The rapid expansion of artificial intelligence is on a collision course with a fundamental physical limit: the availability of electricity. A significant gap is emerging between the power required by America's data centers and what the nation's energy grid can supply, creating a potential bottleneck for the entire AI industry.
According to analysis from Morgan Stanley, U.S. data centers are projected to demand an additional 57 gigawatts (GW) of power between 2025 and 2028. However, with only 6 GW of new power generation currently under construction and an estimated 15 GW in spare grid capacity, the nation faces a staggering 36 GW shortfall.
Key Takeaways
- A projected 36-gigawatt power shortfall threatens to slow the growth of the U.S. AI industry by 2028.
- The gap stems from the massive electricity demands of new data centers outpacing the construction of new power generation facilities.
- This energy bottleneck is shifting investment focus from AI chipmakers to companies that can solve the power crisis, including energy producers and infrastructure builders.
- Small modular reactors (SMRs) are emerging as a key potential solution due to their rapid deployment time compared to traditional power plants.
The Emerging Energy Bottleneck
For years, the conversation around AI's growth has centered on computing power, specifically the production of advanced graphics processing units (GPUs). Companies like Nvidia have become household names by supplying the chips that train and run complex AI models. However, a new constraint is now taking center stage: raw energy.
Data centers, the physical backbone of the AI revolution, are incredibly power-hungry. Each query, model training session, and data analysis task consumes a significant amount of electricity. Without a reliable and massive supply of power, even the most advanced AI hardware is useless.
What is a Gigawatt?
One gigawatt (GW) is a unit of power equal to one billion watts. To put the 36 GW shortfall in perspective, it is roughly equivalent to the electricity required to power more than 30 full-scale nuclear power plants or the entire state of California.
The projected 36 GW deficit represents a critical challenge. If this gap is not closed, some of the multi-trillion dollar data center projects planned by major tech companies may be delayed or unable to operate at full capacity. This shifts the investment landscape, suggesting that the next wave of winners in the AI boom may not be software developers, but the companies that provide the essential kilowatts.
A Multi-Faceted Approach to Powering AI
Addressing a 36 GW shortfall by 2028 requires a diverse and aggressive strategy to boost energy supply. Analysts believe the solution will come from a combination of traditional, renewable, and emerging technologies. No single source can bridge the gap alone.
Immediate and Transitional Solutions
To meet near-term demand, natural gas power plants are seen as a likely first response. These facilities can be built relatively quickly, typically within two to four years, and provide consistent baseload power. It is estimated that natural gas could contribute approximately 12 GW toward closing the deficit. Companies like Vistra and Constellation are positioned to play a role in this expansion.
Simultaneously, the buildout of renewables and energy storage will continue. Major technology companies are already signing large-scale power purchase agreements for solar and wind energy. However, the intermittent nature of these sources presents a challenge. A 20 GW solar installation may only provide about 5 GW of reliable, firm power. Battery storage can help, but it adds significant cost. This sector is expected to contribute an effective 8 GW.
The Role of Data Center Efficiency
Alongside building new power sources, hyperscalers are investing heavily in efficiency. Innovations in cooling systems, server utilization, and chip design can reduce overall energy demand. While not a complete solution, these improvements could account for avoiding 2 to 4 GW of new power demand, providing crucial breathing room for the grid.
The Nuclear Option: Small Modular Reactors
Perhaps the most transformative solution on the horizon is the deployment of small modular reactors (SMRs) and microreactors. Unlike massive, traditional nuclear plants that can take over a decade to permit and build, SMRs offer a faster and more flexible alternative.
These advanced reactors are designed to be manufactured in a factory and transported to a site for installation. This approach drastically cuts down construction time to just two to five years, a timeline that aligns perfectly with the urgent needs of the AI industry.
"SMRs are tailor-made for this moment. Their ability to be deployed quickly and sited directly next to data centers bypasses many of the traditional grid bottlenecks," an industry analyst noted. "They provide the 24/7, carbon-free power that data centers require."
Key advantages of SMR technology include:
- Speed: Deployment in 2-5 years versus 10-15 years for large-scale reactors.
- Flexibility: Can be located adjacent to data centers, reducing reliance on long-distance transmission lines.
- Reliability: Provide constant, weather-independent power, unlike solar or wind.
- Government Support: The U.S. Department of Defense and Department of Energy are actively backing SMR development, with projects like the Army's Project Janus already underway to deploy microreactors on military bases.
Companies like Oklo, backed by tech industry leader Sam Altman, and BWX Technologies, which already builds reactors for NASA, are at the forefront of this movement. Analysts project SMRs could fill the largest portion of the power gap, potentially contributing 12 to 15 GW by 2028.
Investment Implications of the Power Crunch
This impending energy shortage fundamentally alters the investment thesis for the AI megatrend. The focus is expanding beyond the creators of AI models to the enablers of the underlying infrastructure. Electricity has effectively become the new strategic commodity for the 21st century, much like oil was for the 20th.
The companies poised to benefit fall into several key categories:
- Nuclear Innovators: SMR and microreactor developers like Oklo (OKLO) and BWXT (BWXT), along with uranium suppliers such as Cameco (CCJ).
- Grid and Infrastructure Builders: Companies like Quanta Services (PWR) and Eaton (ETN) that are essential for modernizing and expanding the electrical grid.
- Power Producers: Utilities and independent power producers (IPPs), particularly those with natural gas and nuclear assets like Constellation (CEG) and Vistra (VST).
- Data Center REITs: Real estate investment trusts like Equinix (EQIX) and Digital Realty (DLR) that own and operate data centers will see increased demand for powered facilities.
The AI boom is not just about code and algorithms; it is about kilowatts and infrastructure. The challenge of the 36 GW power shortfall is immense, but it also creates a significant investment opportunity in the companies that will build the energy foundation for the future of technology.
