How Much Electrical Power Does AI Require?

2025-03-26

In the race to develop increasingly sophisticated artificial intelligence systems, one crucial factor often goes undiscussed: the substantial electrical power needed to fuel these computational marvels. As AI capabilities grow exponentially, so too does their hunger for energy—creating both challenges and opportunities for technological advancement.

AI systems, sustainable electrical power – artistic concept. Image credit: Alius Noreika / AI

The Growing Energy Appetite of Artificial Intelligence

Artificial intelligence systems require significant computational resources, which translates directly into electrical power demand. According to recent research, global AI data center power requirements could reach an astonishing 68 gigawatts (GW) by 2027—nearly doubling the total global data center capacity from 2022 levels. By 2030, this figure might climb to 327 GW, approaching the power capacity of major industrialized nations.

To put this in perspective, 68 GW is close to California’s 2022 total power capacity of 86 GW. Individual AI training runs could demand up to 1 GW in a single location by 2028 and potentially 8 GW—equivalent to eight nuclear reactors—by 2030 if current training compute scaling trends continue.

Why AI’s Power Requirements Are Escalating

Several factors contribute to AI’s increasing energy consumption:

Training vs. Inference: The Two-Phase Power Profile

AI development consists of two primary phases: training and inference. The training phase, where models learn from vast datasets, is particularly power-intensive. Training a large language model like GPT-3 consumes approximately 1,300 megawatt-hours (MWh) of electricity—roughly equivalent to the annual power consumption of 130 American homes.

For context, streaming an hour of Netflix uses about 0.8 kilowatt-hours (kWh) of electricity. You would need to watch 1,625,000 hours of streaming content to consume the same amount of power required to train GPT-3 once.

The inference phase—when trained models generate outputs based on user prompts—consumes less power per operation but can accumulate significant energy usage at scale. Research conducted by Hugging Face and Carnegie Mellon University found that generating a single AI image can use nearly as much energy as charging a smartphone.

The Scaling Trajectory

AI models have grown dramatically in size and complexity over recent years. This scaling trend directly impacts power consumption, as larger models with more parameters require more computational resources to train and run.

Alex de Vries, a PhD candidate at VU Amsterdam who studies the energy implications of emerging technologies, projects that by 2027, the AI sector could consume between 85 to 134 terawatt-hours annually—comparable to the entire electricity demand of the Netherlands, or about half a percent of global electricity consumption.

The Real-World Impact on Infrastructure

The United States currently leads the world in data centers and AI computing capabilities. However, the exponential growth in power demand is creating significant infrastructure challenges:

Grid connection requests in key regions like Virginia now take four to seven years to process
Transmission line projects face complex multi-state permitting hurdles
Data centers struggle with local and state permits for backup generators and environmental assessments
Environmental commitments limit readily available power sources

These bottlenecks could potentially compel U.S. companies to relocate AI infrastructure abroad, potentially compromising America’s competitive advantage in the computing sector and AI while increasing intellectual property security risks.

Energy Efficiency: A Crucial Counterbalance

While demand for AI continues to grow, energy efficiency improvements offer some counterbalance to rising consumption. Historically, data center energy usage remained relatively stable between 2010 and 2018, accounting for approximately 1-2 percent of global consumption. During this period, hardware efficiency gains helped offset increased computing demands.

However, the AI paradigm introduces a new dynamic. The trend toward solving problems by simply deploying larger models and more data creates natural incentives to continuously add computational resources. When models or hardware become more efficient, developers often respond by building even larger systems—potentially negating efficiency gains.

As Microsoft’s CTO for cloud operations and innovations, Judy Priest, noted, the company is “investing in developing methodologies to quantify the energy use and carbon impact of AI while working on ways to make large systems more efficient, in both training and application.”

The Path Forward: Balancing Innovation and Energy Impact

AI’s electrical power requirements highlight important considerations for the technology’s continued development:

Research Priorities

More research is needed to address bottlenecks in U.S. data center expansion and identify solutions. Priorities include:

Modeling future power grid supply against data center demand
Researching efficiency improvements that could reduce AI power requirements
Studying compute scaling bottlenecks
Analyzing how environmental review processes affect infrastructure development
Evaluating emerging power sources for AI workloads, including small modular reactors and geothermal energy

Transparency in Energy Usage

Researchers like Sasha Luccioni from Hugging Face point out that as AI has become more commercially valuable, companies have grown increasingly secretive about their training regimes, including hardware specifications and duration. This lack of transparency makes it difficult to accurately assess energy consumption.

Luccioni suggests implementing energy star ratings for AI models, allowing consumers to compare energy efficiency much as they do with household appliances.

Strategic Implementation

Perhaps most importantly, organizations must carefully evaluate where AI adds genuine value. As de Vries notes, “If we’re going to be using AI, is it going to help? Can we do it in a responsible way? Do we really need to be using this technology in the first place? What is it that an end user wants and needs, and how do we best help them? If AI is part of that solution, okay, go ahead. But if it’s not, then don’t put it in.”

The Essential Role of AI in Our Future

While the electrical power requirements of artificial intelligence are substantial, they represent a necessary investment in technology that promises to transform virtually every aspect of human society. From improving healthcare outcomes to accelerating scientific discovery, enhancing productivity, and creating entirely new economic opportunities, AI’s potential benefits are immense.

The energy consumed by AI systems should be viewed in context—as part of the infrastructure cost of building fundamentally transformative capabilities. Throughout history, technological revolutions have required significant resource investments that were ultimately justified by their long-term benefits. The same principle applies to artificial intelligence.

As we continue developing more powerful AI systems, the focus should be on maximizing energy efficiency and ensuring that electricity is sourced responsibly, preferably from renewable resources. With thoughtful planning and continued innovation in both AI algorithms and energy infrastructure, we can harness the full potential of artificial intelligence while minimizing its environmental impact.

If you are interested in this topic, we suggest you check our articles:

Sources: RAND, SciAm, TheVerge

Written by Alius Noreika