A boom that the electrical grid cannot sustain
The surge of artificial intelligence has sparked an unprecedented race to build ever-larger and more powerful data centers. These giants, known as hyperscale data centers, are the “digital factories” of the 21st century. Inside, thousands of GPUs work in parallel to train language models like ChatGPT, Gemini, or Claude, which demand immense amounts of computing power.
The issue is that this boom clashes directly with an uncomfortable reality: global electrical infrastructure is not prepared to absorb such demand in such a short time. Traditionally, energy consumption growth followed a predictable curve that allowed governments and electric companies to plan new plants, transmission lines, and network upgrades. But the rise of AI has broken that predictability.
According to the International Energy Agency (IEA), data centers—including those dedicated to AI—accounted for about 2% of global electricity demand in 2022. Projections for 2030 double that figure, nearing 1,000 TWh annually. To put it in perspective, that’s comparable to the total electricity consumption of an entire country like Japan or Germany.
The case of the United States is especially illustrative. In Virginia, a global cloud hub with over 275 data centers in Loudoun County, demand is growing so rapidly that Dominion Energy has had to delay shutting down coal and gas plants, contradicting climate goals.
The operator PJM Interconnection, responsible for 13 states, warns that if construction of data centers continues at the current pace, a structural deficit could develop in just five years: demand will surpass actual generation and distribution capacity.
This imbalance isn’t just an abstract problem. It has direct consequences:
– Rising electricity prices impacting households and businesses
– Increased risk of outages or scheduled blackouts in regions where the grid cannot handle peak loads
– Greater environmental pressure, as the response to demand often involves quick-start coal or gas plants with high CO₂ emissions
The AI boom has become a phenomenon comparable to past industrial revolutions. While 19th-century textile or steel factories transformed coal usage, today servers are redefining the amount and manner of energy needed. The difference lies in speed: what once took decades is now happening within two or three years.
This is not solely a technical challenge but a structural shift in the relationship between digitization and energy. As society demands sustainability and a green transition, AI poses a dilemma: how to power a technology that promises efficiency and progress without jeopardizing the planet’s energy stability.
A clear example of this tension is found in the US, where the electrical grid is approaching a critical point due to the rapid buildup of data centers. In Virginia, known as the global cloud hub, demand is growing so fast that Dominion Energy has had to postpone closing coal and gas plants, contrary to climate commitments. PJM predicts new AI projects will require an additional 10 GW of power in just a few years—the equivalent of Maryland’s entire electricity capacity. Wall Street now warns that the real bottleneck for AI won’t be chips but cheap, abundant electricity.
Even forgotten nuclear projects, like Three Mile Island, are being reactivated, with Microsoft showing interest in developing small, safe reactors to power data centers. Meanwhile, the average citizen already begins to feel the pressure; by 2025, states like Maine (+36%) and Connecticut (+18.4%) saw significant hikes in electricity bills.
Thus, the US becomes a laboratory for a looming global problem: how to sustain the AI revolution without collapsing the energy network or raising living costs.
What’s happening behind the scenes?
The hidden face of AI isn’t just the algorithms but the infrastructure that supports them. The operation of a modern data center reveals this energy voracity:
– Thousands of high-performance GPUs work tirelessly, 24/7, to train and run models like GPT or Gemini. Each chip can consume hundreds of watts, collectively representing the power demand of a small city.
– Cooling systems are another major consumer. Maintaining proper temperatures requires as much energy as the servers themselves, often coupled with large water usage.
– Electrical redundancy adds further pressure: diesel generators and backup batteries ensure continuous operation even during outages or grid failures.
In practice, a single hyperscale facility can consume more electricity than a city of 100,000 residents. And as AI continues to expand, the demand curve grows exponentially, foreshadowing increasingly severe stresses on the grid.
Given this scenario, leading tech companies are now building their own power plants. Previously a domain of utilities and governments, energy production is now becoming part of their business model. In the US, agreements are underway to develop modular nuclear reactors (SMRs) exclusively for powering AI data centers. Microsoft, for example, has hired nuclear engineers to assess the feasibility of small, safe reactors integrated into their facilities. Meanwhile, Amazon Web Services (AWS) invests in dedicated solar and wind farms to ensure green energy for its global network, though experts warn these solutions may not always cover the intermittent demand.
The movement extends beyond renewables and nuclear. In Virginia, some data center operators keep large diesel generators active as backup, despite environmental criticisms. In Asia, Alibaba and Tencent are exploring private hydroelectric plants in mountain regions of China.
The message is clear: major tech firms are no longer fully relying on national power grids—they are moving toward a future where megacenters are both energy consumers and producers, marking a deep shift in industry power dynamics. It’s a historic turn: tech companies are competing not only in algorithms but in megawatts.
The direct impact on citizens and businesses is significant.
The explosion of AI data centers arrives at a critical juncture: the push for a decarbonized energy future. Europe and the US have set ambitious emission reduction targets, but the voracious demand of hyperscalers threatens to overrun these plans.
Every megawatt allocated to a data center is a megawatt unavailable for electrifying homes, industries, or transportation. For example, in Ireland, citizen groups warn that national climate commitments are at risk because data centers already consume over 18% of the country’s electricity.
The tension isn’t just ecological but social as well. In the Netherlands, rural communities protest against giant tech facilities on farmland, while in Brazil and Chile, concerns grow about higher electricity costs for households during drought periods.
The dilemma is evident: AI offers productivity, innovation, and new jobs but also risks slowing the green transition and raising energy bills. Governments, pressed by climate urgency and global competition, must strike a balance between powering AI and ensuring clean, affordable energy for their populations.
The energy challenge extends globally. Ireland, now a major European data hub, has limited new data center connections due to grid capacity issues. The Netherlands faces public opposition to projects consuming vast water and energy resources with minimal local benefit. China accelerates in the opposite direction, expanding solar, hydro, and nuclear capacity to support its AI growth in provinces like Guizhou, aspiring to build a “Silicon Valley” of cloud computing.
In Latin America, countries like Brazil and Chile see data centers as economic opportunities, but structural weaknesses—aging grids and reliance on hydroelectricity vulnerable to drought—pose risks.
The current situation in the US is just the first chapter of a global story: AI threatens to reshape not just the digital economy but also the world’s energy map.
The environmental paradox is clear. AI is promoted as a tool to accelerate energy transition—optimizing grids, forecasting demand, integrating renewables—yet it can impede this very goal by consuming a significant share of available energy.
The risk is slipping into a cycle where digital growth hampers climate objectives. The pressing question remains: can technological innovation and sustainability coexist?
Possible solutions involve a range of strategies:
– Nuclear and geothermal energy as stable sources for data centers
– Dynamic demand management, adjusting load according to renewable availability, acting as “virtual batteries”
– Strategic placement near abundant clean energy sources like Iceland (geothermal) or Norway (hydro)
– Improving technological efficiency: developing more efficient chips and algorithms that reduce AI training energy requirements
Time is now the scarcest resource. As digital demand grows exponentially, electrical infrastructure develops much more slowly—taking 5 to 10 years just for major grid expansions. Tech giants, driven by the need for increasingly powerful models, cannot wait for governments. They are building private plants, signing long-term power purchase agreements, or exploring drastic options like small nuclear reactors or green hydrogen on-site.
Every delay in public infrastructure widens the gap. If AI advances faster than the capacity to generate and transport electricity, there’s a risk of partial network collapses, localized blackouts, or soaring costs affecting millions of homes and industries.
In this global race, governments and corporations face the pressing question: can the energy sector keep up with AI’s rapid evolution, or will energy become the bottleneck of future technological revolutions?
Frequently Asked Questions
1. How much energy does an AI data center consume?
A hyperscale facility can use more power than a city of 100,000 residents, with thousands of GPUs working continuously and intensive cooling systems.
2. Why are tech companies building their own power plants?
Because current grids cannot supply enough energy quickly enough. Private plants, including nuclear ones, guarantee stability and security for their expansion needs.
3. How does this boom affect ordinary citizens?
It translates into higher electricity bills and increased public health costs due to reliance on fossil fuels and diesel generators near urban areas.
4. Is this phenomenon exclusive to the US?
No. Ireland, the Netherlands, China, and Brazil face similar limitations due to rapidly expanding data centers and energy demands.
5. Can AI help resolve this crisis?
Potentially, yes. When applied to energy management, AI can optimize grids, improve demand forecasts, and facilitate renewable integration—though its own energy consumption must be managed carefully to ensure it doesn’t negate these benefits.