Why Does AI Run on Electricity? Why Silicon Valley Is Betting on Nuclear Power

Tracking the massive energy transition triggered by the insatiable energy appetite of data centers, the invisible engine of the AI revolution.

• Understand the exact scale of power consumption by AI data centers.
• Learn why Silicon Valley has shifted its energy strategy toward nuclear power.
• Discover the realities and challenges of Small Modular Reactors (SMRs), a new alternative in nuclear energy.

Main Text

The Invisible Engine of the AI Revolution

Imagine the magical moment when you ask generative AI to write a poem or create a fantastic image from just a few words. Behind that curtain, however, lies a massive physical engine: countless servers running hot, the data center. Every simple question we ask wakes this giant machine, causing it to gulp down enormous amounts of electricity. Asking a chatbot 26 questions consumes as much energy as heating lunch in a microwave, and a single question uses enough power to charge a smartphone for 24 minutes.

Here lies the greatest paradox of modern technology. The AI revolution, once seen as a symbol of an intangible future, is creating the largest and most centralized energy demand in human history. This endless appetite for energy forces a fundamental reevaluation of global energy policies and has even led Silicon Valley—once a staunch advocate of “100% renewable energy”—to embrace a previously shunned energy source: nuclear power.

A Billion-Watt Brain: Quantifying AI’s Insatiable Appetite

Why has AI become an “electricity hippo”? First, AI computations rely on running thousands of high-performance GPUs simultaneously. Training a large model like GPT-3 requires about $1.3 \text{GWh}$ of electricity—equivalent to the daily power consumption of thousands of households.

Second, the bigger problem is ‘heat.’ Intensive computations generate massive heat, and cooling systems consume nearly as much electricity as the computing itself. Currently, about 40% of data center power is used solely for cooling.

The International Energy Agency (IEA) projects global data center electricity use will more than double from 460 TWh in 2022 to 1,050 TWh by 2026—an amount comparable to Japan’s entire annual electricity consumption. Approximately 80% of this demand will be concentrated in the US and China, turning energy policy into a key tool in the geopolitical competition for AI dominance.

Table 1: Global Data Center Electricity Consumption Forecast

The challenge is how to meet this explosive demand. AI data centers require stable, uninterrupted power 24/7, 365 days a year—a typical baseload demand. Intermittent renewables like wind and solar, which fluctuate with weather and time of day, cannot reliably meet this demand alone. Consequently, the IEA forecasts that over 40% of new data center power demand by 2030 will still be met by natural gas and coal. This is the ‘climate paradox’ where the AI revolution directly conflicts with humanity’s decarbonization goals.

The Great U-Turn: The World Embraces Nuclear Power Again

After the Fukushima nuclear disaster in 2011, the world experienced a decade-long “nuclear winter.” However, the combined pressures of climate change, energy security, and AI have dramatically reversed this trend.

• USA: The Inflation Reduction Act (IRA) offers unprecedented tax incentives for nuclear power, and California extended the life of the Diablo Canyon nuclear plant, which was slated for closure amid power shortages.
• Europe: France announced plans to build up to 14 new reactors, and the UK unveiled a roadmap to quadruple its nuclear capacity.
• Asia: Even Japan, the site of the Fukushima disaster, has shifted toward restarting and extending the life of nuclear plants, while South Korea has formalized plans to build three new reactors and commercialize next-generation SMRs.

An Unexpected Alliance: When Silicon Valley Meets Nuclear Reactors

The most surprising development is that Silicon Valley tech giants have become some of nuclear power’s most enthusiastic supporters. Once vocal advocates of RE100 (100% renewable energy), they are now actively reaching out to nuclear power. This pragmatic shift acknowledges that renewables alone cannot reliably supply the massive, 24/7 power demands of data centers.

This strategic pivot is reflected in concrete investments. Amazon acquired a data center next to a nuclear plant and invested in SMR developers; Microsoft signed a long-term contract including the restart of the troubled Three Mile Island (TMI) reactor; even Google has made large power purchase agreements with SMR startups, boldly betting on the future of nuclear power.

Table 2: Big Tech and Nuclear Power Alliances

Small and Beautiful? Is SMR a Panacea?

At the heart of the new nuclear renaissance is the Small Modular Reactor (SMR). Promoted for its ‘3S’ advantages—Safety, Scalability, and Speed—it is hailed as a technology that could transform the nuclear paradigm.

However, the collapse of the first commercial SMR project by US-based NuScale due to economic issues has dampened enthusiasm. Rising construction costs led to higher electricity prices, causing customers to cancel contracts. This highlights the challenge SMRs face: overcoming the “valley of death”—the huge upfront costs and risks of the initial unit—to achieve commercial viability.

A Harsh Reality: The Eternal Obstacles on Nuclear’s Path

Despite gaining a powerful ally in AI, the future of nuclear power remains fraught with challenges.

• Cost Issues: New nuclear plants, like the Vogtle plant in the US, often face massive cost overruns and construction delays beyond forecasts.
• Nuclear Waste: There is still no permanent disposal solution anywhere in the world for spent nuclear fuel, which remains dangerously radioactive for tens of thousands of years. This poses serious ethical and technical dilemmas, effectively passing the problem to future generations.
• Social Acceptance: Regardless of technical safety, NIMBYism (“Not In My Backyard”) and political volatility remain the biggest barriers to nuclear projects.

Conclusion

AI’s insatiable energy demand leads us to an era that requires not a choice between renewables or nuclear, but both renewables and nuclear. Variable power from renewables must be complemented by nuclear’s stable, 24/7 baseload capacity in a mutually supportive system. Currently, nuclear power is the only proven large-scale zero-carbon technology capable of fulfilling this role.

We stand at a critical crossroads. Decisions about energy infrastructure over the next decade will determine not only whether climate goals are met but also the ultimate limits of the AI revolution. Humanity’s quest to create artificial brains paradoxically compels us to fundamentally redesign the physical world that powers those brains.