% residual value of its accelerators for the duration of loans to clients. If projects fail, Nvidia will incur losses of at least 25% of the cost of supplied GPUs. The company argues that older GPUs like H100 and A100 remain in demand, and it regularly updates CUDA software to support them. The consortium will help Nvidia expand access to capital, as partners will resell debt obligations and distribute investment risks, moving debt off Nvidia's balance sheet.
- Abbreviations
- GPU = Graphics Processing Unit — графический процессор
- CUDA = Compute Unified Device Architecture — вычислительная архитектура NVIDIA
Source: 3DNews —
original