AI, not in the cloud... in space

Welcome to AI Collision 💥,

In today’s collision between AI and our world:

Can’t cool, can’t scale
Data in space
Queenie beatboxing

If that’s enough to get the rockets launching, read on…

AI Collision 💥

“In computing, the laws of thermodynamics are the laws of scaling. If you can’t cool it, you can’t run it — and if you can’t run it, you can’t scale it.”

The world’s data centres already consume more electricity than entire nations.

McKinsey thinks that by 2030, that figure will triple. And we could be looking at 12% of electricity demand globally just coming from data centres.

The reason for this isn’t the foundations of the internet (lovable cat videos) but the insatiable, power-hungry, GPU-rammed AI factories that will underpin our world for the next thousand years.

So, what happens if here on earth, we can’t keep up?

What if we can’t get the energy to scale, we can’t get the infrastructure to scale at the speed in which AI wants to expand?

The answer is we go beyond earth…

This week, Nvidia dropped one of the most radical announcements in the history of computing — and barely anyone outside the tech world took notice.

They unveiled Starcloud, a startup from their Inception program, building data-centre satellites.

In short, AI compute in orbit.

Their first spacecraft, Starcloud-1, will carry a 60-kilogram module powered by an Nvidia H100 GPU. It’s small (fridge-sized) but has 100× the compute power of any satellite that’s ever flown before. And that’s just the start.

The company’s long-term plan? A 5 GW orbital data-centre cluster spanning four kilometres of solar panels and radiators.

If that sounds crazy, give me a second to lay it out, because what’s crazy sounding is actually quite a logical idea.

On Earth, AI’s biggest cost isn’t the chips, it’s the energy and cooling that feed them.

In orbit, both are free.

Above the clouds, sunlight is constant and unfiltered. It’s why astronauts have solar protected outfits, and are required to be out of the direct energy coming from the sun.

There’s no night, no weather, no intermittency. In space, solar panels are up to five times more efficient than their Earth-bound cousins. Energy costs that hover around 10 cents per kWh on the ground could drop to 0.1 cents in orbit — a 97% reduction in operating costs once built.

Cooling?

Well forget air-conditioning or water. Space itself is a freezer.

The vacuum acts as a perfect heat sink, letting radiators dump excess heat straight into the void.

Starcloud says its orbital cooling arrays could cut total energy consumption by an order of magnitude. And there’s no need for water, whereas a single hyperscale data centre on Earth can consume up to a million litres a day just for cooling.

Infinite clean power, free cooling, and no land footprint.

The economics work. The only catch is the cost to get these things space-ready and into orbit. But even that’s looking like a more economic prospect now.

For decades, the idea of space-based compute was a thought experiment for futurists and physicists. But the economics have flipped.

SpaceX’s reusable rockets have driven launch costs from $20,000 per kg to near $10 per kg. Optical satellite networks like Starlink are now achieving 30ms latency which is faster than many terrestrial broadband networks. And AI workloads, once latency-sensitive, are increasingly about batch inference and training, where distance doesn’t matter.

That’s a key point.

You don’t need to run ChatGPT in orbit. But you can train it there.

Massive, GPU-intensive workloads that chew through energy for months on end are perfect candidates for space compute.

And that’s where Nvidia’s involvement becomes pivotal. They’re seeding an entire off-planet compute economy, from energy generation to cooling, to the AI hardware that powers it all. And of course it’s Nvidia chips doing the grunt work.

Starcloud is the proof of concept. And one day maybe it goes public. The real play here for investors is the AI cloud above the cloud.

Data once lived on floppy disks, then in server rooms, then in the “cloud.”

So is it so wild, with economics that now make sense, that the next step for data is in space?

The implications are staggering:

Sustainability: zero-emission compute that doesn’t drain rivers or grids.
Security: orbital infrastructure is isolated from terrestrial attacks and natural disasters.
Scalability: no zoning laws, no cooling constraints, infinite expansion potential.
Latency: with optical links, orbit-to-Earth round trips could drop to 20ms, good enough for most interactive applications.

It’s not just Starcloud. Lonestar Data Holdings is already running servers on the International Space Station. Axiom Space is building the first commercial orbital data module, scheduled to dock in 2026. Thales Alenia and the European Space Agency are funding ASCEND, a plan for a 1 GW space-based data centre by 2050.

And of course, Jeff Bezos has his eyes on it too. He recently told Italian Tech Week this month that gigawatt-scale data centres in space are “not just possible, but inevitable.”

You can see where this is heading.

AI is hitting physical limits. Energy, water, and regulation are becoming issues here on Earth. In space, those issues disappear.

And the companies that solve this, that can launch, power, cool, and communicate with space compute, will define the next big investment frontier.

Yes, we’re in the midst of an AI-infrastructure revolution but just remember it’s not only on earth where this revolution will take place.

Most People Will Miss It.

They missed the dot-com boom…

They missed the crypto wave…

And now, they’re too distracted by recession headlines to notice what’s really happening.

A new “Wealth Window” could be opening and Britain may be right at the centre of it.

Don’t be the one who misses out again

Capital at risk.

Boomers & Busters 💰

AI and AI-related stocks moving and shaking up the markets this week. (All performance data below over the rolling week).

man in black suit jacket and black pants figurine

Boom 📈

SK Hynix (KOR:A000660) up 16%
AMD (NASDAQ:AMD) up 9%
Micron (NASDAQ:MU) up 8%

Bust 📉

Rigetti (NASDAQ:RGTI) down 28%
D-Wave (NYSE:QBTS) down 25%
IREN (NASDAQ:IREN) down 20%

From the hive mind 🧠

The end of the AI boom. The more you see these kinds of articles, the more you should be preparing to invest. There is no end of the boom. There will be pauses, pullbacks, but the boom is not over… it’s barely started.
This tanked Google’s stock when the announcement dropped. And then people figured out it was Chromium anyway with ChatGPT plugged in. The browser wars are here… but also still dominated by Google.
How comfortable would you be with your “poo data” sitting in some random database in the cloud somewhere?