Nvidia CEO outlines US$2,000 annual software cost per GPU
The company has not explained how the figure relates to existing AI Enterprise prices or when the pricing might take effect.
Nvidia CEO Jensen Huang put the annual cost of its AI-agent software at about US$2,000 per GPU during an October 22 question-and-answer session at GTC Berlin. He said software licensing would account for an increasingly important share of Nvidia’s business.
The Register first reported Huang’s remarks. His pricing example covered Nvidia software used to operate agents in AI-focused data centers, which the company calls AI factories. Nvidia has not specified when that pricing might become available or explained its relationship to the company’s existing software licenses.
Nvidia’s AI Enterprise package, which businesses use to develop and operate AI on its hardware, currently lists an annual price of US$4,500 per GPU, including support. A five-year commitment costs US$18,000 per GPU, equivalent to US$3,600 annually. Both rates exceed the figure Huang cited.
Nvidia has also introduced the OpenShell Reference Architecture, extending a safety platform unveiled in September. It places agents on Vera CPUs within OpenShell, an open-source sandbox that restricts access to files, networks and tools. BlueField-4 networking chips use Nvidia’s Sentry software to monitor agents externally and can quarantine them within milliseconds if they breach those restrictions.
