BERLIN – Nvidia is set to buttress its already ballooning revenue streams with proprietary enterprise software licensing potentially amounting to around $2,000 per GPU per year to run all of its agent software inside AI factories.
That figure was stated by Jensen Huang during a press conference at GTC Berlin last week, during which the CEO said software licensing will “become increasingly a major part of our company's business.”
The event in Germany saw Nvidia unveil the latest arrow in its burgeoning quiver of agent-focused offerings in the form of a reference architecture based on OpenShell, the open-source runtime software. While a lot of Nvidia’s software is designed to be open, Huang told press outlets, including this title, that proprietary parts of its software stack will be available to enterprises via licensing.
The CEO described Nvidia’s software business as “already quite significant,” placing it at “well over a billion dollars a year.” But a number like $2,000 per GPU could quickly add up for sites deploying tens of thousands of chips, something Nvidia is anticipating.
“Now that AI is going into the world's enterprise, it's going to accelerate tremendously, and so this, our software business, will be quite large, just simply because the amount of software necessary to run these systems is quite significant.”
Huang doesn’t just have his eye set on the big GPU clouds, either, telling press that if Nvidia wants to deploy AI factories outside the large clouds, then “we're going to have to continue to support all the software.”
The Nvidia CEO used his platform at GTC to rally against the GPU company label it's been synonymous with for years now, instead proclaiming during his one and a half hour keynote that the company was a “full stack, end-to-end, complete AI factory platform.”
Referencing the growing importance of software to its offerings beyond its existing golden child in CUDA, Huang said that “nothing works” inside AI factories “unless you have software that activates it all.”
The company’s grand plans to push further into software licensing were hinted at prior to the Berlin show, with the Nvidia CEO telling BEP Research that the business model “will be accelerated in the near future.”
“For the CSPs [cloud service providers], they want to run their own stack, and so we license, or we give them our software open source, and then they take it, and they modify it, and they integrate into theirs, and then they operate their services. But for a lot of the world's companies, they don't have the IT department. They don't have the software department to do it, so they need me to maintain it.”
Huang contended that the shift away from training to inference and actual AI deployment means wider enterprises need to operate their infrastructure as factories, adding: “They need to know that I'm on the hook, I'm on the SLA (service level agreement) to support it and maintain it, keep it secure, keep it safe for as long as we shall live. And so software license for an enterprise business will be very, very large.”
It's important to stress that Huang's comments during the press briefing on a proposed fee have not been formally confirmed by Nvidia. Also unclear at the time of writing is how such a software license works in relation to overlapping between product lines. But given the CEO's eagerness to talk about it in the build up to as well as during GTC Berlin, we may find out soon.
Comments