Why the AI Power Trade Is Still Cold
AI needs the kilowatts, so why are the power stocks dead money? The whole basket trades below trend with nothing overbought — yet the Street targets +30% to +52% on the IPPs. The gap is not demand. It is the clock: chips reprice on next quarter, power on rate-base math, PPAs, and multi-year interconnection queues.
The upside the cold tape is ignoring
+41%
▲ +41 Median analyst upside to target across Vistra, Constellation and NextEra — the AI-power names the market treats as dead money. Basket 14-day RSI sits near 43; 13 of 16 trade below their 20-day average; not one is overbought.
For two weeks the chips ran 24–100% while the layer that powers them sat flat-to-down. The market read that as a verdict on power. The data says it is a verdict on timing. The AI-power basket is cold on the tape — median 14-day RSI near 43, thirteen of sixteen names below their 20-day average, not one overbought — but analysts model +52% upside on Vistra, +41% on Constellation, +30% on NextEra to mean target. The reason the market will not pay today: power earns slowly. Independent producers sign multi-year private contracts; utilities earn a regulated return on assets approved over years; new plants wait in interconnection queues. So the demand is real but the cash is back-loaded. You can see it in the growth curve — Constellation decelerates from +33% revenue growth this year to roughly flat next year, Vistra from +31% to +9%, while regulated utilities sit at single digits. The one exception is cooling: Vertiv compounds ~36% to ~28% and is the only name that is both cheap and still accelerating. Meanwhile insiders are buying the slowest names — Dominion +$9.7M, AEP +$2.5M, Duke, Constellation — and selling the momentum grid names, Quanta −$481M and Eaton −$46M. The model rates the entire layer 37–52: nothing broken, nothing compelling. A coiled spring waiting for a capex print.
The contradiction
AI can't run without power — so why are the power stocks the coldest corner of the AI trade? It isn't demand. It's timing: chips get paid next quarter; power gets paid on a 2027–2030 clock of contracts, rate cases, and permitting queues.
What the headline says
The AI power trade looks like dead money
Median 14-day RSI ~43, 13 of 16 names below their 20-day average, nothing overbought
What the data says
But the Street models a double-digit-to-50% rerate
VST +52%, CEG +41%, NEE +30% upside to mean target; Vertiv compounding revenue ~36% → 28%
Chapter 01
Cold Tape, Hot Targets: The Street Sees +30% to +52% in the Names Nobody Wants
The AI-power basket trades like dead money — yet it carries the widest analyst upside in the AI complex. Vistra sits 52% below its mean price target, Constellation 41%, NextEra 30%, Vertiv 21%. At the other end, GE Vernova and Carrier have already closed the gap (0% and −2%). The contradiction the rest of the story explains: how can the "cold" layer also be the one the Street says has the most room?
Start with the punchline, because it is the whole story in one chart. These are the power, grid, and cooling names that enable AI — the layer the tape has left for dead — and the Street models more upside here than almost anywhere in the AI trade. Vistra trades at $155 against a $235 mean target: +52%. Constellation at $266 against $374: +41%. NextEra at $84 against $109: +30%. Vertiv, the cooling leader, at $323 against $390: +21%. These are not fringe analysts; they are consensus mean targets. The skeptic's reply is fair: price targets lag, and a wide gap can mean the stock is broken, not cheap. So note the other end of the bar. GE Vernova has run so hard its price now sits at its target — 0% upside left. Carrier is actually above its target, −2%. The market has fully paid for the names it loves and refuses to pay for the names it does not. That split — huge modeled upside on one side, none on the other — is not random. It maps almost exactly onto how fast each business turns AI demand into cash. The rest of this piece traces that mechanism.
Analyst upside to mean price target
Source: MarketDecode scanner + analyst targets, 2026-06-02
Widest Upside
VST 52%$155 vs $235 mean target
No Room Left
GEV / CARR0% and −2% — already at target
IPP Avg Upside
41%VST, CEG, NEE — the "cold" names
Chapter 02
Nobody Is Hot: The Whole Layer Sits Below the Line
The "cold" is not a vibe; it is in the tape. Across the basket, 14-day RSI clusters in the low 40s — NextEra and Duke near oversold in the mid-30s, the warmest name (Vistra) only at 51. Not one of the sixteen power names is overbought. The same two weeks, the chip layer they power ran to RSI 75–90. Momentum investors have simply left this aisle of the store.
If section one is "the Street says these are cheap," section two is "and the tape agrees nobody wants them." Relative strength — the standard 0-to-100 momentum gauge, where above 70 is hot and below 30 is washed out — tells the story without a single price. NextEra sits at 33, Duke at 35, both a stone's throw from technically oversold. GE Vernova is at 41, Constellation at 42, Vertiv at 48. The single warmest name in the entire group, Vistra, is only at 51 — dead center, not hot. Thirteen of the sixteen names with price data trade below their own 20-day average. Zero are overbought. Contrast that with the layer this electricity actually feeds: over the same stretch, Dell, Micron, Astera Labs, and Super Micro ran to RSI readings of 79 to 90. So the market is not confused about whether AI is real — it is expressing a preference about WHEN it wants to be paid. It crowds into the names where the revenue is already printing and abandons the names where the revenue is coming but slowly. A cold RSI is not a fundamental signal here. It is a patience signal.
14-day RSI across the power basket
Source: MarketDecode scanner, 14-day RSI as of 2026-06-02
Coldest
NEE / DUKRSI 33 and 35 — near oversold
Warmest
VST · RSI 51Not one name above 60
Below 20-Day Avg
13 of 16The whole layer trades under trend
Chapter 03
Why Chips Get Paid First: The Growth Curve Bends the Wrong Way
The market pays up for acceleration. The power layer offers the opposite. Constellation grows revenue +33% this year, then roughly flat (−0%) next year. Vistra goes +31% to +9%. Even the grid names step down. Regulated utilities never leave single digits. The lone exception is cooling: Vertiv holds +36% this year and +28% next — the only name in the basket whose growth does not fade.
Here is the engine under the cold tape. A semiconductor stock reprices when next year's growth looks faster than this year's — acceleration is the whole game. Run that test on the power layer and you see why the market hesitates. Constellation's revenue is set to grow about 33% this year and then, on consensus, roughly flat next year as the comparisons get hard and contracted volumes reset. Vistra steps down from +31% to +9%. Quanta, the grid build-out name, moderates from +23% to +13%. The regulated utilities — Duke, AEP, Dominion, Exelon — never leave the single digits in either year, by design: a regulator-approved return on a slowly growing asset base is a feature, not a miss. None of that is deceleration into weakness; it is what a contracted, regulated, capital-heavy business looks like. But it does not give a momentum buyer the accelerating curve they pay premiums for. The exception proves the rule. Vertiv — cooling and power management, sold as equipment, not regulated, not contracted-for-decades — grows about 36% this year and 28% next. It is the one place in the basket where AI demand still compounds visibly. That is precisely why it is also the one "cold" name with both upside to target and a real growth story attached.
Revenue growth: this year vs next year
Source: MarketDecode scanner, forward revenue estimates, 2026-06-02
Sharpest Fade
CEG 33% → 0%Growth front-loaded into this year
Only Durable Grower
VRT 36% → 28%Equipment, not regulated or contracted
Regulated Utilities
Single digitsDUK, AEP, D, EXC — by design
Chapter 04
Re-Rate Bet or Growth Bet: Where Each Name Actually Sits
Plot next-year growth against upside to target and the basket splits into clean quadrants. The IPPs (Vistra, Constellation, NextEra) sit top-left: huge upside, fading growth — a bet that the market re-rates them, not that they grow faster. Vertiv sits alone top-right: cheap AND still compounding. The grid names sit bottom-right: still growing but already priced. Knowing which quadrant you are buying is the whole decision.
The same numbers, arranged as a map, settle the "is cold an opportunity?" debate name by name. The vertical axis is upside to analyst target; the horizontal axis is next-year revenue growth. Four corners, four very different trades. Top-left — high upside, low forward growth — is where the independent power producers live: Vistra at +52% upside but only +9% growth next year, Constellation at +41% on roughly flat growth, NextEra at +30% on +10%. Owning these is a bet that the market eventually re-rates a slow, contracted cash machine higher — a valuation call, not a growth call. Top-right — high upside AND high growth — has exactly one resident: Vertiv, +21% upside on +28% growth. It is the only name you can buy for both reasons at once. Bottom-right — strong growth, little upside left — is the grid build-out crowd: GE Vernova (+14% growth, 0% upside) and Quanta (+13% growth, +9% upside) have already been paid for their growth. Bottom-left — modest growth, modest upside — is the regulated-utility ballast: Duke, AEP, Carrier. The lesson is not "buy the cold layer." It is "know whether you are buying a re-rate, a grower, or ballast," because they behave nothing alike when the catalyst lands.
Next-year growth vs upside to target
Source: MarketDecode scanner, forward estimates + targets, 2026-06-02
Re-Rate Bets
VST · CEG · NEEBig upside, fading growth — valuation call
Cheap + Growing
VRT+21% upside on +28% growth — alone
Already Priced
GEV · PWRStrong growth, upside spent
Chapter 05
Who Is Buying the Cold: Insiders Accumulate the Slowest Names
Follow the people closest to the rate-base math. Over 90 days, insiders are net buyers of the regulated utilities — Dominion +$9.7M, AEP +$2.5M, Constellation +$0.6M, Duke +$0.2M — the slowest, most back-loaded part of the basket. They are sellers of the momentum grid and cooling names, led by Quanta at −$481M and Eaton at −$46M. The smart-patience money is rotating toward the clock the market hates.
Insider flow is the quiet tell under a cold tape, because insiders are not trading the chart — they are trading the multi-year plan they can see and you cannot. Over the last 90 days the net buying clusters in exactly the names the market calls dead: Dominion, the regulated utility, +$9.7M net; AEP +$2.5M; Constellation +$0.6M; Duke +$0.2M. These are the businesses whose AI upside is locked in rate cases and long contracts that will not show up in a stock chart for quarters. On the other side, the selling concentrates in the names that already ran: Quanta Services, the grid build-out story, saw −$481M in net insider selling — the heaviest in the group by an order of magnitude — and Eaton −$46M, with Trane, Carrier, and Johnson Controls all net sellers in the millions. Read together, it is a rotation: out of the momentum names that have already been paid for their growth, and into the slow regulated names that have not. That does not guarantee the regulated trade works — insiders are early as often as they are right. But when the tape is cold and the people who know the contracted backlog are accumulating, the "cold means broken" story gets harder to defend.
Insider net buying/selling, last 90 days
Source: MarketDecode scanner, insider transactions (90d net), 2026-06-02
Biggest Buyer
Dominion +$9.7MRegulated — the slowest clock
Biggest Seller
Quanta −$481MGrid momentum name, already run
The Pattern
Buy slow, sell fastRotation into the back-loaded layer
Chapter 06
The Read: A Coiled Spring Waiting for a Capex Print
The model rates the entire power layer between 37 and 52 out of 100 — GE Vernova 37, Vistra 39, Duke 44, Vertiv 47, Carrier 52. Nothing is broken; nothing is compelling. That is the signature of a coiled layer: priced for nothing, waiting for a catalyst. The catalyst is dated — Broadcom prints Wednesday June 3 after the close, and summer grid-load data builds through July.
Step back to the composite — the blended 0-to-100 read across momentum, value, quality, and risk — and the picture is not bearish, it is coiled. GE Vernova scores 37, Vistra 39, Constellation 41, Quanta 42, Duke 44, Vertiv 47, Carrier 52. The whole layer sits in a tight neutral band: nothing the model flags as broken, nothing it flags as a screaming buy. That is what "priced for nothing" looks like — a spring under light tension, waiting for a reason to move. The reasons are on the calendar. Broadcom reports Wednesday June 3 after the close; if it confirms that AI capex is accelerating again rather than plateauing, the read-through to power demand is direct, and the layer that runs on that capex gets a fresh catalyst to reprice. Beyond this week, summer arrives — the season when data-center load and air-conditioning demand stack on the same grid, the clearest real-world test of the scarcity thesis. And rate-case decisions and new long-term supply contracts trickle out through the quarter, each one converting "back-loaded" into "booked." None of that moves the tape today. But it is why the honest verdict on the AI power trade is not "cold, avoid" — it is "cold, and the people who can see the contracts are buying. Watch the capex print."
Composite rank across the power basket
Source: MarketDecode composite score (0–100), 2026-06-02
Next Catalyst
AVGO · Wed AMCDoes AI capex accelerate again?
Slow-Build Test
Summer loadData-center + cooling demand stack
Model Read
Coiled, 37–52Priced for nothing — waiting
Resolution window — 1 month
What would confirm or invalidate this read
Confirmation
Within 30 days the AI-power layer begins to reprice off the catalyst: Broadcom's June 3 print confirms AI capex is accelerating again, and by early July the IPPs (VST, CEG) close at least 10% of their gap to mean target and/or the basket median 14-day RSI moves above 55. Insider net buying in the regulated names (D, AEP, DUK) continues, and Vertiv holds its ~28% forward-growth estimate. That sequence would confirm the "cold = timing discount, not demand verdict" thesis.
Invalidation
The basket stays cold or gets cheaper even after Broadcom confirms capex — median RSI remains below 45, the IPPs make fresh 30-day lows, and analyst targets are cut rather than chased (net downgrades across the regulated names). That would mean the discount is structural — a regulated, low-multiple, hyperscaler-bargained business the market is correctly refusing to re-rate — not a lag. A break of VRT's growth estimate below ~20% next year would remove the one clean grower and weaken the bull case further.