The AI power trade is almost three years old. It started with a simple observation (data centers need electricity, and you can't build a power plant in a quarter) and became one of the best-performing groups in the market. We ranked 14 of its best-known names against their own five-year valuation history, as of October 5, 2026. The trade has split into distinct groups, and the most interesting change is one you won't see in a price chart: for two of the biggest names, earnings caught up with the stock.
If you only read one thing
The short version: GE Vernova and Vertiv now sit at the 11th and 13th percentile of their own five-year P/E range, because their earnings grew faster than their share prices, while their sales multiples remain in the upper range. The grid-equipment makers (Eaton, Quanta, nVent) sit near five-year highs on every measure. The generators (Constellation, Vistra) are in the lower-middle on earnings. The speculative end (Bloom, NuScale) trades at 86th to 99th percentile sales multiples.
Where the trade started
Before looking at today, it helps to know where these stocks began. We recomputed each name's EV/Sales percentile as of January 2024, using only the five years of data available at that date.
| Ticker | EV/Sales, Jan 2024 | 5y %ile then | EV/Sales now | 5y %ile now |
|---|---|---|---|---|
| VST | 1.5x | 47 | 3.4x | 60 |
| NRG | 0.8x | 38 | 1.3x | 72 |
| ETN | 4.6x | 99 | 6.3x | 96 |
| PWR | 1.8x | 95 | 3.3x | 93 |
| HUBB | 3.6x | 97 | 4.6x | 84 |
| NVT | 4.9x | 100 | 6.0x | 93 |
| VRT | 3.2x | 99 | 8.6x | 87 |
| MOD | 1.4x | 100 | 3.0x | 76 |
Where these stand now
Current readings update separately from the article’s dated analysis.
As of
VRT's P/E of 48.0x is lower than 94% of its weekly readings since November 2021.
As of
GEV's P/E of 26.6x is lower than 97% of its weekly readings since April 2025.
As of
ETN's P/E of 40.1x is higher than 93% of its weekly readings over the last five years.
Two different setups hid inside one trade. The power generators (Vistra, NRG) started from the middle of their own range. The equipment makers were already at their five-year highs in January 2024 and kept rising anyway, because revenue grew alongside the multiple.
Notice that several equipment names show a lower percentile today despite a higher multiple. Vertiv's EV/Sales nearly tripled, yet its percentile fell from 99 to 87. The five-year window now includes the 2024 and 2025 highs, so the bar moved. We explain that effect in why a valuation percentile can move without a price move.
The takeaway: In early 2024 the power generators were mid-range on valuation and the electrical-equipment makers were already near five-year highs. Both groups rallied. A lower percentile today can just mean the five-year window now contains the recent highs.
Where each name sits today
| Ticker | Role | P/E (TTM) | P/E 5y %ile | EV/Sales | EV/Sales 5y %ile | Rev TTM YoY |
|---|---|---|---|---|---|---|
| GEV§ | Turbines and grid equipment | 28.4x | 11 | 6.2x | 76 | 13% |
| VRT | Data-center power and cooling | 57.1x | 13 | 8.6x | 87 | 26% |
| CEG† | Nuclear generation | 25.1x | 26 | 4.5x | 65 | 20% |
| VST | Power generation | 23.7x | 30 | 3.4x | 60 | 19% |
| HUBB | Grid components | 28.1x | 48 | 4.6x | 84 | 7% |
| NRG | Generation and retail power | 25.0x | 68 | 1.3x | 72 | 12% |
| MOD | Thermal management | 67.0x | 86 | 3.0x | 76 | 29% |
| NVT | Electrical enclosures and cooling | 46.5x | 89 | 6.0x | 93 | 46% |
| PWR | Grid construction | 77.5x | 90 | 3.3x | 93 | 26% |
| ETN | Electrical equipment | 44.4x | 98 | 6.3x | 96 | 16% |
| BE† | Fuel cells | n/m | n/a | 38.1x | 99 | 58% |
| SMR | Small modular reactors | n/m | n/a | 164.0x | 86 | −81% |
| OKLO | Advanced nuclear | n/m | n/a | n/m | n/a | pre-revenue |
| CCJ‡ | Uranium | 144.4x | n/a | n/a | n/a | −7% |
Earnings did the compressing
GE Vernova's trailing earnings per share rose about eightfold over the past year, and Vertiv's roughly doubled. Their share prices did not keep pace with that earnings growth, so their P/E ratios fell to the 11th and 13th percentile of their own range. Their sales multiples did not fall: GE Vernova sits at the 76th percentile on EV/Sales and Vertiv at the 87th.
That is the healthy kind of compression: the stock didn't crash, the business grew into it. It also explains why these names drifted around since we first ranked the AI complex in August, in our analysis of which AI stocks actually got cheap. Vertiv's P/E percentile fell from the 23rd to the 13th as earnings kept arriving. GE Vernova's moved the other way, from the 5th to the 11th.
The caveat: a low P/E percentile built on a sudden jump in earnings depends on those earnings holding. Margins that expand this fast can also contract fast. See GE Vernova's valuation history and Vertiv's valuation history to compare the P/E and sales views.
The takeaway: GE Vernova and Vertiv look cheap on earnings because their profits grew faster than their stock prices. On sales, both still trade in the upper part of their range. The low P/E reading depends on those profits staying high.
The grid builders never got cheap
Eaton sits at the 98th percentile on P/E and the 96th on EV/Sales. Quanta and nVent sit at the 89th to 93rd on both. nVent's revenue grew 46% over the trailing year, the fastest of the established equipment names, so part of its premium has revenue behind it. These companies were priced as winners in January 2024 and still are. Check Eaton's valuation history for the cleanest example of a multiple that has stayed near its top for years.
The takeaway: The companies that build and equip the grid (Eaton, Quanta, nVent) have traded near five-year-high multiples for most of the AI trade and still do. Revenue growth supports part of that, especially at nVent.
The generators sit in the middle
Constellation (26th percentile on P/E) and Vistra (30th) are in the lower-middle of their own earnings range, with revenue up about 20%. NRG sits higher at the 68th. Constellation carries a data-quality flag in our dataset, and its history only starts in 2023, so read its row with care. Vistra is the cleaner read: Vistra's valuation history shows how far its P/E has come down from its 2024 peak.
The takeaway: The power producers Constellation and Vistra trade in the lower-middle of their own range on earnings, with revenue still growing around 20%. NRG sits higher.
The speculative end
Bloom Energy trades at 38 times sales, the 99th percentile of its own range, with revenue up 58%. NuScale trades at 164 times sales while its trailing revenue fell 81%. Oklo has little or no revenue, so no sales multiple means anything yet. These stocks trade on what investors expect the next decade to look like, and a valuation percentile can't tell you much about that.
The takeaway: Bloom, NuScale, and Oklo are priced on long-range expectations, not current sales. Bloom and NuScale sit near the top of their own ranges; Oklo has too little revenue to measure.
Why cyber brings this trade back
On All-In #291, David Friedberg argued that AI-driven cyber attacks will turn defense into a compute race. If attackers and defenders run similar models, the side with more machines wins, and in his words "the cyber defense necessity actually mandates that we continue to build out data centers" (the cyber segment of All-In #291). We covered his full argument in what Friedberg said about AI agents and cyber defense.
That point matters more for this group than for the security stocks. Cyber platforms have no capacity constraint; power does. If security becomes one more reason to build data centers, the demand lands on the same generators, turbines, and grid equipment in this table. Our ranking of AI cybersecurity stocks by valuation percentile found the security platforms already at the 96th to 98th percentile. Several power names are not.
The takeaway: If Friedberg is right that cyber defense becomes a race for computing power, the extra demand flows to data centers and the power behind them. That links the cyber trade back to this one, and several power names are priced lower in their own range than the cyber platforms.
Every name above updates weekly on the AI power theme page. The AI infrastructure theme ranks the wider AI build-out by where each multiple sits in its own history, and the AI cyber defense theme page tracks the security side. New to percentiles? Read what a valuation percentile is and how we compute it.
Data: SEC EDGAR XBRL filings and market prices, computed by StockResearch as of October 5, 2026. January 2024 percentiles were computed from the same weekly series using only the trailing five years available at that date. Computed multiples are cross-checked against independent snapshots, with divergences flagged in the table. This article summarizes arguments David Friedberg made on the cited podcast; those views are his. Jake is the founder of StockResearch.app, where he writes data-first research on valuation dislocations. Nothing here is investment advice or a recommendation to buy or sell any security. Do your own research.