The AI Power Trade Destroyed Vistra. The Three Infrastructure Stocks That Actually Deliver.

The market has spent two years treating every utility and power generator as an AI play. Goldman Sachs projects data center power demand surging 220% by 2030. The International Energy Agency says data centers will drive roughly half the increase in U.S. power demand through 2030. Big tech has signed over 2.7 gigawatts of nuclear power purchase agreements alone. The narrative is everywhere: artificial intelligence needs electricity, and anyone who generates it is a winner.

That narrative is wrong. Or rather, it is incomplete in a way that has already cost investors dearly. Not every infrastructure company is built for the AI power surge. Some are built for the theme trade, and the market has already punished them. The difference between the survivors and the casualties comes down to one question: does the company have contracted revenue from the demand, or is it hoping demand shows up?

The False Narrative in Action

I've been very surprised that the market has conflated exposure to data center markets with actual contracted revenue from data centers. The distinction is not subtle. It is the difference between a business model that is already being paid for growth and one that is betting on it.

Vistra is the case study. At its 52-week high of $219.82, Vistra was trading on the assumption that every megawatt of its Texas and competitive-generation fleet would be snapped up by AI data centers. The stock carried a $62 billion market cap, an 8.45x price-to-book ratio, and the market was willing to look past a 341.6% debt-to-equity ratio because the AI power story was too good to pass up.

Today, Vistra trades at $140.59 — down 36% from its peak and down 31% over a rolling 12-month return. The market cap has been cut to $47.4 billion. Revenue growth is 3.8% year-over-year. The dividend yield is 0.65%, which is not a dividend stock; it is a growth stock wearing a dividend costume. That being the case, the thesis has shifted from structural to speculative.

Vistra's free cash flow did jump 219% year-over-year to $1.26 billion trailing twelve months, and its operating cash flow of $5.12 billion is respectable. But capex of $3.86 billion is eating almost all of it. And that debt-to-equity ratio of 341.6% — total debt of $35.7 billion against equity of only $5.6 billion — means leverage is doing heavy lifting on the 172.7% ROE. When market conditions turn, as they have, highly leveraged competitive generators get squeezed first. The AI power story was priced in before the contracts materialized.

What Actually Works

The companies that have survived the theme trade — and the ones that will benefit from the actual structural shift in power demand — share two characteristics: contracted revenue from data center customers, and balance sheets that can absorb the massive capex cycle required to build the generation and transmission infrastructure to serve it.

That filters the field quickly.

Constellation Energy (CEG): The Nuclear Contract Machine

Constellation operates the largest nuclear fleet in the Western world. That is not a growth story — it is a cash flow story. Nuclear power runs at 92.3% capacity factor, producing 40 million megawatt-hours in Q1 2026 alone. It does not need new plants to generate revenue from existing ones. More importantly, it has the only asset class that hyperscalers actually want for continuous 24/7 baseload power.

Microsoft committed $16 billion to restart Three Mile Island, targeting 835 megawatts by 2028. Meta signed a 20-year power purchase agreement for 1.1 gigawatts from the Clinton Clean Energy Center. Google has signed the first U.S. corporate small modular reactor fleet deal with Kairos Power. Amazon is investing over $20 billion in converting Susquehanna for direct data center power. These are not term sheets or expressions of interest. They are binding long-term contracts.

Constellation's Q1 2026 GAAP earnings hit $4.49 per share, up from $0.38 a year earlier. Full-year adjusted operating earnings guidance sits at $11 to $12 per share. Free cash flow is forecast at $8.4 billion for 2026–2027, rising to $11.5–13 billion for 2028–2029. Management also repurchased 1.2 million shares at an average of $285 per share in Q1, which shows confidence in the cash flow trajectory.

The stock trades near $275 as of early August, and the market has pulled back from its highs as Q2 2026 GAAP earnings came in at $1.42 per share, down from $2.67 a year earlier. That sequential weakness is worth watching — but it does not negate the contracted revenue pipeline. The difference between Constellation and Vistra is that Constellation's revenue is already under contract. Vistra's is a hope.

NextEra Energy (NEE): Scale Through the Dominion Deal

NextEra Energy announced a $67 billion all-stock acquisition of Dominion Energy in May 2026, creating the largest regulated utility business by market capitalization. The deal is straightforward: NextEra brings scale, renewables leadership, and execution discipline. Dominion brings Virginia — the corridor that powers the largest data center market in the world.

Northern Virginia alone accounts for roughly a quarter of all data center capacity in the United States. Dominion is the utility that serves it. That geographic moat is not replicable through construction; it requires interconnection capacity, transmission infrastructure, and regulatory relationships built over decades. The NextEra-Dominion combination creates a company with an unmatched position in the single most important power market for AI infrastructure.

NextEra's financials support the thesis, though the capex reality is sobering. Free cash flow of $3.03 billion trailing twelve months declined 19.6% year-over-year. Capex of $10.77 billion is massive, and with the Dominion acquisition adding more, the combined entity will be in a multi-year investment cycle. Total debt stands at $164.6 billion, with net debt of $107.3 billion and a debt-to-equity ratio of 161.7%.

But the dividend tells the real story. NextEra yields 2.8% and has grown its dividend for 24 consecutive years. That is 23 years of consecutive dividend increases — the mark of a company that has navigated multiple cycles of heavy investment and still returned cash to shareholders. The 29.54% operating margin on an 81.67% gross margin structure shows the regulated utility model provides stable operating cash flow even during aggressive expansion. The dividend is funded from operating cash flow of $13.8 billion, which comfortably covers the annual dividend outlay even as capex temporarily reduces free cash flow — a standard feature of regulated utilities whose returns are set by state commissions, not market cycles.

Revenue growth of 10.82% year-over-year is strong for a utility of this scale. The P/E of 19.0 on a trailing basis looks attractive relative to a forward P/E of 30.8, which reflects the market's view that the merged entity has significant earnings ahead of it. The stock is up 5.4% year-to-date and has delivered a 16.3% rolling annual return — outperforming Vistra's collapse.

What About Duke Energy?

Duke Energy yields 3.41% and has 22 consecutive years of dividend growth, making it one of the more attractive income plays among large utilities. Revenue grew 6.33% year-over-year, and the 18.76 P/E looks cheap on the surface.

But Duke is bleeding free cash flow. Trailing twelve-month FCF is negative $4.27 billion, down 659% year-over-year. Capex of $15.84 billion has overwhelmed operating cash flow of $11.56 billion. That is a utility spending more on infrastructure than it generates in cash, with no near-term offset from contracted data center revenue that I could identify. The debt load is $144.2 billion, with a debt-to-equity ratio of 160.5%,comparable to NextEra's.

Duke is a solid regulated utility for income. It is not an AI infrastructure play. The negative free cash flow means it cannot fund its own growth without external financing, and that constrains its ability to capture the data center demand surge. I rate Duke as a Hold for yield-seekers who do not need growth, but it does not belong in the infrastructure-for-AI category.

The Ranking

Of the three companies most frequently cited in the AI infrastructure power trade, I rank them by contracted revenue visibility and dividend commitment — the two metrics that separate structural beneficiaries from theme players.

First: Constellation Energy (CEG) — Buy. The nuclear PPA pipeline from Microsoft, Meta, Amazon, and Google represents billions of dollars in contracted revenue that is already priced into multi-year agreements. The $8.4 billion FCF forecast for 2026–2027 is real, not speculative. The stock pullback from its highs is a function of quarterly earnings volatility, not a change in the underlying thesis. In my opinion, this is the clearest play on AI power demand because the revenue is already contracted.

Second: NextEra Energy (NEE) — Buy.The Dominion acquisition gives NextEra access to the Virginia data center corridor, the single most important geographic market for AI infrastructure power. The 24-year dividend growth streak shows the company can manage through heavy capex cycles. The merged entity will be large and leveraged, but the regulated revenue model provides stability that competitive generators like Vistra cannot match. The 2.8% yield makes it an attractive compounder.

Third: Vistra (VST) — Sell. The stock has already told investors what the market thinks of its thesis: down 36% from highs, 341.6% debt-to-equity, 0.65% dividend yield, and revenue growth of 3.8%. The FCF jump of 219% is a nice headline, but it is a one-year comparison on a base that does not change the structural problem — Vistra does not have the contracted revenue from hyperscalers that its peers have. The market corrected for that. Comparisons with Constellation are not only unjustifiable and irrational; in my opinion, they are irresponsible. One company has signed power purchase agreements worth billions; the other is hoping data centers show up in its Texas market.

The Structural Point

The AI power demand story is real. Goldman's 220% projection through 2030, the IEA's estimate that data centers drive half of incremental U.S. power demand, and the $6.7 trillion in required data center infrastructure investment through 2030 — none of that is fabricated. The demand is structural, not cyclical.

But the investment thesis is not "buy any power company." It is "buy the companies whose revenue from that demand is already contracted." That distinction has already separated winners from losers. It will continue to do so as the capex cycle accelerates and competitive generators face margin pressure from capacity oversupply in deregulated markets.

For investors who want exposure to the AI infrastructure buildout through power generation, Constellation and NextEra are the plays. Vistra is what happens when you buy the theme before you verify the contracts.