Tech Titans Turn to Atomic Power to Fuel the Future

Juan Manuel Ortiz de Zarate
Mar 30
10 min read

In a bold and unprecedented shift, major technology companies—Amazon, Google, and Microsoft [1]—are embracing nuclear energy to meet the surging demand for electricity fueled by artificial intelligence (AI). This move signals a new era where data, innovation, and atomic power intersect, reshaping the energy landscape and redefining how we think about sustainable infrastructure in the digital age.

The Energy Challenge of AI

Artificial Intelligence is revolutionizing modern society. From healthcare and finance to logistics and entertainment, AI models are being deployed at breakneck speed. But this progress comes with a hidden cost: massive energy consumption.

Training large-scale AI models requires staggering computational power. Consider GPT-3, a model with 175 billion parameters: its training consumed approximately 1,287 megawatt-hours (MWh) of electricity and emitted over 500 metric tons of CO₂—equivalent to the lifetime emissions of five average cars. BERT, another well-known model, required energy comparable to a transcontinental flight just to train.

This is not an isolated issue. The computational resources needed for training cutting-edge models have been doubling every 3.4 months, creating exponential growth in energy consumption. Data centers supporting AI workloads now consume increasing amounts of electricity, not only for computation but also for cooling, data storage, and model deployment.

Why AI Needs Clean Energy Now

The environmental implications of this energy use are significant. Carbon emissions, water usage, and electronic waste are all rising due to the scale and intensity of modern AI development. For instance, Google reported a 48% increase in its greenhouse gas emissions [17] over five years, partly driven by AI operations. Meanwhile, data centers in arid regions strain local water supplies used for cooling. Frequent hardware upgrades also contribute to growing e-waste.

Estimated electricity consumption of data centers worldwide. Source [15]

These impacts have led AI companies to seek sustainable, reliable energy sources to power their operations—and nuclear energy is rapidly emerging as the frontrunner.

Why Nuclear?

Nuclear energy currently supplies about 18% of the United States’ electricity [2] and far more in countries like France. Its unique combination of zero carbon emissions during operation and consistent, high-output generation makes it ideally suited for powering energy-intensive AI infrastructure.

Unlike renewables such as wind and solar, nuclear energy is not intermittent. It provides steady power regardless of time of day or weather conditions—critical for the always-on demands of cloud computing and AI model training. For companies that have pledged carbon neutrality or 24/7 carbon-free energy (like Google and Microsoft), nuclear energy offers a viable path to meeting those goals without sacrificing performance.

Amazon’s Nuclear Vision

Among the tech giants, Amazon is taking the lead in building nuclear capacity for its operations. The company spearheaded a $500 million investment in X-energy [3], a startup specializing in small modular reactors (SMRs)—a promising new class of nuclear technology designed to be safer, more cost-effective, and faster to deploy than traditional reactors.

X-energy’s reactors use TRISO fuel [4], which encases uranium particles in layers of carbon and ceramic, enabling them to withstand high temperatures and reducing the risk of meltdown. The International Atomic Energy Agency considers SMRs to be a safer advancement over earlier reactor models [6], although some groups, like the Union of Concerned Scientists, urge caution and demand rigorous testing [7].

TRISO particles can withstand extreme temperatures well beyond the threshold of current nuclear fuels. Source [4]

Amazon's investment goes beyond funding. The company has partnered with Energy Northwest to deploy a 320-megawatt X-energy reactor in Washington State, with an option to expand to 960 megawatts. Additionally, a separate agreement with Dominion Energy will see another SMR built in Virginia, adding 300 megawatts to support Amazon Web Services’ data centers.

These investments demonstrate Amazon’s commitment not only to meeting its own energy needs, but also to reshaping the future of industrial-scale power generation.

Google’s Strategic Partnership with Kairos

Google, similarly, is investing in nuclear energy through a partnership with Kairos Power [8], another startup developing small modular reactors. Although financial details have not been disclosed, the collaboration is a cornerstone of Google’s plan to operate on 100% carbon-free energy 24/7 by 2030.

Kairos is pioneering fluoride-salt-cooled high-temperature reactors, which promise increased safety and efficiency. These reactors operate at lower pressures and use molten salt as a coolant, reducing risks and improving thermal performance.

In 2024, Kairos broke ground on a demonstration plant in Tennessee, the first SMR project to be approved by the U.S. Nuclear Regulatory Commission. The demonstration plant is expected to come online in 2027, with additional units projected for 2030 and beyond, ultimately supplying up to 500 megawatts of power.

Google’s broader sustainability efforts include sourcing renewable energy, improving data center efficiency, and optimizing AI algorithms—all complemented by this nuclear initiative.

Microsoft’s Revival of Three Mile Island

In a move both symbolic and practical, Microsoft signed a 20-year power purchase agreement with Constellation Energy [10] in 2024 to restart Unit 1 of the Three Mile Island nuclear plant in Pennsylvania. While Unit 2 was the site of the infamous 1979 meltdown, Unit 1 was unaffected and operated safely for decades before being shut down in 2019 due to economic challenges.

Three Mile Island, which closed in 2019, is soon reopening. Andrew Caballero-Reynolds/AFP/Getty Images

Microsoft’s deal will help bring the plant back online by 2028, offering a substantial boost to the company’s carbon-free energy portfolio. This approach reflects a pragmatic strategy: rather than building from scratch, Microsoft is revitalizing dormant infrastructure to meet immediate energy needs while fulfilling its climate commitments.

Mitigation Strategies Beyond Nuclear

Nuclear energy isn’t the only strategy being employed to address AI’s environmental impact. A multifaceted approach is emerging, encompassing algorithmic innovation, hardware development, and smarter infrastructure design.

Algorithmic Efficiency

One of the most effective ways to reduce energy use is to optimize the algorithms themselves. Techniques like model pruning, quantization, and knowledge distillation help shrink model sizes without compromising performance. This not only reduces training time but also lowers the computational burden during deployment.

Startups and researchers are actively pursuing this path. For instance, DeepSeek's R1 model [12], developed in China, matches the performance of ChatGPT while using far fewer resources. As an open-source alternative, it also lowers the barrier to entry for developers around the world.

Next-Gen Hardware

Advancements in energy-efficient hardware offer another promising route. Neuromorphic chips, which mimic the structure and function of the human brain, are gaining traction. Intel’s Loihi chip [14], for example, is designed to be vastly more efficient than traditional CPUs or even GPUs for certain AI workloads.

The Spanish company Multiverse Computing is also making strides in this area, leveraging quantum computing to compress AI models to 10% of their original size. This technology, backed by a €67 million investment from the Spanish government [13], has the potential to cut data center energy consumption in half.

Sustainable Data Center Design

The architecture and operation of data centers themselves are critical. Companies are investing in advanced cooling systems, energy management platforms, and optimized server layouts to improve energy efficiency. Integrating renewable energy sources like wind and solar into these facilities is becoming increasingly common, though challenges with consistency remain.

For AI workloads that require uninterrupted power, hybrid models combining renewables with nuclear or other baseload sources are seen as the most reliable approach.

Biological and Alternative Computing

Some innovators are even looking beyond silicon. Cortical Labs [11, 18], an Australian startup, developed the CL1, a biological computer using lab-grown human neurons. These systems can process information with a fraction of the energy used by GPUs, offering a glimpse into a radically different—and greener—future of computing.

The neuron is self programming, infinitely flexible, and the result of four billion years of evolution. — The neuron is self-programming, infinitely flexible, and the result of four billion years of evolution. Source [18]

Policy, Transparency, and Industry Standards

As the environmental impact of artificial intelligence becomes more evident, policy frameworks, industry standards, and transparency measures are emerging [15] as essential tools to guide the responsible development of AI technologies. While private companies are driving much of the innovation, governments and regulatory bodies must set the boundaries that ensure sustainability, accountability, and equitable access to resources.

The Need for Regulatory Oversight

Historically, energy policy has been shaped by governments, particularly when it involves nuclear power. But the recent push by tech giants into the energy sector is shifting the balance. When companies like Microsoft, Google, and Amazon begin negotiating directly with utilities, funding new reactor construction, or reviving dormant nuclear facilities, the traditional roles of public and private actors become blurred.

This shift underscores the urgent need for updated regulatory frameworks that can accommodate the evolving nature of energy consumption in the AI era. Current regulations often lag behind technological progress. For example, existing nuclear licensing procedures in the U.S. were designed for large-scale reactors built over decades—not for the rapid deployment of modular, private-sector-funded SMRs.

The legislation signed by President Biden in March 2024 [5], which streamlines the permitting process for nuclear facilities, represents an important step forward. However, long-term governance will require more comprehensive approaches that address not only energy generation but also environmental monitoring, cybersecurity risks, and equitable resource distribution.

Mandatory Energy Transparency

A key pillar of responsible AI development is transparency in energy usage. At present, companies are not uniformly required to disclose the energy costs of training or operating AI models, nor the source of that energy. This lack of transparency makes it difficult to assess the true environmental footprint of AI technologies.

There is growing momentum behind initiatives that would mandate energy disclosures for AI models, similar to nutrition labels on food or energy ratings on appliances. These could include:

Total electricity consumption during model training.
Carbon emissions associated with that consumption.
The share of energy sourced from renewables, nuclear, or fossil fuels.
Water usage for data center cooling.

Such measures would help consumers, regulators, and researchers better understand the trade-offs involved in AI development—and allow governments to benchmark progress toward sustainability goals.

Global Standards and Cooperation

Given that AI development is a global enterprise, there is also a pressing need for international standards around sustainable AI practices. Organizations such as the International Telecommunication Union (ITU), the International Energy Agency (IEA), and the OECD [16] have begun convening working groups to define best practices for green AI development.

These efforts could eventually lead to global sustainability certifications for AI systems and data centers, much like LEED certification in architecture or ISO standards in manufacturing. Companies that meet these criteria could be eligible for tax incentives or preferential access to public cloud contracts.

Additionally, multilateral cooperation will be critical in managing nuclear waste, setting safety protocols for new reactor technologies, and ensuring that the benefits of AI-powered infrastructure are distributed equitably across nations—not just concentrated in wealthier, tech-dominant economies.

Incentivizing Sustainable Innovation

Finally, public policy can play a proactive role by incentivizing green innovation. This includes:

Research grants for low-energy AI model development.
Subsidies for building or upgrading energy-efficient data centers.
Carbon credits or tax deductions for companies using carbon-free energy sources.
Procurement policies that prioritize AI services from companies with strong environmental records.

These tools can help create market advantages for sustainability and accelerate the adoption of best practices across the tech industry.

The Role of the “Energy New Deal”

Perhaps the most ambitious proposal on the table is the so-called “Energy New Deal”, backed by Microsoft, OpenAI, and Nvidia [9]. This proposal calls on the U.S. government to commit hundreds of billions of dollars in public investment to modernize the country’s energy infrastructure.

The plan would fund the construction of new nuclear, solar, and wind plants; improve grid resilience; and build AI-optimized data centers powered by clean energy. If realized, it could serve as a model for other countries looking to align AI innovation with climate policy—effectively creating a blueprint for the next generation of industrial policy in the digital era.

The Strategic Stakes of AI and Energy

At the heart of these developments is a fundamental realization: energy is the new bottleneck in AI. As competition in the AI space heats up, access to scalable, carbon-free electricity is becoming a critical differentiator. Companies that can ensure reliable power for their data infrastructure will have a decisive edge in training larger models, serving more users, and launching new products.

This dynamic also raises broader societal questions. Should private corporations be allowed to shape the direction of national energy policy? How do we ensure that innovation does not eclipse safety and environmental responsibility? And how can we guarantee that the benefits of AI are not outweighed by its ecological footprint?

Conclusion: Toward a Smarter, Cleaner Future

The convergence of AI and nuclear energy marks a turning point. What began as an effort to solve an engineering problem—how to power AI—has grown into a sweeping transformation of the global energy economy.

By investing directly in nuclear power, Amazon, Google, and Microsoft are not just meeting their own operational needs; they are redefining the infrastructure of the digital age. At the same time, innovations in algorithms, hardware, and data center design are helping reduce the overall environmental impact of AI.

The road ahead will require continued innovation, regulatory foresight, and international collaboration. But if done right, the fusion of artificial intelligence and clean energy can deliver on its promise: a smarter, faster, and more sustainable future.