Hidden cost of AI: Energy crisis behind digital revolution

Artificial intelligence is hailed as the defining technology of the 21st century. Large language models, generative tools and conversational chatbots dominate the public imagination, promising to transform entire industries and daily life. However, behind this excitement lies a less glamorous truth: AI is hungry for electricity. Without abundant, reliable power, the much-heralded AI revolution risks stalling.

The numbers are stark. According to the International Energy Agency (IEA), global data centers consumed around 415 terawatt-hours of electricity in 2024, equal to 1.5% of total global use. The United States and China alone accounted for 70% of that demand, underscoring how AI power is already concentrated in a handful of geographies. By 2030, global data center demand could nearly double to 945 TWh, more than Japan’s current consumption. And unlike earlier waves of digitalization, AI workloads are the dominant driver of this surge. So at this point, the challenge is clear: how can states remain at the forefront of AI while meeting climate commitments and avoiding energy crises? That dilemma is fast becoming one of the central geopolitical questions of our time.

AI’s spiraling appetite

AI’s electricity needs dwarf those of traditional digital services. A single ChatGPT query can consume 10 times the energy of a Google search. Training a cutting-edge model can require megawatts of continuous power for weeks. In the U.S., data centers consumed 176 TWh in 2023, or 4.4% of national electricity, a figure projected to rise to 6-12% by 2028. China’s demand, growing 15% annually since 2015, has reached 100 TWh and now represents a quarter of the world’s total. Together with Europe, these giants account for roughly 85% of global consumption.

Of course, we must understand that this is not just about servers overheating in Silicon Valley or Shenzhen. It is about entire grids being reshaped. In Ireland, where hyperscale centers already use nearly a fifth of the country’s electricity, the country’s energy operator has restricted new data center connections. Similar concerns are emerging in the Netherlands, Singapore and even parts of the U.S. The question policymakers face is not whether AI will consume more power, because it will, but whether grids can adapt quickly enough without breaking climate pledges or raising public backlash over rising bills.

Search for power

Where will this electricity come from? For now, AI data centers draw from the same fossil-heavy grids that power the rest of the economy. Coal, oil and gas still make up more than 80% of primary energy worldwide. In the U.S. and Europe, this dependence is politically unsustainable. Both Washington and Brussels are pushing tech firms to sign long-term contracts for renewables. The EU’s Green Deal and the U.S. Inflation Reduction Act are designed not only to decarbonise transport and industry but also to green the digital sector.

Elsewhere, however, the picture is different. China and India still rely on coal, but both are experimenting with nuclear expansion and record renewable deployment to match AI’s growth. The Gulf states are seizing the moment, tying digital investments directly to massive solar and nuclear projects. The United Arab Emirates (UAE), for instance, is marketing itself as a hub for “green compute,” hoping to export AI services powered by clean electricity.

However, it is understandable that renewables alone cannot shoulder the burden. Wind and solar are cheap and fast to deploy but intermittent. The IEA expects natural gas generation to increase by 175 TWh by 2030 just to cover new data center demand, locking in fossil infrastructure at the very moment when it should be phased out. Furthermore, nuclear is being revived as a low-carbon baseload option. U.S. tech giants from Google to Amazon are already signing power purchase agreements with nuclear plants, while policymakers bet on small modular reactors to arrive by the early 2030s.

And then comes fusion. Long ridiculed as “always decades away,” it is suddenly attracting billions. Google’s 2025 deals to back and eventually purchase fusion power signal a belief that AI will require entirely new energy paradigms. Furthermore, China is spending $1-1.5 billion annually on its fusion program. While commercialisation is far from certain, the mere fact that fusion is now on the agenda shows how AI is bending the energy imagination of states and companies alike.

Can AI solve its energy use?

There is a paradox: AI is not only an energy guzzler, but it could also be an energy saver. Advanced machine learning can forecast demand, balance renewable supply, and manage distributed storage far more efficiently than traditional systems. The World Economic Forum estimates that AI-enabled tools could cut global emissions by up to 10% by 2030, equal to the EU’s annual output. To provide a specific case, Abu Dhabi National Oil Company offers a glimpse: its AI-driven energy management saved $500 million and avoided emissions equivalent to removing 200,000 cars from the road.

But efficiency carries risks. Jevons’ Paradox, the idea that efficiency gains drive greater overall consumption, looms large. Cheaper, smarter computation encourages more of it. Unless governments set boundaries, AI could accelerate energy demand beyond what grids and climate policies can bear. This tension, between AI as both savior and stressor, lies at the heart of the policy debates lately.

Geopolitics of AI’s hunger

The AI revolution is redrawing the global energy map. The ability to train and deploy frontier models is no longer determined only by chips, data or talent. It increasingly depends on something more basic: who controls enough reliable electricity to run them. In this sense, AI has transformed energy security from a traditional question of keeping the lights on into a strategic question of sustaining digital dominance, and therefore into a national security issue.

China understands this better than most. Its AI strategy is inseparable from the state-led expansion of coal, nuclear and renewables. By securing electricity at scale, Beijing ensures that its firms can train the next generation of models without interruption, even if the grid faces wider strains. The U.S. is pursuing a distinctive strategy that intertwines digital industrial growth with its broader decarbonization agenda. The Inflation Reduction Act delivers substantial incentives, such as grants, tax credits and loans, to stimulate clean energy and low-carbon manufacturing. Meanwhile, the CHIPS & Science Act provides targeted subsidies and tax credits to secure semiconductor and AI hardware production on U.S. territory. Together, these policies aim to anchor digital and high-tech growth within a domestic, low-carbon industrial base, although significant project delays some policy implementations have occurred.

The Gulf states are trying to leapfrog, branding themselves as providers of “green compute” through solar megaprojects and nuclear capacity. By doing so, they aim to shift from oil exporters to data exporters, selling not barrels but compute cycles powered by clean energy. Europe, meanwhile, once again, risks being caught in the middle. Its strong climate rules may restrict AI deployment unless it can scale renewables even faster or import clean electricity from neighbours.

The likely outcome of these developments is a fragmented world of AI-energy zones. Energy-intensive training may be offshored to countries with cheap and abundant power, Iceland with geothermal, Canada with hydro, or the UAE with subsidised solar. Inference, which requires lower latency, might remain closer to consumers. But this division of labour introduces new dependencies. Just as oil once shaped strategic alliances, access to clean compute could dictate future diplomatic ties. Who hosts whose data centers may become as politically sensitive as who hosts whose military bases.

This raises a deeper danger: a widening digital divide rooted in energy inequality. Advanced economies that can subsidise clean energy for AI will race ahead, while middle and low-income states risk being left behind, unable to both decarbonize and compete in AI simultaneously. The political fallout could mirror past energy politics, such as resentment, dependency and vulnerability, only this time revolving around electricity rather than oil.

Türkiye’s opportunity

For middle powers, the challenge is magnified, but so too is the opportunity. Türkiye offers a vivid example. Its current data center capacity is still being built, with around 250 MW, consuming less than 0.5% of national electricity. But President Recep Tayyip Erdoğan’s 2025 AI strategy sets an ambitious target: 1 GW of capacity by 2030, supported by over $10 billion in investment.

This is no small bet. In early 2024, renewables already provided 53% of Türkiye’s power. Solar alone doubled in just over two years to nearly 20 GW, surpassing the government’s 2025 target ahead of schedule. The Akkuyu nuclear plant, with four reactors set to generate 35 TWh annually, will soon cover 10% of national demand. This combination of abundant renewables and nuclear baseload gives Türkiye a unique chance to power its digital economy with low-carbon electricity.

Crucially, Türkiye’s geography adds leverage. At the crossroads of Europe, the Middle East and the Caucasus, Ankara could market itself as a regional computing hub. Low-latency, low-carbon AI services could be exported to three continents, positioning Türkiye not just as a consumer of digital tools but as a provider of critical infrastructure.

AI is changing the energy question from one of supply security to one of strategic advantage. The race for AI is becoming, at its core, a race for energy, who can marshal the watts to fuel the weights of AI models. For leading powers, it means aligning industrial policy, climate strategy and digital ambition. For middle powers like Türkiye, it means seizing the window to turn geography and energy transition into geopolitical leverage.

What is clear is that the AI revolution cannot be separated from the energy revolution. Those who treat them in isolation will fall behind. Those who connect them, politically, technologically and strategically, will shape not just the future of AI but the future balance of power itself.