I have previously touched on how a great deal of conversation on artificial intelligence (AI) focuses on GPUs, software models, and datacentres.
But one of the most significant constraints shaping the future of AI may be far more fundamental.
Electricity.
As AI systems scale, the physical infrastructure required to power them is rapidly becoming one of the defining challenges of the AI economy.
AI Is Becoming an Energy Infrastructure Challenge
Training modern large language models requires enormous computing clusters made up of thousands of GPUs operating simultaneously.
These systems consume vast amounts of electricity.
Large AI training clusters can require hundreds of megawatts of power, and the next generation of hyperscale AI campuses may approach gigawatt scale energy demand (comparable to the electricity consumption of a small city).
According to the International Energy Agency, global electricity demand from datacentres could reach approximately 945 terawatt hours (TWh) by 2030, representing roughly 3% of global electricity demand.
Demand from datacentres is expected to grow at around 15% per year, more than four times faster than overall global electricity demand.
This growth is being driven largely by AI workloads.
For readers less familiar with electricity scale, it may help to understand what a terawatt hour represents.
What a TWh Means
A terawatt-hour (TWh) equals:
1,000,000,000 kilowatt-hours (kWh) or 1 trillion watt-hours.
It is a unit commonly used when discussing national electricity consumption, power grids, or large industries such as datacentres, mining, or manufacturing.
Simple Breakdown
Unit | Equivalent |
1 kWh | Energy used by a 1,000W appliance for 1 hour |
1 MWh | 1,000 kWh |
1 GWh | 1,000 MWh |
1 TWh | 1,000 GWh = 1 billion kWh |
Real-World Examples
To make it easier to visualise:
1 TWh of electricity can power roughly 150,000 – 200,000 Australian homes for a year.
Australia’s total electricity consumption is roughly 200–220 TWh per year, according to data from the Australian Energy Market Operator, the Australian Energy Regulator, and the Department of Climate Change, Energy, the Environment and Water.
Forecasts suggest that electricity consumption from AI datacentres alone could reach around 34 TWh by 2050 in some scenarios, highlighting the potential scale of demand growth.
(Source: International Energy Agency energy demand modelling for AI and datacentres.)
The Scale of AI Infrastructure Expansion
The infrastructure required to support AI is expanding rapidly.
Analysis by McKinsey & Company suggests that meeting AI infrastructure demand could require 50 to 60 gigawatts of additional datacentre capacity in the United States alone by 2030.
That represents a massive expansion of electricity demand, transmission infrastructure, and grid capacity.
Financial research from Goldman Sachs also estimates that global datacentre electricity consumption could grow by as much as 165% by 2030, driven largely by AI workloads.
This places significant pressure on power generation, transmission networks, and regional grid capacity.
When Power Becomes the Bottleneck
One of the emerging realities of the AI boom is that power availability is increasingly determining where new datacentres can be built.
While computer hardware can be deployed relatively quickly, building new electricity infrastructure can take years.
In several regions globally, new hyperscale datacentre developments are already facing delays due to limited grid capacity and the long lead times required to build new transmission infrastructure.
For example, industry reporting has highlighted that some datacentre developments in parts of the United States and Europe have experienced delays because regional electricity networks cannot currently support the additional demand from large AI facilities.
(Source: International Energy Agency and industry reporting on datacentre power constraints.)
This highlights an important shift in the AI ecosystem.
The next constraint on AI scaling may not be computer hardware.
It may be energy infrastructure.
Infrastructure Behind the AI Economy
The AI boom is often described as a software revolution.
It is also one of the largest infrastructure build outs of the digital era.
Scaling AI requires multiple physical systems to expand together:
• electricity generation
• transmission networks
• datacentres
• optical fibre connectivity
• cooling systems
Each of these systems forms part of the infrastructure that enables modern AI.
A Strategic Question
As AI becomes a defining technology of the next decade, the countries able to deploy infrastructure quickly will have a significant advantage.
Electricity powers AI.
Computer hardware performs AI.
As I have written previously, infrastructure such as optical fibre enables AI.
The challenge ahead is not simply building better models.
It is building the infrastructure capable of supporting them.
The AI race may ultimately be decided not only by algorithms or chips, but by the systems capable of powering them.
Key Takeaways
• AI infrastructure growth is increasingly constrained by electricity availability rather than computer hardware.
• Large AI training clusters can require hundreds of megawatts of power, with future hyperscale campuses approaching gigawatt-scale demand.
• Global datacentre electricity demand could reach ~945 TWh by 2030, according to the International Energy Agency.
• Building electricity infrastructure can take years longer than deploying computer hardware, making power availability a critical factor in AI infrastructure deployment.
• Countries that invest early in energy systems, datacentres, fibre networks and transmission infrastructure will be best positioned to host the next generation of AI infrastructure.
This article is part of a series exploring the physical infrastructure behind the AI economy.
References
Research and data referenced in this article include:
• International Energy Agency – Energy demand from AI and datacentres
• McKinsey & Company – AI infrastructure and datacentre capacity forecasts
• Goldman Sachs – Datacentre electricity demand projections
• Australian Energy Market Operator – Electricity demand statistics
• Australian Energy Regulator – Energy consumption data
• Department of Climate Change, Energy, the Environment and Water – National electricity consumption statistics
Footnote
This article is part of a series exploring the physical infrastructure behind the AI economy.
Disclaimer
The views expressed in this article are my own and are intended for general information and discussion purposes only. They do not represent the views of any employer, organisation, or client.
© 2026 Rodney Terry – Digital Backbone. All rights reserved.