Intel / video

Brad Gerstner and Sunny Madra on the Economics of AI Inference

Source: youtube.com
Author: Stanford Online
Date: Indexed Aug 16, 2026

Context

Cheaper text generation can coexist with a larger AI bill. In this discussion of inference economics, AI investor Brad Gerstner and Nvidia's Sunny Madra argue that longer reasoning and continuously working agents increase the amount of computation users consume. A falling price per generated unit therefore does not settle the question of total spending.

Madra describes hardware progress as hard to sustain on chips alone: manufacturing limits, power and ever-larger models mean that even large efficiency gains can be outpaced by demand. Their argument is that efficiency gains and growing demand can happen together. For evaluating an AI workflow, this points toward considering how much work the system performs and consumes, alongside the price of individual .

Terms in this piece · Glossary

inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.

Key quotes

“You got to make yourself bionic with AI”
Brad Gerstner

“It's going to a billionx.”
Jensen Huang

“IQ gets commoditized and EQ becomes super valuable.”
Brad Gerstner

Read the source youtube.com