The electricity crisis facing AI data centers has created an unusual winner: modular gas power plants that can be ordered, built, and connected to the grid in just 12 to 18 months. While nuclear energy remains the cleaner long-term solution, it won't arrive fast enough to meet the surging demand from artificial intelligence infrastructure. This timing mismatch is reshaping how utilities and tech companies think about powering the AI revolution.

The problem is stark and quantifiable. In December 2025, PJM Interconnection, the grid operator serving 13 states and Washington, D.C., with about 67 million people, held its annual capacity auction and failed to secure enough power for the first time in its history. The shortfall reached roughly 6,600 megawatts, equivalent to six large nuclear plants' worth of generation. PJM expects peak demand to grow by 32 gigawatts between 2024 and 2030, with all but two of those gigawatts coming from data centers.

Why Is Speed to Power Becoming the Real Bottleneck?

The electricity demand from AI data centers is growing faster than the grid infrastructure can handle. According to 451 Research, part of S&P Global, U.S. data center grid demand is projected to reach 75.8 gigawatts in 2026, climbing to 134.4 gigawatts by 2030. This explosive growth has created what the industry now calls "speed to power," a challenge where demand arrives at a pace the grid was never designed to match. The bottleneck isn't money; it's how quickly new generation can physically connect to the system.

Rolls-Royce Power Systems, the division that builds engines for ships, trains, and backup generators, launched a solution at the E-world energy trade fair in Essen, Germany, in February 2026. The company introduced turnkey modular gas engine power plants configured in 10, 20, or 30-megawatt blocks. A single plant can range from five megawatts to several hundred, depending on customer needs, and can be ordered, factory-tested, and connected to the grid within 12 to 18 months.

"Utilities and data centers around the world rely on our solutions," said Tobias Ostermaier, president of Stationary Power Solutions at Rolls-Royce Power Systems, pointing to more than 17 gigawatts of mtu generating capacity already installed.

Tobias Ostermaier, President of Stationary Power Solutions at Rolls-Royce Power Systems

The modular approach offers a critical advantage over traditional power plants. Instead of building one massive central facility, operators can order preconfigured blocks, stack as many as needed, and connect them to the grid. These gas engine plants, not gas turbines, can be switched on and off individually and held at their efficiency sweet spot, making them ideal for absorbing the swings in wind and solar power.

How Does This Compare to Nuclear Energy?

Nuclear power remains the obvious clean answer to an electricity problem of this scale. Rolls-Royce itself is building small modular reactors (SMRs), and in April 2026, the UK government approved the company's first three SMRs at Wylfa in North Wales, each rated around 470 megawatts, together enough for roughly three million homes. However, those reactors are not expected to feed the grid until the mid-2030s, with the final investment decision not due until around the turn of the decade.

A conventional, full-size nuclear plant takes well over a decade to build. Even the gas plants that PJM fast-tracked to address its own capacity shortfall are mostly not due online until 2030 or later. For a hyperscaler that needs megawatts before its lease is half over, "available in 2027" beats "carbon-free in 2035" every time.

The nuclear industry is responding to AI demand, though. Three original equipment manufacturers in the nuclear energy space are positioning themselves for growth: Mirion Technologies, BWX Technologies, and GE Vernova. Mirion is working with small modular reactor developers to solve nuclear measurement, safety, and security challenges. BWX Technologies secured contracts valued at more than 1.4 billion dollars in May 2026 supporting the U.S. Naval Nuclear Propulsion Program, strengthening its position in the nuclear industry. GE Vernova stands to gain from global nuclear power momentum, particularly through its strong position in producing SMRs.

What Makes These Modular Gas Plants "Hydrogen-Ready"?

Rolls-Royce frames its modular gas plants as a bridge solution, not a permanent answer. The hardware is built so that when hydrogen becomes available, the engines can be converted to run on it. The company had this hydrogen-readiness independently certified by TÜV Süd for its mtu Series 4000 gas engines in 2024. However, it's important to be precise about what this means: the plants do not run on hydrogen today. They run on natural gas, and the conversion option is penciled in for a future when green hydrogen is cheap and plentiful enough to feed a 30-megawatt engine.

The first mtu engine actually designed to burn 100 percent hydrogen is a separate, much smaller combined-heat-and-power unit of around one megawatt, slated for a container port in Duisburg. Green hydrogen at the volume and price a 30-megawatt engine would need is not something you can buy today.

Steps to Understanding the Power Infrastructure Challenge

• Demand Growth Rate: Data center electricity demand is projected to more than double from 75.8 gigawatts in 2026 to 134.4 gigawatts by 2030, creating unprecedented pressure on grid infrastructure.
• Timeline Mismatch: Modular gas plants connect in 12 to 18 months, while conventional nuclear plants take over a decade and small modular reactors won't arrive until the mid-2030s.
• Grid Operator Concerns: PJM Interconnection failed to secure adequate capacity for the first time in its history, coming up 6,600 megawatts short of its reliability target for 2027.
• Hydrogen Transition Path: Current modular gas plants are "hydrogen-ready" but cannot run on hydrogen today; the conversion option exists for future deployment when green hydrogen becomes economically viable.

What Are the Risks and Limitations of This Strategy?

The modular gas approach solves the speed problem but creates new challenges. The plants still burn natural gas, so the climate case rests entirely on that hydrogen conversion actually happening. The 12-to-18-month timeline also starts at the order, not the announcement, which means manufacturing becomes the real bottleneck. If every utility and data center that's short on power decides modular gas is the answer simultaneously, that queue gets longer fast.

The demand picture itself is shakier than headlines suggest. One analysis from the Information Technology and Innovation Foundation notes that new data center deals fell more than 40 percent between the third and fourth quarters of 2025, that only about a third of announced construction is actually being built, and that OpenAI's flagship Stargate project in Texas has stalled. The 2027 crunch PJM is staring at is real, but what demand looks like in 2032 is anyone's guess.

Beyond the power supply question, communities are raising concerns about data centers themselves. A Carnegie Endowment survey found that a plurality of Californians think AI data centers are mostly bad for the quality of life of people living nearby. Noise is the number one reason for opposition in cases where projects were ultimately canceled. As a climate concern, pollution from data center power generation is significant, particularly when less efficient turbines are used due to supply shortages.

The broader issue is what happens when today's data centers become stranded assets. Technological advances and land use constraints are already making the next generation of data centers smaller and more efficient. When current facilities are abandoned through bankruptcy or consolidation, they could dot the landscape like empty shopping malls and warehouses, creating a repurposing challenge that communities may want to address ahead of time.

For now, Rolls-Royce holds both ends of the timeline. The company's modular gas plants offer a bridge that can run continuously until a grid connection or another source like nuclear comes online, then drop back to backup duty. This dual approach reflects the reality of the AI energy crisis: the perfect solution doesn't exist yet, so companies are building with what they can get today and planning for what they hope arrives tomorrow.