The hot AI trade is not another model or chip. It is electrons, and the gear that safely moves them to racks of accelerators. Energy companies are raising money at IPO at their fastest pace this century, according to Dealogic figures cited in Ars Technica, and the linkage to AI fits on a whiteboard. Every new compute cluster needs power, then more power, then redundancy, and the market is trying to fund each step where that need meets real infrastructure.

The bottleneck investors can model

The fundraising surge sits beside a scramble for access to the huge amounts of energy needed to run data centers, which Ars Technica describes as a bottleneck inside a multi-trillion-dollar AI investment boom. That framing matters. It shifts attention to the measurable: how much capacity exists, how quickly equipment can be built, and whether capital reliably turns into available megawatt hours.

Analysts quoted by Ars Technica say the trade moved from chip names to the picks and shovels that feed those chips. “Investors started by buying AI-linked names like Nvidia. Then they said, ‘hold on, every chip needs energy to power it,’” said RBC clean energy analyst Chris Dendrinos. That logic has pulled a mixed set of companies to market, each posted at a specific choke point.

Where AI touches the grid

Start with the last mile of power delivery. Forgent Power Solutions, which designs and manufactures electrical distribution equipment used in data centers, raised 1.7 billion dollars in February. Ars Technica notes that the company benefited from strong demand and long wait times for technologies such as transformers and switchgears, the protective and routing gear that makes high-density compute possible without frying the rest of a facility. If you believe AI buildouts will persist, you can underwrite a backlog for the boxes that gate new capacity.

Another touchpoint is the choice to bypass a strained grid. Innio, a German gas engine manufacturer, completed a nearly 2.8 billion dollar flotation in June, riding a trend of data centers powering themselves on-site. This is not a bet on abstract electrification. It is a bet that operators will trade grid interdependencies for the control that comes with their own generation.

Then come the capital-heavy swings at new supply. Ars Technica lists nuclear and geothermal among the projects that have found public market buyers this year, alongside businesses trying to develop new technologies. Fervo, which went public in May and raised nearly 2.2 billion dollars, is developing next-generation geothermal that adapts oil and gas drilling methods. According to its prospectus, the company plans to spend 1.2 billion dollars over the next year to develop its Utah power station. That is a direct pipeline from IPO proceeds to future electrons.

“This is a moment in which speculative projects are being funded and underwritten,” said Julien Dumoulin-Smith of Jefferies.

The demand case tying these together is large but not uniform. Ars Technica cites an estimate that a typical AI-focused data center uses around 876,000 megawatt hours per year, and it references a consultancy projection that US electricity demand could increase 39 percent between 2026 and 2035 in large part due to data centers. Those are directional stakes, not guarantees, which is why the instruments vary. Some investors are reaching for complexity premiums in nuclear and geothermal. Others are buying nearer-term cash flows in distribution equipment or on-site engines.

Picking the pickaxes, then checking the blade

The pace is striking. Initial public offerings for energy firms raised 12.6 billion dollars in the first half of this year, the highest half-year level since late 1999 and the highest first-half figure on record, per Dealogic. ETF providers are packaging the theme as well. Ars Technica reports a new power infrastructure ETF from GMO aimed at generation, grid, and electrification names. Bill Smith of Renaissance Capital frames 2026 as the year that financed the AI revolution’s infrastructure, alongside the marquee of SpaceX.

It would be neat if the story ended there. It does not. Despite the surge in demand and the pipeline of offerings, early trading has been rough. Nearly two-thirds of the energy companies that floated this year and last are now below their offer price, compared with less than 40 percent across all sectors, according to Dealogic data cited by Ars Technica. The article lists several examples, including companies tied to small modular reactors, gas generation, and data center energy that are trading below issue prices.

That divergence explains the tone from allocators. Manish Kabra at Société Générale described “power-capacity expansion, US reshoring, [and] AI-related infrastructure investment” as central allocations. The phrase captures the spread of bets, from upstream generation to local equipment to macro themes like reshoring. It also suggests that investors are trying to own the transmission of value from AI enthusiasm into regulated and physical assets, without assuming that every name will work.

How to read the next filings

The lesson for the next batch of S-1s is to trace the electron. If the company gets paid when a data center adds megawatts or reliability, then its revenue is likely to be indexed to the actual cadence of deployments. If it depends on unproven technology or multi-year development cycles, then early trading may not reflect the eventual demand that AI could create. Ars Technica notes that energy companies often trade at lower valuations than tech peers, which is part of the appeal, but the path from prospectus to profit still runs through project execution.

Treat AI energy as a map, not a monolith. The dependable pieces sit where compute ambition meets a meter, a transformer, or a purchase order for engines. The speculative pieces sit where physics and permitting timelines still have the last word. The market is funding both at the fastest pace this century, then it is checking the blade in secondary trading.