Elon Musk did something unprecedented in Memphis, Tennessee, last year.
He built one of the world’s most powerful AI supercomputers (100,000 cutting-edge GPUs) in a matter of months.
But here’s the twist: he didn’t wait for the local utility to connect it to the power grid. Instead, he parked a fleet of mobile generators on-site and fired them up immediately.
This wasn’t a backup plan. This was the plan.
And it’s becoming the new normal across the entire AI industry.
Why Speed Became More Valuable Than Efficiency
Here’s a number that explains why Musk, and every other tech giant racing to build AI infrastructure, is willing to do whatever it takes to get power immediately: $12 million per megawatt, per year.
That’s the estimated annual revenue a single megawatt of AI computing capacity can generate.
For a 100-megawatt data center campus, we’re talking about $1.2 billion in annual revenue potential.
Now consider this: the average wait time to connect a new data center to the electrical grid in the United States is 8+ years.
Do the math. A one-year delay on that 100 MW facility? That’s roughly $1 billion in lost revenue.
A five-year delay? You’ve just watched $5 billion evaporate while you wait for utility bureaucrats to process paperwork.
Suddenly, paying a premium for expensive, fuel-hungry generators that can be up and running in weeks instead of years doesn’t seem so crazy.
The “inefficiency penalty” might cost you $500,000 per megawatt annually in extra fuel costs, but you’re making $12 million. The difference is absurd.
This is why traditional metrics like “cost per kilowatt-hour” and “thermal efficiency” have been thrown out the window. In the AI era, there’s only one metric that matters: time to power.
Time to Power is the time it takes for a new data center to receive the necessary power from the grid. This process can take years, leading to delays and increased costs for companies waiting to become fully operational.
Why the Grid Can’t Keep Up
The U.S. electrical grid was already stretched thin. AI demand just exposed how unprepared it really was.
The demand side is exploding. A traditional server rack in a data center draws about 5-10 kilowatts of power. An AI rack packed with NVIDIA’s latest chips? Try 60-132 kilowatts or more. We’re talking about 10-20x the power density in the same physical footprint!
The supply side is collapsing. Old coal plants are being retired faster than new generation comes online. The electrical transmission lines connecting power plants to cities are already maxed out. And the regulatory approval process for new infrastructure moves at a glacial pace….measured in YEARS, not months.
The result? A massive, multi-year bottleneck.
Utility companies have “interconnection queues” that have waiting lists that would embarrass a DMV. Projects that submitted applications in 2018 are still waiting for approval in 2026!
For a hyperscaler like Amazon, Google, or Microsoft engaged in an existential battle for AI dominance, this timeline is completely unacceptable.
Waiting five years for power while your competitor spins up their AI infrastructure today is a death sentence.
The Unconventional Solution
So what do you do when you can’t get power from the grid? Simple: you bring the power plant to you.
This has sparked one of the most fascinating supply chain pivots in modern industrial history. Tech companies are sourcing power generation equipment from the most unlikely places.
Here’s one example: jet engines.
Yes, the same turbine technology that powers a Boeing 767 across the Atlantic is being modified to generate electricity for AI training clusters.
These aerospace-derived generators can be installed and operational in a fraction of the time it takes to get a grid connection approved.
But that’s just the beginning. Data centers are also turning to equipment originally designed for oil fields, ocean vessels, and industrial facilities.
Technologies that were never intended for this purpose but happen to have the one characteristic that matters most: speed.
These aren’t elegant solutions. They’re not cheap. They often burn more fuel and emit more carbon than traditional grid power. But they have one overwhelming advantage: they can be deployed in MONTHS, not years.
And in an industry where every month of delay costs tens or hundreds of millions of dollars, that speed is worth almost any price.
The AI Trade You’re Missing
While everyone is piling into NVIDIA, AMD, TSM, ASML, Sandisk, Micron, and other semiconductor stocks, a quiet industrial boom is happening in the background.
Companies that manufacture internal combustion engines and other industrial equipment are seeing unprecedented demand.
We’re not talking about trendy startups or speculative tech plays. These are old-school industrial manufacturers.
Many of them trade at reasonable valuations because the market hasn’t fully priced in this structural shift yet.
Here’s what makes this opportunity particularly compelling:
It’s not a short-term trend. Grid interconnection timelines aren’t improving, they’re getting worse. The backlog is projected to persist through 2030 and beyond. This means the “temporary” solutions being deployed today will be running for years, generating recurring revenue through maintenance contracts, fuel supply agreements, and equipment upgrades.
The economics are locked in. As long as AI computing generates $10-12 million per megawatt annually, data centers will pay almost any price for immediate power. This gives equipment suppliers extraordinary pricing power.
The addressable market is massive. AI workloads could represent half of all data center operations by 2030. We’re talking about hundreds of billions of dollars in infrastructure investment. And a meaningful chunk of that is going to power generation equipment.
The Companies Winning the On-Site Power Rush
There are several publicly traded U.S. companies with significant exposure to this trend.
They range from massive industrial conglomerates diversifying into data center power to pure-play specialists that have essentially become “picks and shovels” for the AI gold rush.
The common thread? They all have the ability to deliver power fast, and they’re capitalizing on the grid crisis in different ways.
This isn’t speculative. Major contracts worth billions of dollars have already been signed.
For example, let’s look at GE Verona (GEV):
GE Vernova sells mini power plants that data centers can run on-site to get all the electricity they need without waiting for the local power grid.
GEV has transitioned from consolidation into renewed trend expansion.
The recent sharp move higher suggests that institutional buyers are stepping in. The price is moving more strongly now than it did during the earlier sideways phase, which often signals increased participation and conviction.
As long as higher lows continue to form above prior breakout levels, weakness is an opportunity to buy the dip. You could look to buy GEV on a pullback into the previous breakout level and position for trend continuation.
Subscribe to Babypips Premium to get our full analysis, including:
- Detailed profiles of seven stocks positioned to benefit.
- Breakdown of each company’s data center revenue exposure.
- Technical comparison of equipment types and deployment timelines.
- Risk analysis and regulatory considerations.
The AI revolution isn’t just about software and chips. It’s about the unglamorous, capital-intensive infrastructure that makes it all possible.
And right now, that infrastructure is being built with equipment from the most unexpected industries.
👉 Subscribe to Babypips Premium today.
Feed from Babypips.com