AI Power Stock Divergence Intensifies as Investors Shift Toward Near-Term Certainty
N.R. Finch
The AI power trade has split — investors no longer pay up for decade-out stories. Whether a company can book real revenue before 2030 is now the valuation dividing line.
The "anything data-center" rally — why did it stop working?
Until recently, any stock linked to data-center power rode the AI valuation wave.
Now investors draw a sharp line: companies shipping today versus companies that won't deliver for a decade.
This means → the sector has moved from "rising tide lifts all" to "show me the product" — the valuation logic has changed.
Far-out stories are getting punished — who fell hardest?
Nuclear-reactor developer Oklo, backed by OpenAI's Sam Altman, has signed over 14 GW of power agreements — but still lacks key permits, let alone a built plant. Since its brief peak above $25 billion market cap last October, the stock has dropped roughly 75%.
Uranium-fuel producer Standard Nuclear completed its IPO last week but had to cut its fundraising target on weak demand, raising just $150 million. Shares have since fallen 36% below the offer price.
In plain terms = big customer lists and celebrity endorsements no longer prop up valuations when the product hasn't reached the ground.
Has anyone been unfairly caught in the sell-off?
Geothermal company Fervo Energy is under pressure too, but Jefferies analysts argue lumping it with advanced-nuclear names "isn't fair."
Fervo's first plant is expected to come online this year and reach 100 MW capacity by early 2027 — a timeline far shorter than any nuclear project.
This means → a broad sentiment sell-off can hit everyone equally, but differences in delivery schedules will eventually re-split valuations.
Even the sector's best seller saw its stock drop?
Gas-turbine maker GE Vernova reported orders up 88% year-over-year this week — a strong result.
But the company also announced plans to expand capacity through 2030, and the stock fell.
This reflects a market that is hypersensitive to "earning money now" and treats "spending money to grow later" as a negative — expansion means cash out; the market only wants to see cash in.
How does Wall Street sum up this shift?
BNP Paribas analyst Moses Sutton put it this way: "Investors are willing to pay for narratives that play out before 2030, before 2032. They don't want to assign much weight to what happens in the 2030s."
Melius Research analyst James West was blunter: "At some point you have to ask: who is actually selling something to data centers today?"
In plain terms = Wall Street has landed on a simple screen — can you show real revenue before 2030? If yes, stay. If no, move on.
Will the uncertainty around far-out names keep growing?
Reciprocating-engine supplier Innio has also weakened since its recent listing. Its capacity-expansion plans carry outsized risk if demand disappoints.
Data-center construction itself faces bottlenecks including skilled-labor shortages, and actual build-out has slowed below the most optimistic forecasts.
This means → the core mismatch is clear: new power technologies have commercialization timelines longer than data-center build cycles — and those build cycles are themselves slowing. The two schedules are out of sync, compressing certainty around long-dated power demand even further.
Content is for reference only, not financial advice.