The surprising economics of orbital data centres — and the real solution

There has been a growing debate about putting AI data centres into space.

AI needs enormous amounts of power. Space has constant sunlight for solar power. And if launch costs keep falling, maybe it will finally make sense to move the data centres into orbit.

Until recently, this could be dismissed as science fiction. Today, it deserves to be taken seriously — but only if we follow the economics all the way through.

When you do, the answer turns out to be very clear, and very different to what is being discussed.

Why is this question being asked now

This conversation is not driven by AI. It is driven by launch costs.

For most of the space age, lifting large amounts of mass into orbit was prohibitively expensive. That constraint has changed dramatically in the last decade.

Launch costs to low Earth orbit have followed a steep and dramatic decline: from more than US$40,000/kg historically to US$2500–3000/kg today and targeting US$100–300/kg in SpaceX’s new Starship

To stay conservative, this analysis assumes US$150/kg to LEO — not a promise, but no longer a fantasy.

That single shift turns space from an exotic environment into something closer to infrastructure.

The data centre energy stack

To ground this discussion in reality, consider a facility like xAI’s Colossus, operating at roughly 300 MW of continuous power.

The current “best possible” energy stack is a mix of onsite gas turbines, grid connections, a small amount of solar and a few batteries for smoothing

Some of that power is delivered via the grid, some via on-site generation. For a true cost comparison, we can treat the energy stack as if it were fully dedicated to the site.

The cost of building the energy stack is around US$550–1050M

Plus annual maintenance and fuel costs of US$100–180M a year

Gas is not a backup in this model. It is structural.

Also Read: Breaking into the data centre sector: Beyond technical expertise

Why look to space at all?

Because AI needs power at scale, and it needs it to be stable, and we need a route there that doesn’t depend on extracting and burning ever more fossil fuels.

Solar is an obvious solution; however, on Earth, even excellent solar installations deliver only 25–30 per cent of their theoretical output over a year. Solar in orbit benefits from constant sunlight 40 per cent stronger than on the surface of the earth and is effectively firm by default. There is no night, no weather, and no seasonal variation. Once built, it can be 100per cent solar without fuel or large storage.

That single difference is what makes space interesting.

What does a 300 MW space-based solar energy stack weigh?

The cost to get a solar plant in space is the cost per kg we discussed before times the number of kgs it weighs. Modern space-solar designs use ultralight photovoltaic membranes rather than glass-and-steel terrestrial panels. With no wind or gravity, structures can be far lighter.

Consensus estimates a conservative near-term figure of 0.8 kW per kilogram of photovoltaic material is plausible.

At that density, 300 MW requires ~375 tons of panels.

Even in space, you still need structural support, wiring, power electronics, and control systems. These add mass, though far less than on Earth.

Using optimistic but defensible assumptions, non-panel components add roughly one to two times the panel mass.

That puts the total mass required to generate 300 MW in orbit at approximately 750–1,100 tonnes.

At US$150/kg to LEO, and another 15–20 per cent to raise to GEO, it is expensive, but single-time — and crucially, it buys something Earth-based solar cannot: firm power without fuel.

These figures reflect a post-industrialised SBSP cost regime; today, a kind of 300 MW GEO system would cost over a billion dollars, but with repeat builds and learning-curve effects, these ranges are plausibly achievable within ~10–25 years.

Annual operating costs are minimal:

Annual cost of operations and maintenance: US$3–6M / year

No fuel. No price volatility.

Also Read: The AI age is changing the data centre industry – Here’s how Singapore can pivot

What about the data centre?

At this point, now that we know that moving the energy stack to space is feasible, we can look at moving the data centre itself.

This is when the numbers break.

A data centre is not just chips. It also comprises power electronics, cooling systems, structural containment, cabling, and radiation shielding. Even reducing the weight of the structure for zero gravity, we’re looking at a 300 MW AI data centre of 13,000–15,000 tons.

Plugging in our conservative near-term launch costs of ~US$150/kg to LEO, that implies:

• US$1.95–2.25 billion to launch the data centre, before orbital transfer.

And a second factor has to be added: unlike the solar infrastructure, this cost is not one-time.

Chips are replaced every three to five years. That means most of the compute mass would need to be relaunched on that cadence.

No plausible launch-cost trajectory fixes this asymmetry.

That is why putting compute in orbit fails economically — even in a world where space-based energy begins to make sense.

The pivot the numbers force

Once it becomes clear that data centres are too heavy and too short-lived to move economically, the problem reframes itself.

The thing that should move to space is the energy stack.

The thing that should stay on Earth is the computer.

Beaming power is real technology: The basic architecture is straightforward: collect sunlight in space, convert it to microwaves, beam it at low intensity, below that of radar, to a large “rectenna” on Earth, which is a simple large mesh and convert it back to electricity. No exotic physics or speculative materials are required; power beaming has been demonstrated terrestrially and at a small scale in space.

End-to-end efficiency is not 100 per cent. A reasonable near-term assumption is that only about two-thirds of generated power reaches the data centre, which means the orbital solar array must be oversized by roughly 1.5×.

For a 300 MW continuous data-centre load on Earth, that implies ~450 MW of space solar generation, plus transmission hardware.

Also Read: Is Southeast Asia’s data centre boom headed for a PR crisis?

Adding transmission capabilities and increasing the capacity to 450MW changes our costs: (again, after price optimisation, not first of a kind today, which would be two to three times the cost for a prototype)

The logical conclusion

Although the upfront cost of space-based solar is higher, the difference in annual fuel and operating cost is large enough to repay that in a handful of years.

And with no future fuel cost risk.

Why isn’t everyone doing this? Because launch costs only crossed the threshold very recently. Even a 2024 NASA study still assumed Falcon 9–era economics.

And politically, it’s easier to talk about “AI in space” than “power beamed from orbit.”

But narratives follow incentives.

As launch becomes cheap and power demand explodes, the industry will pivot.

Not to data centres in space.

But to something far more powerful: Data centres on Earth, powered by cheap, stable, solar energy from space.

That’s the real solution hiding in plain sight.

—

The post The surprising economics of orbital data centres — and the real solution appeared first on e27.