Nvidia is discussing a $2000 per-GPU annual fee for software, which could significantly increase operational costs for production environments running large-scale LLMs and agents on Nvidia hardware, potentially forcing a reevaluation of infrastructure expenses and vendor lock-in strategies.