WATTS THE HOLDUP?

(DUK), (NVDA), (PWR), (AMZN), (POWL), (EME), (ETN), (URI)

Seven years. That’s how long it takes to connect a new data center to the U.S. power grid right now. Not the glossy brochure timeline. Not the consultant’s PowerPoint fantasy. Seven actual years from application to flipping the switch. 

I learned this sitting in a Duke Energy (DUK) conference room in Charlotte three weeks ago, surrounded by infrastructure planners who looked like they hadn’t slept since ChatGPT launched. One guy kept nervously tapping his pen while explaining interconnection queues. I wanted to tell him to switch to decaf, but honestly, caffeine isn’t his problem. Reality is.

Jensen Huang announced NVIDIA (NVDA) might drop $100 billion on OpenAI data centers requiring 10 gigawatts of power. Everyone lost their collective mind, but nobody mentioned the seven-year queue. 

Nobody talked about transformer lead times stretching to three years. Nobody asked the obvious question: where exactly does Jensen think this power is coming from, and when?

I spent the other month touring a Quanta Services (PWR) transmission project in West Texas. Beautiful operation – crews working in that methodical way that tells you they know what they’re doing. 

But what jumped out to me was the information that the massive transformers they’re installing were ordered 32 months ago. Not 32 weeks. Thirty-two months. 

These things aren’t Amazon (AMZN) Prime eligible. Each one is custom-manufactured, requires specialized rail transport, and needs installation crews who understand high-voltage work in ways that would make your average electrician very nervous. 

Quanta’s backlog looks magnificent, and it is, but Wall Street keeps modeling revenue like these projects run on predictable quarterly schedules. They don’t.

The company trades at 40 times earnings, which sounds rich until you consider the secular tailwinds. But then you remember those seven-year interconnection queues and suddenly 40x starts feeling a bit ambitious. 

I’m not saying Quanta’s a bad company. It’s an excellent company facing execution risks that earnings calls don’t adequately capture. 

Powell Industries (POWL) has the same problem, just wrapped in different packaging. They’ve transformed from a cyclical oil and gas play into a legitimate power distribution growth story. Their backlog tripled in 18 months. 

The market noticed – stock’s up over 300% from 2023 lows. Fantastic. But I’m hearing from their actual customers that mid-project specification changes have become routine. 

Utilities are scrambling to accommodate AI power loads they never forecasted, which means Powell keeps redrawing plans halfway through fabrication. That creates margin pressure nobody’s modeling yet. Great long-term story, but short-term volatility is baked in like a certainty.

EMCOR Group (EME) is the one I’m watching most carefully. Last quarter, I attended an invitation-only data center summit in Dallas. EMCOR’s VP of critical facilities walked through their modular construction approach: building major components off-site in controlled environments, then assembling on location in weeks instead of months. 

They’re solving the labor shortage differently than competitors, which matters enormously when interconnection windows are this constrained.

But even EMCOR can’t escape physics. The specialized HVAC systems required for AI server cooling are in short supply globally. Lead times that were 12 weeks in 2023 are now 40 weeks. 

I confirmed this with two of their equipment suppliers who seemed genuinely apologetic about it, like they’d personally invented the shortage. The company’s revenue growth will be spectacular. That’s not the question. 

The question is whether 31 times earnings adequately discounts the inevitable execution slippage. My answer is no.

Every utility executive I’ve talked to in the past six months sounds overwhelmed. Duke Energy alone is evaluating 7.3 gigawatts of new capacity requests, mostly from data centers. 

But utilities operate under regulatory constraints that force them to prove demand before building. They cannot risk stranding ratepayers with unused infrastructure if AI growth disappoints. 

This creates a fundamental mismatch: tech companies need power yesterday, utilities need regulatory certainty before spending billions, and the approval process moves at the speed of continental drift.

Overall, the AI infrastructure thesis is correct. We will absolutely spend trillions upgrading America’s grid. Companies like Quanta, Powell, EMCOR, Eaton (ETN), and United Rentals (URI) will benefit enormously. 

But timing matters, and right now these stocks have priced in several years of flawless execution in a world where flawless execution is physically impossible. 

The smart move is building watchlists, not deploying capital at valuations that assume everything goes perfectly on schedule.

Seven years to interconnection. Remember that number when you’re tempted to chase these stocks at current prices. It’s your edge for finding better entry points in the best infrastructure story of our generation.