News from the IT community on October 26th: Jensen Huang, CEO of NVIDIA ( Jensen Huang ), stated that the annual licensing fee for the software that supports AI is approximately $2,000 per GPU (Note from IT community: The current exchange rate is about 13,355 RMB). This fee is used to "operate the full set of software provided by NVIDIA that supports the operation of AI intelligents in these AI factories." A video of his related remarks has recently been leaked. According to a calculation scenario provided by FourWeekMBA, if 5 million licenses are sold, this pricing model is expected to generate revenue of 50 billion US dollars (about 333.872 billion RMB at the current exchange rate) within five years.
Jensen Huang discussed this pricing at the GTC conference in Berlin on October 22nd, and NVIDIA also announced more ambitious plans at the conference: to build a AI infrastructure business that covers chips, software, and AI intelligent entity security capabilities. "Software licensing will gradually become a very important part of our business. Our licensing model is roughly as follows... For example, to run NVIDIA's full set of software for deploying intelligent entities in AI factories, the cost is approximately $2,000 per chip per year (about 13,355 yuan at current exchange rates)," said Jensen Huang.
The volume of this revenue is easy to estimate. If 5 million GPU are included in the paid licensing scope, and assuming each one is valued at $2,000 per year (which is approximately 13,355 RMB at current exchange rates), NVIDIA could generate $10 billion in revenue annually (about 66.774 billion RMB at current rates). As long as customers continue to renew their subscriptions, the total over five years would amount to $5 billion (approximately 33.3872 billion RMB). However, there are uncertainties here: NVIDIA has not specified the scope of application for the $2,000 per unit rate, nor has it confirmed whether this is a uniform standard pricing. It is also unclear which software this license includes and whether there will be any overlap in permissions between different products.
It is crucial to distinguish between these two points, as NVIDIA already has enterprise-level software products that are subject to fees. The annual subscription price for its AI Enterprise platform is $4,500 per unit (which is approximately 30,049 RMB at current exchange rates). Although the $2,000 mentioned by Jensen Huang seems lower, NVIDIA has not confirmed that the products corresponding to these two prices have identical licensing terms, so it cannot be directly considered an official price reduction.
The background for this discussion on software licensing is that NVIDIA is developing more supporting infrastructure to facilitate enterprises in deploying and managing AI intelligents. Such systems can invoke tools, read data, and automatically execute tasks. When enterprises deploy them on a large scale, new security and management challenges arise.
On October 21st, NVIDIA released the OpenShell reference architecture, which integrates Vera Rubin computing infrastructure, Vera CPU, as well as BlueField-4 data processors (DPU) equipped with NVIDIA Sentry security components. OpenShell provides a secure operating environment for AI agents, defining their boundaries of action and restricting their permissions to call tools and access resources. NVIDIA explains that Sentry can independently monitor agent behavior, and if an agent attempts to overstep these boundaries, it will be isolated. This architecture brings security controls down to the underlying infrastructure, rather than relying solely on the protection mechanisms within AI models or applications.
NVIDIA launched an open agent security platform as early as September. This new reference architecture is an extension of the previous one, extending the control concepts to enterprise infrastructure and edge devices.
Other manufacturers are also incorporating similar control capabilities into their products. Anthropic combines OpenShell and BlueField hardware with Claude managed agents to facilitate enterprises' control over the content that AI agents can access within the sandbox. Salesforce has developed a Slack management bridging tool that is compatible with OpenShell; administrators can directly review agent access requests, approve or deny permissions within Slack, while also monitoring operational dynamics.
The pricing of this new software also reflects Jensen Huang's consistent viewpoint: NVIDIA is no longer just a GPU supplier. The company's business has expanded to include network devices, complete systems, enterprise software, and infrastructure for AI loads. Recently, Jensen Huang mentioned in a BG2Pod podcast program, "One day, people might no longer categorize us as a chip company." He gave the example that NVIDIA has entered multiple industries such as healthcare, automotive, communications, and life sciences, providing a complete technology stack, not just chips.
Even so, NVIDIA has not announced when the fee scheme of $2,000 per GPU per year will take effect, nor is it clear which software is included in this scheme, nor how this pricing will be integrated with their existing enterprise software business.
