Venture capitalists have placed increasingly bigger bets on AI startups, investing over half a trillion dollars into the sector over the last five years.

But these days, the smartest AI investment might be in energy, according to a report by Sightline Climate. Researchers found that up to 50% of data center projects that have been announced might be delayed. One of the biggest culprits is access to power. 

Of the 190 gigawatts worth of data centers the company is tracking, only 5 gigawatts are under construction. About 6 gigawatts of data center projects in Sightline’s database came online last year. A far larger percentage — about 36% — saw their timelines slip in 2025. The delays may eventually trickle down and affect large enterprises and other companies that use AI for their businesses. 

That supply-demand squeeze is an opportunity for investors. Here’s why. 

Big tech companies like Google and Meta have devoted large parts of their balance sheets to develop solar, wind, and nuclear projects. These companies are also supporting emerging technologies like Form Energy’s 100-hour battery through direct investments and working with utilities to accelerate their adoption.

Dozens of startups are pursuing technologies that tackle the power problem. For instance, Amperesand, DG Matrix, and Heron Power are developing new power conversion technologies, while companies like Camus, GridBeyond, and Texture are building software that can manage the flow of electrons.

Power remains one of the most significant constraints for data centers, a shortfall that isn’t likely to change anytime soon. AI is expected to drive data center power consumption up 175% by 2030, according to Goldman Sachs. 

These shortages on the grid are unprecedented in modern times, and they’ve been driving up electricity prices around the country. That has pushed many tech companies to explore alternative ways of powering their data centers. (The Trump administration, sensing a looming political crisis, is urging tech companies to build their own power source, pay higher rates, or both. Most had already made plans to do so, of course.)

Grid alternatives

Amazon, Google, Oracle, and other large tech companies have been working to minimize their dependence on the grid. Several data centers are being planned using on-site power or a hybrid approach that blends on-site power with a grid connection. 

The biggest data centers are leading the charge. Less than a quarter of projects that have identified a power source will use on-site or hybrid; together they represent 44% of total capacity.

The shift has been driven in part by shortages of power generation equipment — namely gas turbines — and an antiquated grid. That’s opened a path for alternative energy sources.

Google’s latest deal to power a new data center in Minnesota shows one approach to tackling the problem. The company will blend wind and solar with a massive 30 gigawatt-hour battery from Form Energy. Google also worked with Xcel Energy to devise a new rate structure that it says will help encourage the adoption of new technologies in the utility’s planning process.

Form Energy’s battery isn’t the only example. Grid-scale batteries are poised to take a big bite out of the power market. By the end of this year, the U.S. should have nearly 65 gigawatts of battery storage capacity, according to the U.S. Energy Information Administration. Like many of its peers, Form Energy is looking to capitalize on the momentum by raising a $500 million round in advance of an eventual IPO. 

Underrated tech

Energy supplies are only part of the story. Once the power hits the grid or the data center, it needs to be managed, a task that mostly falls on the humble transformer. 

Most of today’s transformers use massive blocks of iron wrapped in copper wire, a technology that is about 140 years old. It’s reliable, but it’s becoming far too bulky as data center power demands ramp up. By the time server racks hit 1 megawatt in power density, the power equipment needed to run them will occupy twice as much space as the rack itself, one expert told TechCrunch. 

It’s why investors have been flocking to back solid-state transformer startups recently, which are hoping silicon-based power electronics can supplant the ancient iron-and-copper tech. They’re more expensive than existing transformers, but they are also flexible enough to replace several pieces of equipment in a data center, which should make them cost competitive. 

Altogether, the scale of investments in battery and transformer companies has been much smaller than some of the blockbuster rounds we’ve seen in the AI industry. 

That’s not a bad thing — those rounds are more tractable for investors. Plus, as the world electrifies everything from transportation to heavy industry, the need for power is only going to grow, giving investors a hedge against an AI bust. Maybe the best AI investment isn’t in AI at all.