China Eyes Nuclear, Hydrogen Power for AI Data Centers

China Launches Bold Plan to Power AI With Nuclear and Hydrogen Energy

China's National Energy Administration (NEA), along with 3 other top government agencies, has issued a landmark policy framework exploring the direct connection of nuclear power and hydrogen energy to AI computing facilities. The 'Action Plan for Promoting Mutual Empowerment Between AI and Energy,' jointly released with the National Development and Reform Commission (NDRC), the Ministry of Industry and Information Technology (MIIT), and the National Data Administration, signals Beijing's aggressive push to solve one of the AI industry's most pressing challenges: the insatiable demand for electricity.

The directive arrives as global AI infrastructure faces an unprecedented energy crisis. Data centers worldwide are projected to consume over 1,000 terawatt-hours of electricity annually by 2026, according to the International Energy Agency — roughly equivalent to Japan's entire national consumption.

Key Takeaways From the Policy

• Direct energy connection: China will explore linking nuclear and hydrogen power directly to AI computing facilities, bypassing traditional grid distribution
• Grid-forming storage: Data centers are encouraged to deploy grid-forming energy storage systems for enhanced power stability
• New standards: Authorities will establish planning and construction standards for energy supply to computing facilities
• Multi-source power: The policy promotes diversified electricity supply capabilities for AI infrastructure
• System support: Computing facilities should actively support the broader power grid, not just consume from it
• Customized approach: Energy solutions will be tailored based on facility scale, voltage requirements, renewable penetration rates, and business types

Why Direct Nuclear-to-Data-Center Connections Matter

The concept of direct energy supply — connecting power generation directly to computing facilities without routing through the public grid — represents a paradigm shift in how nations think about AI infrastructure. Traditional data centers draw power from the electrical grid, competing with residential, commercial, and industrial users for capacity.

Direct connection eliminates transmission losses, which typically account for 5% to 8% of generated electricity. For a large-scale AI training cluster consuming 100 megawatts or more, this efficiency gain translates to millions of dollars in annual savings.

Nuclear power offers a particularly compelling proposition for AI workloads. Unlike solar and wind, nuclear delivers baseload power — consistent, 24/7 electricity generation with capacity factors exceeding 90%. AI training runs, which can last weeks or months on thousands of GPUs, demand exactly this kind of unwavering reliability.

China Follows a Global Trend — But With a Twist

China's policy mirrors moves already underway in the United States. Microsoft signed a landmark deal in September 2024 to restart the Three Mile Island nuclear plant in Pennsylvania, securing 835 megawatts of carbon-free power for its AI operations. Amazon Web Services purchased a $650 million nuclear-powered data center campus from Talen Energy in Pennsylvania. Google announced agreements with Kairos Power to deploy small modular reactors (SMRs) to power its AI infrastructure by 2030.

However, China's approach differs in one critical aspect: it is a top-down, government-coordinated strategy rather than a collection of individual corporate deals. By issuing a unified policy framework across 4 agencies, Beijing can align energy planning, grid infrastructure, industrial policy, and data governance simultaneously.

• Microsoft's Three Mile Island deal: 835 MW, expected online by 2028
• Amazon's Susquehanna campus: 960 MW nuclear-adjacent facility
• Google's Kairos SMR plan: First reactor targeted for 2030
• Oracle's planned nuclear campus: 3 SMRs for a 1-gigawatt data center
• China's new policy: National framework covering all AI computing facilities

The scale of China's ambition is notable. Rather than negotiating one-off deals, the government is building an entire regulatory and planning ecosystem to connect advanced energy sources to AI infrastructure nationwide.

Hydrogen Energy: The Wild Card in China's AI Power Mix

While nuclear power gets most of the headlines, the inclusion of hydrogen energy in the policy is equally significant — and arguably more innovative. Hydrogen fuel cells can provide backup and supplementary power to data centers, potentially replacing diesel generators that currently serve as emergency backups at most facilities worldwide.

China is already the world's largest hydrogen producer, generating approximately 33 million metric tons annually. Most of this is currently 'grey hydrogen' produced from fossil fuels, but the country has ambitious plans to scale up 'green hydrogen' production from renewable-powered electrolysis.

For AI data centers, hydrogen offers several advantages over traditional backup power:

It produces zero carbon emissions when used in fuel cells. It can be stored for extended periods, unlike battery systems that discharge over time. It can scale to provide sustained power for hours or even days during grid outages. And it can be produced on-site at facilities with access to water and renewable electricity.

The challenge remains cost. Green hydrogen currently costs between $4 and $7 per kilogram to produce, compared to roughly $1 to $2 for grey hydrogen. But as electrolyzer costs fall and renewable energy prices continue declining, the economics are improving rapidly.

Grid-Forming Storage: Making Data Centers Part of the Solution

One of the most forward-thinking elements of the policy is the encouragement for AI computing facilities to deploy grid-forming energy storage systems. Traditional data centers are purely consumers of electricity — they take from the grid but give nothing back.

Grid-forming storage changes this dynamic entirely. These advanced battery systems can actively stabilize the power grid by providing frequency regulation, voltage support, and inertia — services traditionally provided only by large spinning generators at power plants.

This approach transforms data centers from grid liabilities into grid assets. As renewable energy penetration increases in China — the country added a record 217 gigawatts of solar capacity in 2023 alone — the grid needs more sources of stability. AI data centers equipped with grid-forming storage could help fill this gap.

The financial implications are also significant. Data center operators deploying grid-forming storage could potentially earn revenue by selling grid services, offsetting some of their enormous electricity costs. In markets like Texas and parts of Europe, grid services can command premium prices during periods of high demand or system stress.

What This Means for the Global AI Race

China's policy has profound implications for the global AI infrastructure competition. Energy availability is increasingly becoming the primary bottleneck for AI development — not chip supply, not talent, not capital.

Countries and companies that solve the energy problem first will have a decisive advantage in training the next generation of AI models. A single frontier AI training run now consumes as much electricity as a small city uses in a year. OpenAI's rumored next-generation models and xAI's Memphis supercluster — reportedly consuming 150 megawatts — illustrate the trajectory.

By creating a national framework for diverse energy supply to AI facilities, China is attempting to ensure that energy constraints do not slow its AI ambitions. The country already operates 55 nuclear reactors with 23 more under construction — the world's largest nuclear construction pipeline.

For Western policymakers, the message is clear: AI energy policy cannot remain an afterthought or be left entirely to market forces. The U.S. Department of Energy has begun exploring similar concepts, but a comprehensive, multi-agency framework comparable to China's has yet to emerge.

Looking Ahead: Implementation Challenges and Timeline

Despite the policy's ambition, significant hurdles remain before nuclear reactors and hydrogen systems are directly powering China's AI data centers at scale.

Regulatory complexity tops the list. Nuclear facilities operate under strict safety protocols that may conflict with the fast-moving timelines of AI infrastructure deployment. Establishing direct connections requires new safety standards, liability frameworks, and operational procedures that do not yet exist.

Technical challenges are also substantial. Matching the output characteristics of nuclear plants or hydrogen fuel cells to the precise power quality requirements of modern GPU clusters — which demand extremely stable voltage and frequency — requires sophisticated power electronics and control systems.

Realistic timelines suggest initial pilot projects could emerge within 2 to 3 years, with broader deployment following by 2028 to 2030. China's advantage in rapid infrastructure development could accelerate this timeline compared to Western counterparts.

The policy also raises questions about international supply chains. As China builds out nuclear-powered AI infrastructure, demand for uranium, specialized steel, and advanced power electronics will increase, potentially creating new resource competition with Western nations pursuing similar strategies.

What is clear is that the convergence of AI and energy policy is no longer theoretical — it is becoming a central pillar of national technology strategy on both sides of the Pacific. The race to power artificial intelligence may ultimately prove as consequential as the race to build it.