Economics of Orbital vs
Terrestrial Data Centers

Motivation and Framing

I love space. I live and breathe it. I'm lucky enough to brush the heavens with my own metal and code, and I want nothing more than a booming orbital space economy that creates the flywheel that makes space just another location we all work and visit. I love AI and I subscribe to maximum, unbounded scale. I want to make the biggest bets. I grew up half-afraid we'd never get another Apollo or Manhattan. I truly want the BigThing.

This is all to say that the current discourse is increasingly bothering me due to the lack of rigor; people are using back-of-the-envelope math, doing a terrible job of it, and only confirming whatever conclusion they already want. Calculating radiation and the cost of goods is not difficult. Run the numbers.

Before we do the classic engineer thing and get nerd sniped by all the shiny technical problems, it's worth asking the only question that matters: why put compute in orbit at all? Why should a watt or a flop be more valuable 250 miles up than on the surface? What economic or strategic advantage justifies the effort required to run something as ordinary as matrix multiplication in low Earth orbit?

That "why" is nearly missing from the public conversation. The "energy is cheaper, less regulations, infinite space" arguments just ring false compared to the mountains of challenges and brutal physics putting anything in space layers on. The discourse then skips straight to implementation, as if the business case is obvious.

Personal Positioning

I'm not here to dunk on anyone building real hardware. Space is hard, and shipping flight systems is a credibility filter. I'm annoyed at everyone else. The conversation is full of confident claims built on one cherry-picked fact and zero arithmetic. This is a multivariable physics problem with closed-form constraints. If you're not doing the math, you're not contributing, you're adding noise and hyping for a future we all want instead of doing the hard work to actually drive reality forward.

Core Thesis

The target I care about is simple: can you make space-based, commodity compute cost-competitive with the cheapest terrestrial alternative? That's the whole claim. Not "space is big." Not "the sun is huge." Not "launch will be cheap." Can you deliver useful watts and reject the waste heat at a price that beats a boring Crusoe-style tilt-wall datacenter tied into a 200–500 MW substation?

If you can't beat that, the rest is just vibes. GPUs are pretty darn happy living on the ground. They like cheap electrons, mature supply chains, and technicians who can swap a dead server in five minutes. Orbit doesn't get points for being cool. Orbit has to win on cost, or it has to admit it's doing something else entirely. If it's an existential humanity play, that's cool too, but it's a slightly different game.

Analytical Lens

So here's what I did. I built a simple model that reduces the debate to one parameter: cost per watt of usable power for compute. The infographic below lets you change the assumptions directly. If you disagree with the inputs, great. Move the sliders. But at least we'll be arguing over numbers that map to reality.

The model is deliberately boring. No secret sauce. Just publicly available numbers and first-principles physics: solar flux, cell efficiency, radiator performance, launch cost, hardware mass, and a terrestrial benchmark that represents the real alternative: a tilt-wall datacenter sitting on top of cheap power. The code is public, please go through everything. github.com/andrewmccalip/thoughts

Findings and Implications

Here's the headline result: it's not obviously stupid, and it's not a sure thing. It's actually more reasonable than my intuition thought! If you run the numbers honestly, the physics doesn't immediately kill it, but the economics are savage. It only gets within striking distance under aggressive assumptions, and the list of organizations positioned to even try that is basically one.

That "basically one" point matters. This isn't about talent. It's about integration. If you have to buy launch, buy buses, buy power hardware, buy deployment, and pay margin at every interface, you never get there. The margin stack and the mass tax eat you alive. Vertical integration isn't a nice-to-have. It's the whole ballgame.

Market and Incentives

Which is why I trend positive on SpaceX here. If anyone can brute force a new industrial stack into existence, it's the team that can reduce $/kg and get as humanly close to free launch as possible. And they need to, because the economics are not close. This is not a 25% mismatch. It's 400%. Closing that is the whole job. Positive does not mean gullible. It needs measurable targets and painful reality checks.

If SpaceX ever goes public, this is exactly the kind of thing shareholders should demand: extreme, barely-achievable goalposts with clean measurement. Tesla did it with the options grant. Do the same here. Pay Elon a king's ransom if he delivers a new industrial primitive: cheap, sustained dollars per kilogram and dollars per watt in orbit, at real cadence, for years.

Broader Interpretation

On strict near-term unit economics, this might still be a mediocre use of capital. A tilt-wall datacenter in Oregon with cheap power, cheap cooling, and technicians on call is hard to beat. Crusoe can park compute on stranded natural gas and turn it into flops with a supply chain that already exists.

But the knock-on effects are why this keeps pulling at people. If you can industrialize power and operations in orbit at meaningful scale, you're not just running GPUs. You're building a new kind of infrastructure that makes it easier for humans to keep spreading out. Compute is just one of the first excuses to pay for the scaffolding. Even if this is a mediocre trade on strict near-term unit economics, the second-order effects could be enormous.

I'll go one step further and say the quiet part out loud: we should be actively goading more billionaires into spending on irrational, high-variance projects that might actually advance civilization. I feel genuine secondhand embarrassment watching people torch their fortunes on yachts and status cosplay. No one cares about your Loro Piana. If you've built an empire, the best possible use of it is to burn its capital like a torch and light up a corner of the future. Fund the ugly middle. Pay for the iteration loops. Build the cathedrals. This is how we advance civilization.

Links to Reports

Everyone is going to copy-paste this into the models, so I've done that part for you. It's a decent way to automate the sanity checks, but it could use more in-depth review.

GitHub: github.com/andrewmccalip/thoughts

Grok: grok.com/share/...
ChatGPT: chatgpt.com/share/...
Gemini: gemini.google.com/share/...
Claude: claude.ai/public/artifacts/...

Overall Conclusion

Even so, irrational ambition doesn't get to ignore physics. The point of this page is to make the constraints explicit, so we can argue about reality instead of vibes. If the numbers close, even barely, then it's worth running hard on the idea. If they don't, the honest move is to say so and move on. Either way, I think some version of this has a feeling of inevitability.

So scroll down, play with the sliders, and try to break it. Change launch cost. Change lifetime. Change specific power. Change hardware cost. The goal here isn't to "win" an argument. It's to drag the conversation back to first principles: assumptions you can point at, and outputs you can sanity-check. Check out the GitHub, run the code, find the errors, and I'll update it live.

After that, we can do the fun part: thermal diagrams, radiator math, orbit beta angles, failure rates, comms geometry, all the shiny engineering details that make this topic so addicting. It's not obviously stupid, and it's not a sure thing. That's why it's worth doing the math.

Technical Engineering Challenges

The governing constraint for orbital compute is thermodynamics. Terrestrial datacenters leverage convective cooling—dumping waste heat into the atmosphere or water sources, effectively using the planet as an infinite cold reservoir. In the vacuum of space, convection is impossible. Heat rejection relies exclusively on radiation.

Every object in space settles to an equilibrium temperature where absorbed power equals radiated power. If heat generation exceeds radiative capacity, the temperature rises until the $T^4$ term in the Stefan-Boltzmann law balances the equation:

$$\dot{Q}_{\text{rad}} = \varepsilon \sigma A T^4$$

The engineering challenge is ensuring this equilibrium temperature remains below the safe operating limits of silicon processors.

Energy Balance and Heat Rejection

To dimension the radiator surface, we must account for the total thermal load managed by the satellite bus. In this model, based on a Starlink-style bifacial architecture (PV on front, radiator on back), the system must reject the aggregate energy of two distinct paths:

1. Incident Solar Flux: The sun delivers $G_{\text{sc}} = 1361\;\text{W/m}^2$ (AM0). With a solar absorptivity $\alpha = 0.92$, the panel absorbs approximately $\sim 1250\;\text{W/m}^2$.
2. Energy Partitioning:
   • Electrical Path ($\sim$22%): High-efficiency cells convert $\sim 275\;\text{W/m}^2$ into electricity. This power drives the compute payload and is converted entirely back into heat by the processors. A liquid cooling loop collects this heat and returns it to the panel structure for rejection.
   • Thermal Absorption ($\sim$78%): The remaining $\sim 975\;\text{W/m}^2$ is not converted to electricity but is absorbed immediately as lattice heat (phonon generation) within the panel structure.
3. Total Heat Load: The radiator must reject the sum of both the immediate thermal absorption and the returned electrical waste heat—effectively 100% of the absorbed solar flux.

Operating Temperature Limits

Modern AI accelerators (H100/B200 class) typically throttle at junction temperatures $T_j > 85\text{–}100\degree\text{C}$. To maintain a junction at 85°C, and accounting for the thermal gradient across cold plates and interface materials ($\Delta T \approx 10\degree\text{C}$), the radiator surface temperature $T_{\text{rad}}$ is constrained to approximately 75°C.

The model below calculates the equilibrium temperature for a bifacial array in a terminator orbit ($\beta = 90^\circ$). It accounts for solar flux, Earth IR ($\sim 237\;\text{W/m}^2$), and albedo. If the calculated equilibrium temperature $T_{\text{eq}}$ exceeds the target radiator temperature, the design fails.