By Judith Murphy
Publication Date: 2026-10-26 05:58:00
Jensen Huang let slip a number that could reshape how Nvidia makes money: roughly $2,000 a year for every GPU running its AI agent software, on top of whatever the chip itself costs.
Nvidia CEO Jensen Huang said at GTC Berlin on October 22 that running the full stack of Nvidia software behind AI agents in “AI factories” costs about $2,000 per GPU per year. He framed it as a rough estimate during a Q&A session, not a price list. But the number travelled fast. Within days it was showing up in headlines from TechPowerUp to Tom’s Hardware. Commentators did the obvious napkin maths: a million GPUs licensed at that rate is $2 billion a year, and Nvidia is shipping GPUs by the million.
Huang’s framing matters here. “Software licensing will become increasingly a major part of our company’s business,” he said. That’s according to a post from the account Wall St Engine, which circulated widely on X and was picked up by multiple outlets. It’s a notable shift in emphasis from a company that has spent the last three years selling almost entirely on hardware scarcity. Nvidia doesn’t need a new revenue stream to hit its numbers this quarter. It’s signalling where the moat goes once chip supply stops being the bottleneck.
Here’s the problem: Nvidia already sells software at more than double that price. Tom’s Hardware reported that Nvidia’s existing NVIDIA AI Enterprise platform lists at $4,500 per GPU for a one-year subscription, or $18,000 for five years. That’s the suite that…
