The AI gold rush meets Wall Street reality check

The artificial intelligence (AI) boom may still be hungry for computing power, but investors are starting to ask harder questions about who will pay for the data centers feeding it.

Companies tied to data center construction, power and land had been expected to dominate the US initial public offering pipeline. That enthusiasm is showing signs of cooling as investors scrutinize valuations, massive capital requirements and the growing political and community resistance surrounding new facilities.

One of the clearest tests is SoftBank-backed SB Energy, which is developing a massive Ohio data center project to be leased to OpenAI.

The company had targeted a valuation of $50 billion or more, but its IPO has been pushed back as bankers struggled to attract enough investors at the prices sought, according to The New York Times.

SB Energy is not a small bet on AI. Its IPO filing showed revenue rose 66.4 percent to $138.7 million in the first half of 2026, while its net loss widened to $3.21 billion. Reuters reported that the company is moving ahead with plans to sell shares, including up to $500 million to Japanese investors as part of its US listing.

Its biggest challenge may be scale. SB Energy has yet to put a data center into operation even as it projects a $439 billion revenue backlog over roughly two decades, much of it tied to its planned Ohio development.

Not just an SB Energy problem

Investor caution has spread beyond one offering.

Holtec Nuclear postponed its planned IPO after previously seeking to raise as much as $900 million at a valuation of about $10 billion. The company explicitly cited “uncertainty over data center development” among the factors hurting market confidence.

Aggreko, another power provider with data center customers, has also slowed its listing plans amid broader market and industry uncertainty, according to The New York Times.

That does not mean AI’s infrastructure boom is disappearing.

The International Energy Agency expects global data center electricity consumption to more than double to about 945 terawatt-hours by 2030, with AI serving as the biggest driver of that increase. Data centers could account for roughly half of US electricity demand growth through the end of the decade.

Big Tech is also still spending heavily.

Meta, Alphabet, Amazon, Microsoft and Oracle have collectively committed more than $1 trillion to infrastructure plans, while estimates cited by The New York Times suggest more than $5 trillion could be needed to build the infrastructure supporting the AI boom through 2030.

What has changed is the tolerance for building at any cost.

Data centers consume enormous amounts of electricity and, depending on their cooling systems, water. Those demands have increasingly drawn scrutiny from residents, utilities and governments concerned about grid capacity, household energy bills and the cost of new infrastructure.

The US House recently passed legislation directing regulators to ensure large data centers shoulder the costs of power and transmission infrastructure built to serve them. An AP-NORC poll also found about 60 percent of Americans supported limiting the number of new data centers, while roughly two-thirds favored requiring them to use clean energy.

The pressure is also reaching state governments. Texas Gov. Greg Abbott on 21 September ordered a halt to new state-issued data center permits pending an audit of their impact, while California Gov. Gavin Newsom signed a package of measures covering issues including electricity costs, water consumption and local oversight.