Beyond Tokens: Why Intelligence Per Watt Is The Metric That Matters

Forbes
Read full story· 5 min
A Stanford study highlights that intelligence efficiency—AI capability delivered per watt—improved 5.3 times from 2023 to 2025, driven by better models and hardware. This metric better reflects AI value than token throughput, influencing industry focus amid rising AI costs. Companies like Uber are limiting AI spending, emphasizing value over token volume.

What happened

A Stanford research team published a study in November 2025 showing that the efficiency of AI systems, measured as intelligence delivered per watt, increased by 5.3 times from 2023 to 2025. This improvement came from advances in both AI models and hardware. Companies such as Uber have responded to rising AI costs by capping their AI budgets and focusing on the value generated rather than the number of tokens processed. The industry is shifting attention from token throughput to intelligence efficiency as a key performance metric.

Key details

  • Stanford's study was published in November 2025.
  • Intelligence efficiency improved 5.3 times between 2023 and 2025.
  • Model improvements contributed a 3.1x gain; hardware improvements added 1.7x.
  • Uber announced in May 2026 that it had already spent its annual AI budget.
  • Server DRAM contract prices are expected to rise 13% to 18% in Q3 2026.

Why it matters

CFOs, CIOs, and CTOs at companies using AI now face higher costs and must justify spending based on measurable value, not just activity. Hardware and memory providers are pressured to deliver better performance per watt and per dollar, as energy and component costs rise. The shift to intelligence efficiency as a metric changes how organizations allocate budgets and design infrastructure, affecting purchasing decisions and operational strategies.


More in Machine Learning