Tokens-Per-Watt Economics

Tokens-per-watt economics is the view that AI infrastructure competitiveness depends on how many useful AI outputs a system can produce per unit of power, not only on raw chip speed or hardware count.

Key points

• Jensen says power is a concern for AI scaling, which is why NVIDIA pushes Extreme Co-Design to improve tokens per second per watt by orders of magnitude ^[src-065].
• He claims computing advanced far faster than Moore’s Law through system-level scaling and co-design, while token costs continue falling because token-generation effectiveness improves faster than hardware price rises ^[src-065].
• The metric links technical design to business model: energy efficiency affects the revenue and economics of an AI Factories operator ^[src-065].
• The concept helps explain why cooling, power delivery, rack design, supply chain, and networking are no longer secondary details in AI infrastructure strategy ^[src-065].

Related entities

• NVIDIA
• Jensen Huang
• NVIDIA Vera Rubin

Related concepts

• AI Factories
• Extreme Co-Design
• GPU Supply as AI Strategy
• LLM Inference Economics

Source references

• ^[src-065] Lex Fridman – “Jensen Huang: NVIDIA – The $4 Trillion Company & the AI Revolution | Lex Fridman Podcast #494” (2026-03-23)