Jensen Huang Says AI Could Need One Billion Times More Compute Than Last Year

In an interview released Oct. 15, Nvidia CEO Jensen Huang said AI could require one billion times more computing power than the level from roughly a year earlier, as reasoning workloads grow and AI is used for more tasks.

Huang described reasoning systems as spending compute to explore options, plan, reflect, check outcomes and prune weaker paths. “All of that thinking … takes a lot more compute,” he said.

He connected the projection to two kinds of growth: more computing inside each task and more tasks handled by increasingly capable AI. Huang compared potential needs with roughly last year’s level but did not specify a date for reaching it or define a compute unit.

The projection comes alongside infrastructure constraints. Huang identified land, power and data-center shells as obstacles to expansion in 2027. He also said supplies of substrate materials and silicon photonics were constrained.

At Nvidia’s March 2026 GTC conference, Huang separately said demand for Nvidia GPUs had risen one million times over “the last few years.” He also said he saw at least $1 trillion in Nvidia revenue from 2025 through 2027.

The figures use different measures and periods: the latest projection concerns potential AI compute needs compared with roughly a year earlier, while the GTC remark concerned GPU demand over several years. The revenue figure covered 2025 through 2027.