For the first time in decades, the US is experiencing a sharp demand in growth for electricity. The exact projections vary, including DOE estimating 20% in the next decade and Grid Strategy forecasting 16% peak load growth in the next five years. By all estimates, after decades of flat growth and minimal activity, the US power market is entering a new age.

Three trends have been the primary drivers: (1) the race for AI and the training of large foundation models; (2) manufacturing onshoring leading to the expansion of industrial plants; and (3) continued electrification in transportation and buildings. Being the newest force, AI has been the biggest factor in new energy demand, with the most uncertainty.

A Unique Challenge

The US went through a seemingly similar period in the last century when electronic appliances were first introduced to households. From the 1950s-1960s, electricity demand grew at an annual rate of 8%. Another decade of grid expansion followed suit until supply finally met demand.

Yet, with AI driving the rise in demand, our modern situation is unique.

  • High electricity density: Most AI-related electricity demand comes from GPU training, which is much more energy-dense than traditional CPU training. The latest AI data centers being built are almost all above 100 MW, if not more, matching only with a giant industrial plant. Even trickier, while industrial plants consume energy mostly during the day, these AI data centers run 24/7. As a result, while the appliance-driven load growth back in the ’60s came in smaller chunks and was more spread out, every new AI data center creates a big shock and requires meaningful upgrades for the system to absorb.
  • High uncertainty: To make the problem more difficult, most new AI data centers planned are built on the premise that AI will change the world. Big tech is investing a seemingly infinite amount of money into AI-related infrastructure, strategically (or speculatively?) to train the latest and greatest machine learning models to lead the AI revolution. But what does the AI revolution look like, and how much computing power and electricity does it eventually need?

There’s additional challenge from the supply side with the ongoing clean energy transition driven by the improving economics of renewable energy and the increasing awareness of climate change. Yet renewable energy alone cannot meet all the demands due to its intermittent nature. With the rest of the less mature clean energy technology still evolving, tackling the change this time requires not just scaling the existing system but creating a new system with novel technology and infrastructure.

Potential Solutions

Most of the promising solutions to meet such ‘power thirst’ fall into two categories:

First, by combining renewable energy with energy storage. The excess energy generated by the intermittent renewable energy during the day will be stored and dispatched at night.

Second, incorporating ‘baseload’ power, i.e., power generated by an energy source that can run regardless of the hours in most conditions. Historically, this role is played by fossil fuel plants. In the clean energy domain, there are two main contenders: hydro and nuclear. Hydropower is limited by water resource availability, making nuclear power the main focus for this category.

Many technologists seem to be increasingly betting on nuclear power as the holy grail of 100% clean energy. Billions of grants and venture money have been invested to develop new technologies, including fission, fusion, and advanced modular reactors such as SMR.

Even the existing nuclear technology is resurrecting at the industry level. Early this year, Amazon announced a $52B data center expansion plan with electricity supplied by clean energy and nuclear. This quickly went into action with Amazon purchasing a nearly 1GW nuclear plant from Talen to power its data centers. Another industry headline came from Microsoft, who entered a 20-year contract with Constellation Energy to power its data centers through the Three Miles nuclear plant. One of the recent releases came from Meta, who announced a $10B investment to build a 2GW data center campus in in Louisiana, partially powered by nuclear.

Reality Check

So, is the age of nuclear energy coming soon? Not quite. Most of the advanced nuclear technology is years, if not decades, away from commercialization. A great many barriers exist even when deploying the existing nuclear technology.

First, all these projects, despite their media attention, will need regulators’ blessing to move forward, ranging from permits to interconnection approval. The Talen-Amazon deal hit a pause this year after FERC rejected the nuclear plant interconnection request. While negotiations are still ongoing, the surprising ruling is a reality check of the regulatory complexity.

In addition, given the size of the projects and the constrained labor and manufacturing capacity in the US, I wouldn’t be surprised if these projects run into meaningful delays. Southern Co. made the last attempt to build a nuclear plant in the US to expand Plant Vogtle. The project was delayed by over 7 years and only recently came online. While none of these new data center projects require building a new nuclear plant, bringing the old plant back is still a huge undertaking requiring labor training, equipment procurement, fuel sourcing, etc. Every component runs the risk of delay, and the lag can quickly stack up.

Nuclear plants were almost all decommissioned in the ’90s due to public perception and safety concerns. When bringing them back, these issues have to be properly addressed. There’s no shortage of clean energy projects being canceled due to community pushback. Getting the community onboard is another hurdle these projects have to clear.

With all these said, I’m confident that nuclear energy will play a pivotal role in the long-term energy transition. Just not yet. In the near term, new demands are still going to be mostly met through renewable energy plus storage solutions for its mature technology and better economics. The continued technology enhancement and infrastructure expansion in energy storage further sustain the traction.

Looking Forward

Conservatively, I estimate around $250B investments needed to build 128GW capacity of power, assuming all the demand is met by deploying the most mature solar and battery technology at around 1.85$/w capex and 54.6$/kw-y Opex based on NREL’s cost forecasts for the next five years. And I’m not even accounting for R&D or additional infrastructure spending. If we were to introduce new technology to our near-term energy portfolio, the number could easily blow up.

With a problem of this size and complexity, we need imagination and pragmatism. We need to strike a balance between scaling working solutions in the near term and investing in long-shot ideas for the long term. I’m encouraged by the latest $20B deal from Google to invest in Intersect Power. The deal focuses on expanding Intersect’s pipeline in clean energy, providing energy to Google’s data center energy park by deploying mature solar and battery technology. It may not be the most exciting deal of the year. But it will be effective.

The road ahead to meet the rapidly rising ‘power thirst’ will see no shortage of challenges, and one has to pick their battles. Advancing and deploying mature technologies specifically renewable energy and storage solutions should remain front and center in then near term.