Hey there, bargain hunter. There is a phrase bouncing around trading desks right now that is worth burning into your brain: you can’t print more electrons. And no amount of venture capital, hyperscaler capex, or Federal Reserve rate cuts changes that physical fact.
Here is what is happening. Vistra (VST) and Constellation Energy (CEG) – two of the most direct plays on the AI power buildout – have both pulled back 25–30% from their all-time highs. Meanwhile, the underlying demand thesis has not weakened. It has accelerated.
The Numbers That Matter
Vistra is projected to grow adjusted earnings by 77% in 2026, climbing from $5.26 per share in 2025 to an estimated $11.44 per share. The company reaffirmed its 2026 outlook when it reported on May 7, and its longer-term 2027 EPS estimate moved higher following that release. Vistra has long-term agreements to supply electricity from its nuclear plants to Amazon Web Services and Meta – real revenue, locked in, not a promise.
Constellation reaffirmed its own 2026 EPS range on May 11 and guided for 20%-plus EPS growth from 2026 through 2029. CEG raised its dividend by another 10% in 2026. The company completed its acquisition of Calpine in January, creating what is now the largest independent power producer in the country – roughly 55 GW of capacity, with nuclear at the core.
Slight tangent, but it matters: Constellation’s long-term nuclear PPAs with Meta and Microsoft aren’t just contracts – they’re decade-long revenue locks that most utility analysts are still underpricing in their models.
Why the Pullback Is Interesting
CEG’s stock has fallen roughly 25% from its October 2025 high of $413. The catalyst was a 2026 EPS guidance range of $11–12 that came in slightly below consensus. At a lower multiple, the stock now trades in a range where 20%+ annual earnings growth looks genuinely cheap relative to what software names with half the growth are carrying.
Vistra, at roughly 10.6x EV/EBITDA, trades at a meaningful discount to Constellation. If you want operating leverage tied to Texas power demand and hyperscaler PPAs without paying the full CEG premium, VST is where that tradeoff lives.
The core issue is simple: training a single large AI model requires as much electricity as 10,000 U.S. homes consume in a year. Multiply that across thousands of simultaneous workloads and billions of daily inference calls. Nuclear provides the 24/7 baseload reliability that natural gas and renewables cannot match at that scale. That is not a 2026 story – it is a decade-long structural mismatch.
What I’m watching: PJM transmission bottlenecks for CEG’s Crane restart timeline, Vistra’s Calpine natural gas synergy execution, and whether any hyperscaler announces a new direct PPA in the next 90 days. Any one of those moves the stock.
The bears will say valuations are still elevated on an absolute basis and that a demand miss from one of the big tech players could reprice the whole sector fast. That is fair. But if the AI capex cycle is real – and the evidence says it is – power producers with contracted revenue and limited new supply are not where you want to be underweight.
