The AI Revolution's Power Problem

We are potentially in the midst of the largest industrial revolution in human history. Artificial intelligence promises to redefine how we work and live, but there is one essential ingredient: electricity. How this is sourced is an essential question for investors.

A Trillion-Dollar Infrastructure Challenge

Trillions of dollars are expected to flow into AI infrastructure over the coming years. While estimates vary, the numbers are staggering. McKinsey estimates that by 2030, $6.7 trillion will be invested in data centre infrastructure, with $5.2 trillion for AI-capable facilities.

Today, data centres already consume approximately 2-3% of global electricity, expected to double within five years. In Ireland, data centres consume over 20% of national supply. The International Energy Agency projects data centre electricity use could equal Japan’s entire annual consumption by 2030, eqating to approximately 110 GW of continuous power.

This demand arrives as decarbonisation efforts require the ‘electrification of everything’. This includes heating systems, industrial processes, and transportation networks. This is creating combined strain on generation and associated grids.

Can Supply Meet Demand?

Despite recent political rhetoric to the contrary, increases in power demand continues to be absorbed predominantly by renewable energy. Global gas capacity additions have averaged just 10 GW annually in recent years, while new nuclear capacity this year will total a modest 3–5 GW. By contrast, 700 GW of renewable capacity will be installed globally this year alone, producing more electricity annually than Japan consumes.

The challenge however is one of timing. AI infrastructure investment and development is happening now. Data centre developers cannot wait for gas plants or nuclear facilities to be permitted, financed, constructed, and commissioned, a process that typically spans a decade or more. Gas supply chains face additional constraints, with turbine manufacturing capacity already stretched and lead times extending to years for new orders.

Solar and renewable energy will likely fill the gap by necessity. Projects can be brought online within 18–24 months, aligning with AI deployment timelines. However, intermittency challenges will require complementary solutions: battery storage, operational flexibility within data centres to shift loads, grid interconnection, and likely intermittent fossil fuel-based flexible generation to manage supply gaps.

The Environmental Cost

The AI boom looks likely to drive emissions in the wrong direction. The International Energy Agency projects data centre emissions will grow from 180 million tonnes of CO2 today to between 300 and 500 million tonnes by 2030-2035, depending on adoption rates and efficiency improvements.

Yet AI appears a double-edged sword. This same technology can optimise grid operations, improve renewable energy forecasting, and accelerate decarbonisation efforts.

The World Economic Forum suggests AI could help mitigate 5-10% of global emissions by 2030 through improved energy management and system efficiency.

Real-world deployment, however, tells a more complex story. In 2024, Elon Musk’s xAI data centre in Memphis installed 35 gas turbines without environmental permits. Optical gas imaging revealed visible pollution plumes, including unburned methane, drifting beyond facility boundaries.

Water withdrawal also presents an additional constraint. Research at the University of California suggests AI cooling could withdraw 4.2 to 6.6 billion cubic metres annually by 2027 – more than half the United Kingdom’s total consumption.

Investment Opportunities

AI and electrification are driving unprecedented demand for grid infrastructure upgrades. Renewable energy developers, energy storage providers, and grid modernisation companies are positioned to benefit – the question is which can deploy at scale while managing environmental and social risks effectively.