英伟达首席执行官黄仁勋在10月22日于柏林举行的GTC大会上表示，公司计划对运行企业AI代理的软件，按每颗GPU每年约2000美元收费。
Nvidia Explores $2,000 Annual Per-GPU Software Licensing Model for AI Infrastructure At GTC Berlin on October 22, 2026, Nvidia CEO Jensen Huang indicated that the company plans to charge roughly $2,000 per GPU annually for software running enterprise AI agents.
若应用于500万颗GPU，该授权模式每年可带来约100亿美元收入，五年累计高达500亿美元。
If applied to five million GPUs, the licensing model could generate an estimated $10 billion annually and up to $50 billion over five years.
英伟达尚未公布生效日期，也未详细说明具体哪些软件产品将纳入这一层级。拟定费率低于现有英伟达AI企业版每年4500美元的订阅价。
Nvidia has not yet announced an effective start date or detailed exactly which software products will be included under this tier. The proposed rate sits below the existing $4,500 annual subscription for Nvidia AI Enterprise.
这一举措符合英伟达进军企业级基础设施的更广泛战略，近期宣布的OpenShell参考架构即获Anthropic和Salesforce支持。
The initiative aligns with Nvidia's broader push into enterprise infrastructure, highlighted by its recently announced OpenShell Reference Architecture supported by Anthropic and Salesforce.