AI’s Dirty Secret: The Power Grid Can’t Keep Up
By Cash Flow University · · 5 min read
AI isn't being capped by chips anymore — it's being capped by power. With 945 TWh of projected data center demand by 2030, a 2,600 GW grid backlog, and Big Tech pouring $300B+ into AI capex, the real trade isn't in AI apps — it's in power producers, nuclear restarts, and grid infrastructure.
The Story
This week, I noticed the conversation shift — from "AI is expensive" to "AI needs electricity at scale."
That's a different problem. And it changes where the money flows.
When the political narrative becomes "don't raise consumer bills to feed data centers," the next move is predictable: the biggest buyers of compute start paying directly for the energy side of the stack. They stop waiting for the grid and start building around it.
I've been watching this shift for months. The signals are everywhere if you know where to look — earnings calls, PPA filings, permitting backlogs. And what I keep coming back to is this: the bottleneck isn't silicon anymore. It's watts.
Why This Matters
Three forces are converging at the same time:
When those three forces combine, the winners are not always the most famous AI names — they're often the "picks and shovels" behind the scenes.
By the Numbers
I like to anchor my theses in data, not hype. Here's what the numbers actually say about the AI-power collision:
Read that table again. Demand is projected to grow over 5x in seven years, while 2,600 GW of generation is stuck in the interconnection queue waiting to come online. That's not a gap — that's a chasm. And chasms create asymmetric opportunities.
The "Shadow Power Grid" Thesis
If the grid can't deliver power fast enough, hyperscalers do one of three things:
- Sign longer-term power agreements (PPAs) — locking in supply at predictable rates
- Backstop new generation buildouts — financing plants that wouldn't otherwise get built
- Pursue on-site or dedicated power solutions — bypassing the grid entirely
That creates a new investable narrative: "AI is a power trade."
The Nuclear Card
Here's what sealed this thesis for me: every major hyperscaler is independently making the same bet — nuclear.
I've been tracking these deals as they've rolled in, and the scale is staggering:
- ▸ Microsoft signed a 20-year deal to restart Three Mile Island's Unit 1 reactor — 835 MW, targeting 2028. The deal is reportedly worth $16 billion over the contract life.
- ▸ Google inked the first-ever corporate SMR (small modular reactor) fleet deal with Kairos Power — 500 MW targeting 2030 and beyond.
- ▸ Amazon launched "Project Spectrum" — a $5B data center campus collocated next to an existing nuclear plant in Texas, and contracted 20.4 GW of clean energy in 2025 alone, including 4.7 GW of nuclear.
- ▸ Meta issued an RFP for 1–4 GW of new nuclear capacity to power its AI ambitions.
☢️ When the four largest tech companies in the world are all independently betting on nuclear, that's not a trend — it's a thesis.
These aren't speculative pilot projects. These are multi-billion dollar, multi-decade commitments. They tell you exactly how serious the power constraint is — and how much capital is going to flow into the companies that can deliver baseload energy at scale.
Who's Actually Buying the Power?
📊 The PPA Concentration Signal
The power purchase agreement (PPA) data tells a story that most people are missing:
- ▸ Big Tech accounted for 49% of all global corporate clean energy deals in 2025 — up from 37% the year prior.
- ▸ 2024 saw a record 68 GW of corporate PPAs globally — a 29% year-over-year jump.
- ▸ Overall PPA volumes dipped ~10% in 2025, but Big Tech's share grew — meaning they're crowding out everyone else.
The concentration is the signal. When four companies consume half the world's corporate energy deals, the downstream beneficiaries aren't hard to find — if you know where to look.
Sources: BloombergNEF (Feb 2026), IEA World Energy Outlook
👀 What to Watch Next
- ▸ New power purchase agreements announced by hyperscalers
- ▸ Interconnection / permitting changes (anything that shortens timelines)
- ▸ Earnings commentary from power producers and grid-adjacent companies:
- ◦ Forward contracting demand
- ◦ Capacity pricing
- ◦ Capex plans
- ◦ Regulatory messaging
How CFU Is Trading It
Inside CFU, we map the full picture:
Want the tickers + the setup logic?
Get full access to our scanners, trade alerts, and elite community. See how we're positioning around the AI power trade in real time.
Join CFU →Disclaimer: This content is for educational purposes only and does not constitute financial advice. Always do your own research before making investment decisions.