Data Centers in Space!

And what that means for earthly energy markets

In February, Elon Musk merged SpaceX and xAI into a $1.25 trillion company. The stated reason: to build data centers in space. He filed with the FCC for permission to launch up to one million satellites as orbital compute nodes. At Davos, he called it a “no-brainer.”

Nice! The man on track to become the world’s first trillionaire is betting a meaningful chunk of that trajectory on the premise that it will soon be cheaper to launch a GPU into orbit than to plug it into the grid in Virginia. And this isn’t just the man saying this, this is a bunch of very smart wall street people going like “yea that’s a trillion dollar business for sure”.

He is not alone. Google is testing radiation-hardened AI chips for a project called Suncatcher: an orbital AI cloud, prototype launching 2027. Sam Altman reportedly explored buying a rocket company to put OpenAI’s compute in space. A Y Combinator startup called Starcloud already launched an NVIDIA H100 into low Earth orbit in November 2025.

That does that mean for us mere earthly energy consumers??

If the demand class that needs the electricity can afford to pay almost anything, it means prices may go up for everyone else. At least until the point where it’s worth it to shoot all of this infrastructure into SPACE.

This is the core insight behind the Compute Heat Rate™ (CHR)¹, a framework I developed to quantify the maximum electricity price that AI workloads can profitably sustain. The math is straightforward: take the revenue a GPU generates per megawatt-hour of electricity consumed, subtract the non-electricity operating costs, apply a required return margin. What you get is a price ceiling.

For frontier AI inference the ceiling is over $53,000 per MWh. For mid-tier inference, around $8,000.

The current average wholesale electricity price in PJM? About $50.

For now, electricity is a rounding error in the economics of AI.

Why Space Actually Might Make Sense

Starcloud claims orbital energy costs work out to roughly $0.002 per kWh. That is the marginal cost of sunlight hitting a solar panel that is already in orbit. It does not include the cost of launching hardware at $2,000 to $5,000 per kilogram, the satellite bus, the solar arrays, radiation-hardened components that degrade in five years, or the fact that you cannot send a technician to fix anything. The honest all-in cost per MWh of orbital compute is obviously dramatically higher than $0.002/kWh.

And that is the entire point. Even if the true all-in orbital cost is $200, $300, or $500 per MWh, it still pencils out, because the revenue per MWh of frontier AI inference is huge.

That is the CHR thesis in a single image: a company literally leaving the planet to avoid terrestrial grid bottlenecks, because the economics of AI compute are so favorable that shooting them into space makes sense.

The Orbital Ceiling Principle

For fun, let’s start referring to this as the Orbital CHR Ceiling. If the all-in cost of generating and consuming one MWh of electricity in low Earth orbit (including amortized launch costs, satellite depreciation, solar array maintenance, and latency penalties) is, say, $1,000 per MWh, then that becomes a theoretical upper bound on what any rational terrestrial data center operator would pay for grid electricity.

Why pay more than $1,000 on the ground when you could pay less in space/

But here is the thing: $1,000 per MWh is still twenty times the current wholesale price. And it is roughly eight times higher than CONE (the Cost of New Entry for building a new gas plant). Which means we have a very long runway between “where prices are now” and “where it becomes cheaper to literally leave Earth.”

What This Means for People Who Don’t Own Rockets

If you are an aluminum smelter in Virginia, your electricity tolerance is about $60 to $80 per MWh. If you are a chemical manufacturer, maybe $100 to $160.

The AI data centers being built next door to you can sustain prices hundreds or thousands of times higher. And they are not going anywhere. Their load is non-curtailable: 99.99% uptime SLAs with liquidated damages. They are contractually obligated to consume power 24 hours a day, 365 days a year.

When those data centers push the clearing price above your margin threshold, you have two options: absorb the loss, or shut down.

The data center’s third option is apparently to leave earth’s gravitational well.

If the cost of electricity almost does not matter to the fastest-growing source of electricity demand on the planet, then the rest of us need a new framework for understanding where prices are going.

The framework is this: computeheatrate.com.

More soon.

Hans Royal is the originator of the Compute Heat Rate™ (CHR) framework. All views are his own and do not represent those of any employer or affiliated organization.

1. Royal, Hans, The Compute Heat Rate: Quantifying AI-Driven Electricity Price Tolerance
and Its Implications for Wholesale Market Repricing (February 28, 2026).
Available at SSRN: http://dx.doi.org/10.2139/ssrn.6322318