Investors' AI Bubble Concern Focuses On Power Providers, Especially Nuclear

As concern about an AI bubble has grown, prescient VCs and other investors have begun to ascertain where the greatest risk may lie. That search has focused on companies considered "AI beneficiaries," those not directly engaged in AI development but ancillary businesses providing support to the Big Tech and startup enterprises through provision of essential services including cloud computing, semiconductors, construction, storage - and power.

And while use of the word 'essential' might suggest immunity from any downturn, the reality is that too many investors looking for indirect means of capitalizing on the AI boom have leapt into the various categories, over-investing relative to potential market size and inflating valuations. But none has seen more of that exuberance than has the industry providing power to AI data centers. And within that, those investing in micro or small modular nuclear reactors appear the most exposed. Many of the nascent power providers are unprofitable; virtually none of the nuclear aspirants has deployed a reactor, let alone generated revenue. The issue is not about whether this is technically and economically feasible, but whether it will happen before a shakeout occurs and irrational projections are exposed. JL

Colin Laidley reports in Investopedia and divsrv reports in Neutron Bytes:

As AI spending has ballooned, so have the companies claiming a piece of it. The AI buildout has turbocharged the sales of slow, steady businesses transformed into buzzy growth names. It has juiced the stocks of nascent companies years from self-sufficiency. Where the AI trade has become most bubbly are "AI beneficiaries:" cloud computing; semiconductors; power providers; and storage/cooling. (But) power has more unprofitable companies than any other category. Five of the 14 expect to report a loss this year. No other category has more than one unprofitable company. None of the small reactor developers have deployed a reactor for a customer. These companies have no revenue. “The exaggerated order books will soon be called out. We will see investors dry up."

AI boom or AI bubble? That's been one of the most pressing debates on Wall Street of late.

Some investors see echoes of the Dotcom Bubble in Nvidia's (NVDA) and OpenAI's recent circular dealmaking bonanza and soaring stock prices. Others note the AI boom is being financed by hugely profitable tech companies, whose valuations aren't nearly as high as their Dotcom peers.

As Big Tech's AI spending has ballooned, so have the ranks of companies claiming a piece of the pie. The AI buildout has turbocharged the sales of unsexy, slow-and-steady businesses and transformed their stocks into buzzy growth names. It has also juiced the stocks of nascent companies that are years from self-sufficiency, creating pockets of exceptional froth within a pricey AI ecosystem.

 

To understand where the AI trade has become most bubbly, Investopedia identified 75 companies regularly referred to as "AI beneficiaries" by Wall Street analysts, and sorted each into one of five categories: cloud computing providers; semiconductor makers; software companies; power providers; and networking, storage, and cooling equipment makers.

A few companies, like Microsoft (MSFT), fit in multiple categories, in which case we've placed them in the one that feels most central to their AI business today. At present, Microsoft's cloud revenue is the best metric for assessing its AI business, so it's classified as a cloud provider instead of a software company.

 

Power Provider Stocks Seem The Frothiest

All five categories have seen their valuations rise over the past few years, but none more than power providers. The median price-to-sales (P/S) ratio of our power basket in 2025 is 4.53, nearly three times the median in 2023 (1.52). The next largest P/S expansions over that period were in networking, storage and cooling (4.45 in 2025 vs. 2.09 in 2023) and cloud providers (10.5 vs 6.34). (Cloud providers have a higher P/S ratio in absolute terms because the category is composed of tech stocks that have historically commanded higher valuations. This is why we've compared change over the past three years rather than absolute P/S ratios.)

Power also has more unprofitable companies than any other category—so many in fact that we’re using price-to-sales as our benchmark valuation metric rather than the more common price-to-earnings ratio. Five of the 14 companies in our power basket are expected to report a loss this calendar year. No other category has more than one unprofitable company. 

The race to build the data centers that train and run AI models has set off an equally frenzied race to generate and transmit the vast amounts of electricity those data centers consume. Nuclear energy has attracted interest from tech companies for its efficiency and small carbon footprint. Microsoft, Amazon (AMZN), Alphabet (GOOG), and Meta (META) have all signed multi-billion dollar deals with nuclear power plant operators like Constellation Energy Corp. (CEG) and Vistra (VST), both of which have seen their stocks surge over the past two years. 

But in their rush to bring reliable sources of electricity online, tech companies and investors have also thrown money at nuclear tech upstarts, some without operational generators or regulatory approvals. Shares of small modular reactor maker NuScale Power (SMR) doubled in value between January and mid-October. At the stock’s peak earlier this month, the company, which reported $37 million in revenue last year and isn’t expected to be profitable until 2029, was valued at more than $15 billion. 

 

The financial health of the AI industry has a direct bearing on the number of data centers expected to be built which in turn potentially drives demand for SMRs and microreactors to power them. It is a daisy chain of connections that appears to contain a fair degree of risk.

A news analysis published in the New York Times on 10/14/25 by former members of the Council of Economic Advisers, Jared Bernstein and Ryan Cummings, warns that the current surge in Artificial Intelligence (AI) investment exhibits classic signs of a speculative bubble, similar to the dot-com era of 2000.

If the two economists are correct, then there is a lurking risk that if the AI bubble pops, that paper commitments, e.g., nonbinding MOUs with SMRs for data centers, may be tossed in the bit bucket. SMR and mircoreactor developers could wind up without the customers they expected to drive the growth of their businesses and provide returns for their stockholders.

A Booming Stock Price With No Revenue?

Meanwhile, nuclear tech startup Oklo’s (OKLO) market capitalization peaked at $25.7 billion earlier this month, a 720% increase from the start of the year. Oklo is the only company of the 75 we’ve included in our analysis that is expected to report no revenue this year. Analysts forecast it will turn a profit for the first time in 2030. 

Though it’s not just nuclear businesses that have achieved rich valuations. Fermi (FRMI), the developer of a massive AI data center campus in the Texas Panhandle, was founded in January and went public in early October at a valuation of more than $19 billion. Fermi plans to build 11 gigawatts of computing capacity powered by on-site nuclear, natural gas, wind, and solar generators. It expects to break ground on its first data center in March, and hopes to have about 1 GW of capacity online by the end of 2026.2

 

AI power stocks just finished a volatile week that may underscore the extent to which their prices are driven by fickle sentiment. Constellation Energy and Vistra both shed more than 10% of their value in the first half of the week, as did GE Vernova (GEV), whose turbines are in high demand from data center clients seeking to draw on the South’s ample supply of natural gas. All three finished the week little changed.

The upstarts were hit even harder. NuScale, Oklo, and Fermi all lost more than 25% of their value between Monday's open and midday Wednesday. But they too rebounded, finishing the week with losses in the low- to mid-teens.  

The falling dominoes that would follow from an AI bubble bursting are that many but not all of the nonbinding MOUs for nuclear energy reactors to power AI data centers would evaporate taking them the over valuation of some of the stocks of nuclear mirco and small modular reactor startups.

In short, the collapse of the AI bubble would negatively affect firms in the data center industry and, following that, impact the nascent nuclear energy industry especially developers of SMRs and microreactors who have built up investor confidence in their future, based on part, on stacks of nonbinding MOUs for AI data centers not yet built.

Given that AI platforms are a significant element of growth of the US economy, the question is whether there is “excessive exuberance” driving the AI industry is germane.

A key risk for investors is that none of the SMR and micro reactor developer have yet to deploy a commercial reactor for a customer. Licensing challenges with the Nuclear Regulatory Commission remain a significant hurdle. These companies currently have no revenue, meaning the investment case rests squarely on future contracts, partnerships, and government support. And with a rally this steep, any delays or regulatory snags could pressure the lofty valuations of their stocks.

What Experts in Nuclear Energy Finance are Saying

A knowledgeable senior level investor emailed Neutron Bytes with the following observation. The person’s name and title are withheld due to their position.

“The pseudo or exaggerated order books or projections will soon be called out. We will also see investors dry up, moving away from the wishful reactor developer class, as decisions get made and real customers choose from two or three, maybe four or so more credible designs.”

This isn’t the only warning. According to a report by the Bloomberg wire service on 10/07/25 “some on Wall Street are starting to get a little dubious about how frothy the market for nuclear stocks has gotten.”

Rama Variankaval, global head of corporate advisory for JPMorgan Chase & Co., acknowledged in an interview with Bloomberg that there is “unrealistic optimism” around the sector.

Bloomberg reports that Chris Gadomski, head of nuclear research at BloombergNEF, said that he, “ominously compared the enthusiasm around SMRs and data centers to the internet boom and bust of the early 2000s” in an interview with the Financial Times.

Gadomski also suggested that Wall Street might be underestimating the enormous costs it will take to get SMRs up and running.

“There’s a lot of cheerleading happening, but the amount of capital that you need to cross the finish line is huge,” he said.

If investor confidence in the future of the AI data center customer segment is negatively impacted, the daisy chain of effects of investors’ response could be one of being less willing to support financing the run from prototypes to products for SMRs and microreactors.

Even the two of leading advanced reactor projects, TerraPpower and X-Energy, which are benefiting from billions in cost shared funding from the Department of Energy’s Advanced Reactor Demonstration Program, know that they will have to raise billions more to build their first of a kind units for customers.