Wall Street Is Worried About an AI Bubble-One Sector Looks Especially Stretched
AI hype has lifted almost everything tied to model training, data centers, and the compute supply chain. But one slice stands out for valuation risk: power providers, especially those linked to nuclear.
Investors are paying up for electricity exposure because AI needs staggering amounts of energy. That's pushed some utility-like names into growth-stock territory and sent early-stage nuclear tech firms to multibillion-dollar market caps-despite thin sales or none at all.
Why This Matters for Portfolios
The stocks that run the hardest in a hot market often drop the fastest when sentiment cools. That's amplified for young companies that need steady access to capital and don't yet produce meaningful cash flow.
If you're positioned for the AI buildout, the risk doesn't sit only in chips and cloud. It's now embedded in the pipes and electrons that keep data centers running.
Where Valuations Have Stretched Most
Across five "AI beneficiary" groups-cloud providers; semiconductor makers; software; power providers; and networking, storage, and cooling-multiples have risen. But power leads the expansion.
- Power providers: median price-to-sales (P/S) 4.53 in 2025 vs. 1.52 in 2023
- Networking, storage, cooling: 4.45 in 2025 vs. 2.09 in 2023
- Cloud providers: 10.5 in 2025 vs. 6.34 in 2023 (higher in absolute terms, but the focus here is the jump)
Power also has the most companies running at a loss. Five of 14 names in the basket are expected to be unprofitable this year; no other category has more than one.
How AI Turned Electricity Into a Growth Story
Data centers that train and serve AI models devour electricity. That's pushed Big Tech to cut long-term deals with nuclear-heavy operators like Constellation Energy and Vistra. Their shares have surged as demand visibility improves.
At the same time, capital has flowed into nuclear startups with ambitious timelines and limited operating history. In a momentum tape, that cocktail can push prices far ahead of fundamentals.
The Outliers: NuScale, Oklo, and Fermi
NuScale Power doubled between January and mid-October. At its peak earlier this month, it was valued above $15 billion, despite reporting $37 million in revenue last year and not being expected to reach profitability until 2029.
Oklo's market cap topped $25.7 billion-up roughly 720% year to date-while being the only company in a 75-name AI basket projected to generate no revenue this year. Street forecasts point to first profits in 2030.
Fermi went public in early October at north of $19 billion after being founded in January. The plan: an 11 GW AI data center campus in the Texas Panhandle, powered by on-site nuclear, natural gas, wind, and solar. Groundbreaking is targeted for March, with about 1 GW online by the end of 2026.
A Reality Check: A Whipsaw Week
Sentiment cuts both ways. Constellation Energy, Vistra, and GE Vernova each fell more than 10% early in the week, then finished roughly flat. The message: even "established" beneficiaries aren't immune to air pockets.
The upstarts swung harder. NuScale, Oklo, and Fermi dropped more than 25% by midweek before paring losses to the low-to-mid teens by Friday. When narratives drive price, volatility follows.
What to Watch Before You Add Exposure
- Multiples vs. history: compare current P/S to three-year norms, not just peers riding the same wave.
- Path to profit: note revenue reality today and credible milestones to break-even; watch for slippage.
- Funding and runway: assess cash, burn rate, and access to capital if markets tighten.
- Contract quality: look for long-duration, take-or-pay deals with hyperscalers-not just MOUs and headlines.
- Regulatory gating: nuclear timelines hinge on licensing and approvals. Review status directly with the U.S. Nuclear Regulatory Commission.
- Project execution: capex, grid interconnects, and transmission bottlenecks can delay revenue recognition.
- Stack diversification: consider spreading risk across chips, cloud, networking/cooling, and power instead of chasing the hottest corner.
- Position sizing: size smaller in pre-revenue names; stress test for 30-50% drawdowns.
If You Need a Primer on AI to Underwrite These Bets
Getting fluent in how models, compute, and data center demand interact can sharpen your thesis work. For practical overviews and tools by job function, see AI courses by job or explore finance-focused tooling here: AI tools for finance.
One Last Angle
AI is real. So is overpaying. In power, both truths can exist at once: durable demand and stretched pricing. Price your risk accordingly.
Your membership also unlocks:
