The New Arms Race: AI, Electricity, and the Paradox of Progress

Beginning of this month, Meta signed a striking deal in the American state Illinois. The parent company of Facebook and Instagram will purchase nuclear power from Constellation Energy for the next 20 years to meet its surging energy needs. The Clinton-based nuclear power plant, once scheduled for closure, will now stay open at Meta’s request to supply power to its AI data centers.

In late 2023, Meta announced plans to build a $10 billion data center in Louisiana, purpose-built to train the next generation of LLAMA models. The facility will require 2 gigawatts of energy, the output of two nuclear power plants.

Electricity as the ultimate strategic advantage

This marks the beginning of a new geopolitical and economic reality: in the race to dominate artificial intelligence, electricity has become the ultimate strategic advantage. As OpenAI CEO Sam Altman succinctly put it, “Eventually, the cost of AI will converge to the cost of electricity.”

Today, data center capacity is no longer measured in square meters, server racks, or GPUs, it’s measured in gigawatts. The key question is: how much energy can an AI factory secure? The answer increasingly determines who leads. Microsoft, for example, signed a landmark deal to bring the Three Mile Island nuclear plant — once dubbed “America’s Chernobyl” — back online. While Google recently locked in 1.8 gigawatts of advanced nuclear power.

Billions are being poured into this transformation. In 2024, Microsoft built data centers at a staggering rate of $1 billion per week. In 2025, Amazon is set to outpace that with $2 billion per week. Elon Musk recently outfitted a former Electrolux factory in Memphis with 200,000 Nvidia H100 GPUs, rebranding it as X.ai’s “AI factory.” Since the local grid couldn’t handle the load, diesel generators were brought in, creating a cloud of smog and a storm of protest.

Just as DeepSeek seemed to offer a hopeful turn, with AI models that train 11 times faster, reality reminded us of an old truth: Jevons Paradox. When James Watt made the steam engine 90% more efficient in the 18th century, many assumed coal use would drop. Instead, it soared. As steam power became cheaper and easier, consumption skyrocketed. Microsoft CEO Satya Nadella recently cited this very dynamic in relation to AI. Greater efficiency doesn’t reduce usage, it amplifies it.

Emissions are rising, too. Largely due to the explosive growth of AI, Microsoft’s carbon output has increased by 23% since 2020, despite promises of carbon neutrality. The AI sector already consumes a significant share of global energy and demand is growing fast. The paradox is clear: we’re building the digital brains of the future on an energy system that still relies heavily on outdated, often polluting sources.

Yet this issue gets very little attention in public debate. We focus on algorithmic bias, automation-driven job loss, copyright, and deepfakes but rarely on the sheer energy appetite of the technology behind it all. We diligently bring reusable containers to zero-waste shops, but also use ChatGPT en masse to write emails and papers.

Our fascination with megabytes and microchips has given way to a romance with megawatts and kilotons. AI isn’t a pipe dream, it’s an industrial revolution. And like every revolution, it runs on raw materials. Electricity is the new oil, and data centers are the new factories.

What about Europe?

So the big question is: where does Europe fit into this picture?

While the U.S. rapidly links data and power infrastructure, Europe remains caught between climate ambitions and multi-year permitting processes. We lead the world in ethics, regulation, and privacy, but without a strong digital and energy backbone, that won’t be enough to compete in the AI arms race.

If Europe wants a real seat at the table in shaping AI’s future, we’ll need to look beyond legislation and think about gigawatt contracts, nuclear power, and digital sovereignty. Today, technology policy is energy policy.

History shows that those who control infrastructure, shape the future. More than a challenge, these types of evolutions are a massive opportunity, as I explore in my brand-new book The Uncertainty Principle (You can order it here) that just came out. It’s time for Europe not just to watch from the sidelines, but to step up, invest boldly, and help build the backbone of tomorrow’s AI economy.

This is an adapted version of a Dutch piece that was originally published by De Tijd.