GPUs in Space are dumb, right?

In February 2026 Elon Musk told Dwarkesh Patel that within 30 to 36 months “the most economically compelling place to put AI will be space.” The whole question reduces to one number. Here is the math.

Solar

In a dawn-dusk sun-synchronous orbit a panel sits in near-continuous sunlight at 1361 W/m²: no atmosphere, no night, no clouds, no seasons. A ground panel, even in a good desert, averages a fraction of that after night, weather, atmospheric loss, and the sun’s angle. Average power per deployed cell is about 5x higher in orbit, ~8x against a mid-latitude site.

Storage

Running 24/7 off ground solar means storing energy for the night. A desert site delivers its day’s energy in a few peak-equivalent hours, so you oversize the panels and store ~12 to 18 hours of load in batteries, which dominates the bill. A terminator orbit has no night: even a normal low orbit spends only ~35 of each ~90 minutes in shadow, and the dawn-dusk orbit nearly eliminates that. Per kilowatt of continuous load:

	Ground (desert)	Terminator orbit
Solar oversizing	~4x	~1x
Storage	~15 kWh	~0.5 kWh
Battery cost (~$300/kWh)	~$4,500	~$150

A ~$4,000/kW swing toward orbit. This is the real structural advantage.

Launch cost

Orbit has one cost the ground does not: lifting the hardware. Per kilowatt, that cost is the system’s specific mass (kg/kW) times the launch price ($/kg), and it dominates everything else. Every orbital LCOE figure in circulation, from $2/MWh to ~$900/MWh, is just that product with a different launch price plugged in. None is an independent result. The only meaningful question is how cheap launch must get to beat the ground.

Ground 24/7 solar-plus-storage is ~$40 to $50/MWh, about $5,000/kW of lifetime cost over 15 years. Subtract ~$1,500/kW for orbital hardware, leaving ~$3,500/kW for launch. Divide by system mass:

System mass	Break-even launch price
7 kg/kW (cells only)	~$500/kg
15 kg/kW (optimistic full system)	~$235/kg
30 kg/kW (realistic integrated)	~$115/kg
60 kg/kW (conservative)	~$60/kg

Current Starship estimates: $500 to $1,500/kg. Musk’s 2028 target: $100/kg. So the outcome hinges on two numbers, system mass and launch price, and a world near 15 kg/kW and a few hundred dollars per kilo makes orbit competitive.

Cooling

System mass is set mostly by cooling. The 7 kg/kW “cells only” figure counts no way to shed heat, and in space heat leaves only by radiation. A radiator near room temperature rejects ~245 W/m² ($\varepsilon\sigma T^4$ at 20°C), and every watt of compute must exit through it:

Radiator	Mass per kW of heat
Thin-film (aspirational)	~1 kg/kW
Aluminium (conventional)	~8 kg/kW

That 7 kg/kW swing is the difference between roughly the 15 and 30 kg/kW rows above, i.e. between a ~$235/kg and a ~$115/kg break-even. Cooling mass is the binding uncertainty.

Two things are not blockers. Shipping the GPUs is cheap: an H100 is ~3 kg/kW of silicon, a few percent of its purchase price even at $1,000/kg. And radiation, long assumed fatal, is handled: Google’s Trillium TPU survived 15 krad(Si) with no hard failures, Starcloud’s orbital H100 ran inference and training without crashes, and ECC overhead is ~5 to 15%.

Throughput

Suppose the economics close. The schedule does not. Musk’s 100 GW/yr at 30 kg/kW is 3 billion kg to orbit per year:

$$\frac{100\,\text{GW} \times 30\,\text{kg/kW}}{100{,}000\,\text{kg/launch}} = 30{,}000\ \text{launches/yr} \approx 80\,\text{/day}$$

SpaceX flew 165 orbital launches in all of 2025, a record. The 100 GW vision needs ~180x the entire global cadence, sustained daily. The cells-only mass figure only divides that by a few. No cost reduction touches a propellant-and-pad limit on a 30-month horizon.

Conclusion

The physics works. Solar and storage favour orbit by a real margin, radiation is solved, and break-even launch prices near $100 to $235/kg are not absurd. The workload is no obstacle: training and batch inference are latency-tolerant, and LEO round-trip latency (~20 to 40 ms) is fine even for interactive serving.

Whether orbit wins reduces to one race: launch price falling faster than ground storage does. Orbit’s main advantage is the battery it doesn’t need, and cheap ground batteries erode that from the other side, so the question is which cost curve drops faster. Ground has the head start and the larger industrial base.

If launch wins that race, the cadence ceiling (80 launches a day for 100 GW/yr) caps how fast the installed base grows, not what can run there, so the orbital share climbs over decades rather than switching on. What the math rules out is the timeline, not the scale. A large fraction of new AI compute built in orbit in the 2040s is plausible. “Ten times cheaper in thirty months,” and “more compute in orbit than on Earth in five years,” are not.