How to Power AI's Massive Energy Needs

Key Concepts

• AI Data Query Energy Consumption: The significant energy required for each individual AI data query.
• Global AI Query Volume: The immense scale of AI data queries being made worldwide per second and per hour.
• Data Center Energy Usage: The substantial power consumption of AI data centers, particularly large-scale ones.
• Gigawatt Data Centers: The newest and largest data centers, characterized by their immense power demands.

AI Data Query Energy Consumption

A significant data point has emerged regarding the energy consumption of AI. It has been reported by several sources that a single AI data query can consume 34 watt-hours (Wh) per query. This figure, while specific, is difficult to fully grasp in isolation.

Global AI Query Volume and Bundled Energy Consumption

To better understand the impact of 34 Wh per query, it's helpful to consider the sheer volume of AI data queries globally.

• There are over 5 billion smartphones in use.
• There are 21 billion connected devices worldwide.
• Globally, people are making 278,000 AI data queries per second.
• This translates to over 1 billion AI data queries every hour.

When this massive query volume is bundled with the 34 Wh per query figure, it represents a massive amount of energy consumption.

Data Center Energy Usage and Gigawatt Data Centers

The primary driver of this energy consumption is the operation of AI data centers, which are tasked with processing these queries. The newest and largest data centers, referred to as gigawatt data centers, are particularly noteworthy for their power demands.

• A single gigawatt data center can consume enough power to power a million homes for a year.
• Alternatively, it can power both the cities of San Francisco and Seattle simultaneously.

This highlights the immense scale of energy required to support the current and growing demands of AI.