The International Energy Agency’s latest report reveals a technology that is simultaneously getting greener and consuming more power than ever before. The world is not ready.
Key Takeaways
- The Energy Paradox: Despite unprecedented annual improvements in energy efficiency per AI task, the aggregate electricity consumption of data centers—especially those focused on AI—is surging. The International Energy Agency (IEA) projects that global electricity consumption from data centers will roughly double by 2030, with AI-focused data center consumption tripling in that period.
- Infrastructure Bottlenecks: The rapid growth of AI is creating a bottleneck crisis where the physical world cannot keep up with digital ambition. Grid constraints could delay approximately 20% of global data center capacity planned for construction by 2030, and shortages of high-bandwidth memory are anticipated to persist through at least 2027.
- AI as the Solution and Policy Imperative: AI is not just an energy taker but is becoming an “energy maker,” capable of helping energy-intensive industries reduce their energy costs by 3 to 10 percentage points and accelerating scientific discovery for materials and battery chemistries. Policy intervention is crucial, requiring approaches that promote electricity system flexibility and remove barriers to AI adoption in the energy sector.
There is a phrase that has quietly become the IEA’s defining axiom for our era: there is no AI without energy. It sounds simple enough. But the International Energy Agency’s newly published report, Key Questions on Energy and AI, shows just how staggeringly complex and urgent that relationship has become.
The numbers alone are arresting. Global electricity demand from data centres, the critical infrastructure for training and running AI models, grew by 17% in 2025. Electricity consumption from AI-focused data centres grew even faster, surging 50% in that year alone. This is not an incremental change. This is a structural reshaping of how the world consumes power, happening at a pace that grids, regulators, and supply chains were never designed to absorb. IEA
And yet the report’s most counterintuitive finding is not the surge; it is the efficiency miracle happening in parallel. Measured per individual task, the energy efficiency of AI is improving at a rate unprecedented in energy history. Software and hardware advances have resulted in the energy use per AI task dropping by at least an order of magnitude annually in recent years. Simple text queries now typically consume less electricity than running a television over the same period of time. IEA
So AI is simultaneously becoming greener per task and consuming vastly more electricity in aggregate. This is the central paradox policymakers must confront.
At Least a Twofold Increase by 2030
The trajectory the IEA projects should concentrate minds in every energy ministry on the planet. Updated projections see electricity consumption from data centres roughly doubling from 485 TWh in 2025 to 950 TWh in 2030, accounting for around 3% of global electricity demand by that date. Electricity consumption from AI-focused data centres grows much faster than overall data centre electricity consumption, tripling in this period. IEA
To put that in context: a typical AI-focused data centre consumes as much electricity as 100,000 households, and the largest ones under construction today will consume 20 times as much. IEA
Driving this is not just usage growth but investment at an almost incomprehensible scale. The capital expenditure of just five technology companies exceeded USD 400 billion in 2025 and is expected to jump by another 75% in 2026. Capital expenditure of just those five companies is now larger than global investment in oil and natural gas production. Let that sink in: five Silicon Valley firms are now outspending the entire global fossil fuel extraction industry on infrastructure. IEA
This unprecedented surge is colliding with aging and under-invested electric grids. In regions like the United States and Europe, permitting new power plants and transmission lines can take over a decade, far too slow to meet the AI industry’s timeline. This has led to a “gigawatt squeeze,” where projects are delayed, and compute clusters sit idle, waiting for a connection to a grid that cannot deliver. Developers are increasingly exploring “off-grid” solutions, bringing their own power sources like natural gas, microgrids, and even nuclear reactors to site. The challenge is not just absolute capacity; it is the grid’s inability to deliver a “certain type” of flexible power that a one-size-fits-all system cannot easily accommodate.
The Bottleneck Crisis Nobody Is Talking About
Behind the headline investment figures lies a quieter, more troubling story, one of physical limits asserting themselves against digital ambition. The speed of the AI revolution is increasingly contrasting with the speed of the physical, social, and economic systems that underpin it. Bottlenecks across energy supply chains and advanced chip manufacturing have tightened. Planning and regulatory systems are being stretched by the wave of project applications for data centres, amid a broader trend of rapid load growth and electrification. IEA
This is not a hypothetical risk. The IEA’s analysis is stark: grid constraints could delay around 20% of global data centre capacity planned for construction by 2030. One in five planned facilities may simply not connect to the grid on schedule. That is a direct constraint on the AI ambitions of nations and corporations alike.
The chip supply chain is equally fragile. A shortage of high-bandwidth memory, integral to AI chip production, has developed over the past six months and is anticipated to persist through at least the end of 2027.
Supply Chain Fragility and Geopolitical Risks The chip shortage is exacerbated by a fragile and highly concentrated supply chain. Global memory production is dominated by a handful of companies, primarily in Asia, creating a single point of failure. Geopolitical tensions, such as conflict in the Middle East and the potential closure of critical trade routes like the Strait of Hormuz, pose a major risk. A disruption could spike energy costs and sever access to essential raw materials like helium and specialized acids, directly impacting chip factories and prolonging the “chip tightness” potentially until the end of the decade. This has triggered a rush by nations and corporations to secure their own supplies, a form of hardware hoarding that mirrors the wider power grab.
Meanwhile, the geographic concentration of these projects is compounding local strain. 50% of data centres under development in the United States are in pre-existing large clusters, potentially raising risks of local bottlenecks. IEA
The Critical Minerals Time Bomb
The energy story of AI is inseparable from a geopolitical one. Apart from bulk materials like steel and concrete, the construction of data centres requires sizeable amounts of several minerals and metals, such as copper, aluminium, silicon, gallium, and rare earth elements. There is a significant overlap between the minerals needed for building new data centres and those that are critical to energy technologies. IEA
The supply concentration figures should alarm any strategist. In 2030, data centre demand for gallium could equal up to 10% of today’s supply, and China accounts for 95% of gallium refining. The high market concentration for critical minerals highlights significant vulnerabilities to supply shocks, whether from extreme weather events, industrial accidents, trade disruptions, or geopolitics. IEA
This is not a niche technical concern. It is the next front in the global competition for technological sovereignty.
AI Is Also the Solution
It would be a journalistic failure to leave this story as pure alarm. The IEA is equally insistent on the other side of the ledger, that AI, if deployed at scale in the energy sector, could help solve precisely the problems its infrastructure is creating.
Proven applications of AI could help firms in energy-intensive industries reduce their energy costs by 3 to 10 percentage points. For energy-hungry manufacturers, that is a transformative margin. In scientific discovery, AI led to a 45,000-fold acceleration in the mapping of protein structures – critical for designing new drugs – and could allow scientists to dramatically accelerate the process of finding and testing promising materials, battery chemistries, and carbon capture molecules. IEAIEA
The tech sector is also beginning to respond to the clean energy challenge it helped create. The tech sector accounted for around 40% of all corporate power purchase agreements for renewables signed in 2025, and is also now a major source of momentum for the nuclear and advanced geothermal industries. The pipeline of conditional offtake agreements between data centre operators and small modular reactor nuclear projects has grown from 25 gigawatts at the end of 2024 to 45 gigawatts today. IEA
As IEA Executive Director Fatih Birol has put it, AI is still an energy taker, but it is also becoming an energy maker – driving forward innovative solutions like next-generation nuclear reactors, flexible data centres, and long-duration energy storage. IEA
The Affordability Question
There is one dimension the IEA raises that deserves far greater public attention than it typically receives: what does all this mean for ordinary electricity bills?
The report finds that if the right mix of policies and infrastructure investment is in place, increases in electricity demand do not necessarily raise prices. However, data centres can create special challenges for electricity affordability, since they have large, concentrated power loads and scale up rapidly, often triggering the need for new generation assets and grid investment. IEA
The political economy here is delicate. Communities are already pushing back. Social acceptability is a growing issue, as communities push back against data centre projects, and concerns about affordability and environmental impacts rise. Policymakers cannot afford to dismiss these concerns as technophobia. They reflect legitimate questions about who bears the cost of infrastructure built to serve a global tech industry. IEA
The Policy Imperative
The IEA’s prescriptions are clear and, frankly, not yet being followed with sufficient urgency. Approaches that promote electricity system flexibility can help accelerate grid connections and ensure electricity affordability. System operators can explore non-firm grid connections and incentivise data centre developers to provide demand response in return for faster connection processes. IEA
On the industrial side, removing barriers to AI adoption in the energy sector can ensure AI is leveraged to enhance energy security and sustainability, with comprehensive policy frameworks that address data availability, cybersecurity, skills, and interoperability, crucial for boosting AI uptake. IEA
And for the developing world – the part of this story most often absent from Western coverage, the stakes are existential in a different way. Emerging and developing economies other than China account for 50% of the world’s internet users but less than 10% of global data centre capacity. Countries with a record of reliable and affordable power will be best placed to unlock data centre growth, localise computing power critical to homegrown AI development, and spur the IT industry more generally. IEA
The AI revolution, in other words, risks deepening the very digital divide it promises to transcend – unless policymakers in both rich and developing nations act with deliberate urgency.
The IEA’s Key Questions on Energy and AI is a landmark document precisely because it refuses false comfort. It does not tell us that efficiency gains will automatically solve the demand problem, nor that the investment surge is inherently catastrophic. What it tells us is that the outcome depends entirely on the choices made now – on grids, on supply chains, on regulation, and on whether the energy sector embraces AI as a tool rather than merely tolerating it as a load.
The technology is moving faster than the physical world it depends on. That gap is the defining energy challenge of this decade.
Other People are Reading
Continue Reading
