NVIDIA CEOJensen Huang revealed at the recent GTC event that the company plans to charge software licensing fees based on the number of GPUs, initially around $2,000 per card per year.
During the event, Jensen Huang said, "Software licensing will increasingly become an important part of our business. Our licensing approach is roughly $2,000 per GPU per year, covering all software provided by NVIDIA for running Agents in AI factories."
If calculated based on 5 million GPUs covered by paid licenses, it could generate $10 billion in annual revenue, totaling $50 billion over five years, assuming customers continue to renew their subscriptions.
However, there are still many unanswered questions about this calculation. NVIDIA has not disclosed how many GPUs will be subject to the $2,000 fee, nor confirmed whether this is the standard pricing.
The background of the software licensing discussion is that NVIDIA is building a complete infrastructure for deploying and managing AI Agents. On October 21, NVIDIA released the OpenShell reference architecture, integrating Vera Rubin computing infrastructure, Vera CPU, and BlueField-4 DPU, running NVIDIA Sentry security software.
OpenShell provides a secure runtime environment for AI Agents, defining boundaries for their behavior and tool access. Sentry independently monitors Agent activities and can isolate agents that cross these boundaries. This architecture embeds security controls into the underlying infrastructure, rather than relying solely on security mechanisms at the AI model or application layer.
Jensen Huang recently stated, "One day, people may no longer classify us as a chip company." He pointed out that NVIDIA has entered fields such as healthcare, automotive, telecommunications, and life sciences, providing a complete technology stack rather than just chips.
ButThis subscription model has also raised concerns among consumers. The shift from one-time purchases to recurring payments is a common path taken by many tech giants, with Adobe's transition from perpetual licenses to subscriptions being a prime example.
Enterprise customers are more accepting of software payments because GPUs are just one part of AI infrastructure. Deploying AI Agents also involves runtime environments, security isolation, permission controls, and management tools. NVIDIA's bundled offering saves significant integration effort.
However, gamers don't need to worry too much. Enterprise customers value stability and efficiency and are already deeply integrated into the NVIDIA ecosystem. If gamers have to pay extra for software features after buying graphics cards, NVIDIA is bound to face severe backlash. Besides, no matter how powerful NVIDIA GPUs are, AMD graphics cards are still a viable option.