Energy IPO sparks a frenzy, raising an unprecedented $12.6 billion in demand—setting a new high this century; AI power-shortage anxiety becomes Wall Street’s new battleground

Power companies’ IPO fundraising in the first half of this year has hit the highest level since the 1999 dot-com bubble—driven by growing concerns about power shortages at AI data centers, which is pushing Wall Street’s hot money away from chip stocks and toward power plants.
(Background: Bloomberg calls “AI a long-term demand”; Caterpillar, a power equipment company, jumped for eight straight days to a record high.)
(Additional context: Goldman Sachs: US stock IPOs are vacuuming up $120 billion “without yet getting bubble-fied”! The crowding-out effect from AI mega-giants has caused a widespread shutdown of crypto companies’ IPOs.)

Table of contents

Toggle

• $12.6 billion in the first half—power stocks become hot commodities
• The logic behind hot money from chips shifting into power
• Behind the estimated 39% electricity-use growth is a reshuffling of capital

Chips determine what AI can do, while power determines how much AI can do—and the latter is quickly becoming the most realistic bottleneck in the entire competition. Capital has obviously already caught the scent: in the first half of this year, energy companies rushed into the IPO market at a pace that set the fastest record of this century.

$12.6 billion in the first half—power stocks become hot commodities

According to data provider Dealogic, in the first half of this year, global energy-sector IPO fundraising totaled $12.6 billion, the highest half-year record since the peak of the 1999 dot-com bubble, and also the highest for any comparable period in history—far outstripping the $4.3 billion total for all of 2025.

The backdrop to this wave of capital is actually not hard to understand: AI data centers are consuming far more electricity than the outside world has imagined, and obtaining enough power has become the biggest gate for this multi-trillion-dollar AI investment boom. What’s more troublesome is that building a data center may only take one or two years, but building a power plant—or even just upgrading the grid so it’s in place—can easily take five to ten years.

The time gap is the real tightening curse of this AI boom.

The market’s response has been direct. This week, ETF provider GMO just launched a “Power Infrastructure ETF,” targeting returns from power generation, grid, and electrification-related stocks. Nuclear startup Standard Nuclear is also expected to list in the United States a bit later in July. Combined with Dealogic’s statistics, these moves add up to the same picture: capital is lining up to buy electricity.

This surge isn’t only reflected in the numbers of listings. Geothermal energy company Fervo Energy raised roughly $1.9 billion in its IPO this year; on its first trading day, its share price jumped more than 33%. It’s viewed as the largest clean-energy IPO in history. Market statistics show that in 2026, at least 10 power-infrastructure and clean-technology companies plan to go public. In the first five months alone, the value of US utility mergers and acquisitions reached $203.6 billion, already surpassing the total for all of 2025.

The logic behind hot money from chips shifting into power

RBC clean energy analyst Chris Dendrinos points to the key behind this round of capital rotation: “At first, investors bought AI concept stocks like Nvidia, and only later did they think that every chip needs electricity to operate.” Behind this is a simple but underestimated logic: chips determine what AI can do, while electricity determines how much AI can do.

This is the so-called shoveler logic. In simple terms, during the nineteenth-century US gold rush, the people who often profited steadily weren’t the miners who struck gold, but those who sold shovels and cowboy jeans to the miners. Put into today’s AI game, no one can guarantee which AI company will win, but it’s almost certain that no matter who wins, they will need more electricity.

Buying power stocks is equivalent to not having to guess which AI company will survive to the end—you’re betting on the entire game itself. As long as you can confirm that the whole game needs more electricity, the tolerance for error of this kind of investment is inherently higher than betting on a single chip company.

The scale of electricity demand is far more enormous than most people think. A typical AI data center uses about 876,000 million kWh of electricity per year. Put simply, this is roughly equivalent to the combined total household electricity consumption of an entire city—such as Glasgow in the UK or Salt Lake City in the US—over one year. One data center consumes the power of an entire city.

And this electricity usage can’t be interrupted. AI training and inference require baseload power. Put simply, it’s not wind and solar that depend on the weather and come and go, but power sources like nuclear and natural gas that can provide stable electricity 24 hours a day—only that can support the data center’s never-stop operating demand. In other words, at its core, this wave of electricity-related IPO hype is the market making up for an infrastructure lesson the AI industry learned too late.

Behind the estimated 39% electricity-use growth is a reshuffling of capital

Consulting firm ICF projects that US electricity demand will grow 39% between 2026 and 2035, with the main driver coming from a surge in electricity demand from data centers. Manish Kabra, head of US equity strategy at Société Générale, said bluntly: “Expansion of power capacity, the return of US manufacturing, investment in AI-related infrastructure… are still our core strategic allocation direction.” This isn’t a bet by a single institution—it’s Wall Street’s collective capital rearranging and queuing up again.

Viewed on the coordinates of history, this scene feels familiar. During the most frantic period of the 1999 dot-com bubble, the truly steady winners weren’t any particular internet company, but the infrastructure firms that sold routers and fiber optics. The difference is that this time, the roles have been replaced by power plants, the power grid, and nuclear reactors. Back then, not every infrastructure company necessarily survived, but the valuations across the entire track did surge along with the number of internet users. This time, the market clearly wants to run the same storyline again.