The rise of AI agents could significantly increase computing demand as companies move from occasional AI interactions to systems that operate continuously in the background.
NVIDIA CEO Jensen Huang said the company is seeing a transition toward increasingly agentic AI, where software agents can reason, use tools and perform tasks with limited human intervention.
According to Huang, AI agents already require substantially more computing power than conventional human-driven AI interactions.
He estimated that the computing requirement for an agent can be 15 to 100 times that of a human using AI, depending on the problem.
Huang also expects the number of AI agents used by companies to eventually dwarf their human workforces.
“In the future, every company will have a whole bunch of agents.”
He described a future in which these agents operate continuously rather than only when a human submits a prompt.
That could have major implications for data-centre demand.
Today’s AI workloads are already driving massive investments in computing infrastructure. But continuously operating agents could add another layer of demand because inference would increasingly happen throughout the day rather than only during user interactions.
NVIDIA believes this trend is already translating into profitable AI workloads. Huang said the key development for the industry is that AI is now doing useful work, generating profitable tokens and creating additional demand for computing capacity.
This also explains NVIDIA’s focus on making computing infrastructure more flexible. Its Vera Rubin platform is designed to support different stages of the AI lifecycle, including data preparation, training, post-training and agentic inference.
The result could be a new phase in data-centre expansion, where the growth of AI infrastructure is driven not only by larger models but by the sheer number of autonomous systems operating continuously.
