The Energy Bottleneck: Why AGI May Starve Before It Thrives

Everyone’s talking about artificial general intelligence (AGI) like it’s a math problem. Like if we just get the algorithms right, the whole world will change. But here’s the truth no one wants to face: AGI won’t fail because it’s dumb. It’ll fail because it’s hungry, and we can’t feed it.

This isn’t a software problem. It’s an energy crisis.

🧠 If this resonates, read my earlier piece: The AI Gold Rush Is Over where I explain why AGI is no longer a research problem, but a resource war

For the last decade, we’ve tricked ourselves. We treated compute like it was infinite. New chips every year, cloud credits handed out like candy, startups burning hundreds of GPU-hours on anime filters and chat apps.

But training GPT-4 took 1.3 gigawatt-hours. The same energy needed to power 1,000 homes for a month. That’s 1 model. 1 run. The subsidies will stop.

Now scale that to real-time AGI:

This is a national grid-scale demand. Thousands of GWh per month, per frontier model, running live.

Zoom out. Here’s the macro picture:

We can’t tax our way out. We can’t inflate our way out. We can’t grow our way out.

The only thing that pencils out is non-human labor. Automation at scale. And fast.

This connects directly to the thesis I laid out before: AGI is an economic necessity. The system collapses without it.

But all that automation needs power. Real power.

Let’s kill the dream gently. Solar is fine for homes, for daytime offsets, for feel-good ESG slides. But for powering AGI? It’s not enough.

So unless you’re planning to cover Nevada in silicon, solar isn’t not enough.

And yet… Elon Musk has been bullish on solar since at least 2022. In his words:

Recently, he’s escalated:

And just last year, he revealed the real frame he’s operating from:

He’s not just thinking about solar farms. He’s thinking Dyson Sphere-adjacent.

In fact, Musk has done the math. As he put it:

But not everyone agrees. Yann LeCun, one of Musk’s most vocal critics, fired back:

The irony is that Musk is in the business of solving that exact problem. Batteries. Lots of them. He’s not hand-waving storage away: he’s quietly building the supply chain to close that gap.

Whatever delivers reliable, scalable energy fastest wins. Whether it’s fusion in orbit or uranium underground.

Musk says you can power the U.S. with solar in a small corner of Texas. And he’s not wrong, on paper.

Yes, the math works. 10,000 square miles of panels could technically generate enough electricity.

But Musk’s model is handwaving the real world:

It’s a fantasy built for spreadsheets, unless Musk knows something we don’t. Maybe he does.

He’s publicly called space-based solar power (SBSP) "one of the dumbest ideas ever", but that might be marketing misdirection. Because if there’s anyone positioned to pull it off, it’s him.

Let’s get serious.

Nuclear is the only credible energy source for the AGI era. Here’s why:

And when it comes to nuclear… Musk’s record is clear:

But nuclear isn’t scaling. Why?

Until these change, AGI has no food.

Could we go around Earth? Sure. It’s called space-based solar power (SBSP).

Put a solar array in orbit. Beam energy down via microwave. No night, no clouds, no intermittency.

Sounds perfect, but there are economic walls we haven't solved yet:

But Musk already owns the stack:

Musk is closer to the federal government than ever, literally BFF with President Trump. SpaceX is indispensable. Starlink is battlefield infrastructure. JD Vance, Peter Thiel’s protégé, is the VP. Thiel and Musk don’t just share ideology, they also share bets. The machinery of state is starting to sync with the Musk stack.

If AGI energy becomes a national security issue, and it wil, Musk can walk into the DoD and say: “I can beam solar anywhere on Earth. I’ve got the hardware, the network, and the launch cadence.”

That’s a Dyson Sphere prelude.

Musk is openly thinking in Kardashev terms:

He’s not just trying to solve the AGI power bottleneck. He’s laying groundwork for a Type I civilization.

Everything he’s building fits:

It’s more than a collection of companies. It’s a Dyson Stack. A vertically integrated, post-nation-state infrastructure monopoly with the capacity to power, network, and shape the cognition of billions, on Earth and in orbit.

He has the regulatory leverage, the cultural capital, and the institutional buy-in to attempt what no sovereign nation has: energy-dominant, orbital-ready civilization infrastructure.

I know this is a deviation from my usual themes about building product… but this is how this all affects you, the engineer, the unconnected builder dreaming of an exit big enough to provide for your family.

You don’t need a rocket. But you better start building like the world you know is ending. Because it kinda is.

Here's your survival stack:

AGI IS coming. Energy is one of the bottleneck, interfaces or reality filters as I call them are the new empires.

🧠 Poll for Builders, Founders, and Operators
I’ve been writing more about how the big players are quietly reshaping the future: Musk, Thiel, state-aligned infrastructure plays, and the energy bottlenecks behind AGI.