AI CHIPS ARE 420,000X MORE ENERGY EFFICIENT THAN THEY WERE A DECADE AGO
And yet electricity demand from AI keeps rising. Here is why.
NVIDIA $NVDA chip energy cost per token:
Kepler (2014): 42,000 joules
Pascal (2016): 17,460 joules
Volta (2018): 1,200 joules
Ampere (2020): 150 joules
Hopper (2022): 10 joules
Blackwell (2024): 0.4 joules
Blackwell Ultra / Rubin (2026): 0.1-0.2 joules
That is a 420,000x improvement in 12 years.
So why is power demand still exploding?
The Jevons Paradox.
When the cost of inference drops, more people use AI. More use cases get unlocked. More tokens get generated. The efficiency gain gets consumed by volume growth.
Cheaper AI does not mean less energy demand. It means more AI.
Tema ETFs: AI chips are 420,000x more energy efficient than a decade ago, and yet electricity demand keeps rising. That’s the Jevons Paradox: efficiency lowers the cost of inference, accelerating AI adoption and driving even greater power demand.
Post highlights a 420,000x improvement in NVIDIA chip energy efficiency over 12 years (from 42,000 joules per token in 2014 to 0.1-0.2 joules projected for 2026), yet notes rising AI power demand due to Jevons Paradox—efficiency gains drive adoption and volume growth rather than reducing total energy consumption. Current NVDA price is $207.41 (-2.37%), with earnings approaching on Aug 25, 2026. The efficiency narrative supports long-term AI infrastructure demand, though the post frames this as industry-wide rather than a direct NVDA trading call.
Post emphasizes massive efficiency gains in NVIDIA chips as foundational to AI infrastructure growth and cites Jevons Paradox as a structural driver of continued demand. However, this is a macro industry observation rather than near-term price catalyst. Current price is $207.41 (-2.37%) with earnings in ~3 months. The bullish case rests on long-term AI adoption trends rather than specific catalysts, and the post does not address current valuation or recent price action. Moderate confidence due to lack of immediate catalysts or financial analysis.
事件 Earnings (EPS est 2.1227) · 2026-08-25
关键要点
NVIDIA chip efficiency improved 420,000x from Kepler (2014) to projected Blackwell Ultra/Rubin (2026)
Despite efficiency gains, AI electricity demand keeps rising
Jevons Paradox: lower inference costs unlock more use cases and drive volume growth
Efficiency does not reduce total energy demand—it accelerates AI adoption
NVDA currently down 2.37% at $207.41, earnings on Aug 25, 2026
