Jensen Huang says AI agents could drive a 5-10x computing boom: “100 billion agents and billions of robots”
The New Stack Amanda Caswell ● Covered by 4 sources
Nvidia's Jensen Huang says chip demand could grow 5-10x this decade as AI agents and robots start outnumbering human computer users. That's the logic behind Nvidia's new $500 billion infrastructure push with South Korea's SK Group.
Based on reporting by The New Stack, Amanda Caswell — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Jensen Huang did not mince words in his latest Bloomberg interview. The Nvidia CEO said the semiconductor industry needs to grow somewhere between five and ten times over the next decade, and the reason is not more people buying laptops. It is agents. Huang painted a picture where a billion human computer users gets dwarfed by 100 billion AI agents and billions of robots, all quietly chewing through compute around the clock. He was careful to frame this as his own estimate rather than gospel, but it is a number he has been circling for months, alongside his repeated claim that traditional coding is on its way out in favor of engineers who build agents instead.
The shift he is describing is not just philosophical. On Nvidia's fiscal Q1 2027 earnings call, Huang described the move from generative AI to what he calls agentic AI — systems that perceive, reason, plan, and act — as the next major industry phase. That has real consequences for backend infrastructure. When most traffic comes from autonomous background processes instead of a person tapping a screen, the old assumptions about rate limiting, session memory, and API gateway concurrency start to crack. Infrastructure teams are already running into this.
Huang laid out where he thinks the physical backbone for this will come from. Speaking at the AI Summit in San Francisco on July 24, he called South Korea's semiconductor and industrial base central to the coming buildout, going so far as to say the country is entering a golden age. Nvidia backed that talk with a partnership with SK Group valued at more than $500 billion, covering large-scale purchases of next-generation High-Bandwidth Memory from SK Hynix, joint development of HBM4 roadmaps built for agentic and physical AI workloads, and deployment of Nvidia supercomputers.
The details of the South Korean buildout are specific and sizable. SK Telecom plans a 2-gigawatt AI data center running on Nvidia's Vera Rubin architecture and SK Hynix's HBM4 memory, with the first facility targeted for 2027. Separately, Nvidia is putting $1 billion into Naver Corp, with Brookfield contributing up to $9 billion as infrastructure capital partner, to help take Naver's data center capacity from 55 megawatts to 200 megawatts by 2028.
None of this is happening in isolation. Nvidia has already disclosed $119 billion in supply-related commitments as it tries to lock down advanced packaging capacity, power, land, and memory years ahead of need. Huang's underlying argument is that the industry has to stop designing for a world where people are the main users of computers and start designing for a world where computers serve other computers. The real bottleneck, in his telling, will not be demand or ambition — it will be whether anyone can physically build the infrastructure fast enough, a constraint already shaping how companies like Nvidia and Palantir think about sovereign AI deployments.
My take — AI-written commentary, not fact-checked reporting
A CEO forecasting that his own industry needs to grow five to tenfold is not exactly a neutral source, and it is fair to treat the number as a sales pitch dressed up as a prophecy. But the South Korean deal is real money committed to real megawatts, not just a slide at a conference, and that is the part worth watching. If agents genuinely start outnumbering humans as computer users, the winners will be whoever locked in memory and power capacity earliest — which conveniently happens to be exactly what Nvidia and its partners are doing right now.
Read more about this at: The New Stack