The Operating System
21 Markets Play

Price on the Value Delivered

Huang argues the cost of each unit of AI output is falling fast, which is how he justifies margins that went from 35% to 75%.

our token generation effectiveness is going up so much faster that token cost is coming down.

The argument on the Lex Fridman Podcast (#494, March 2026) is that the price tag on the hardware is the wrong number to look at. What matters is the cost per unit of output, and Huang says that is collapsing: “It’s just coming down an order of magnitude every year.” On his telling, an expensive system that produces far more per dollar is the cheap option.

His scale claim. “Moore’s Law would have progressed computing about 100 times in the last 10 years,” he said, while “we progressed and scaled up computing by a million times in the last 10 years.” That is his own figure, and it bundles algorithms, numerical precision and scale as well as chips, so treat it as a claim, not a measurement.

The margin arc. In the CUDA years he says Nvidia “was a 35% gross margin company.” GAAP gross margin was 71.1% for fiscal 2026 and 75.0% in Q2 FY2027, per Nvidia’s SEC results.

The limit. China shows where it stops: a $4.5B H20 charge in Q1 FY2026, and guidance that assumes no China data-center compute revenue.

Doesn’t transfer without scarcity. 75% margins on hardware reflect supply shortages, a 20-year software moat and few alternatives, conditions that also draw antitrust scrutiny. Any company can price on value; few can charge this much for it without being the bottleneck.