In Part I, I wrote about one of the counterintuitive problems with putting data centers in space: cooling them.
Space may be cold, but a vacuum doesn’t give you air or water to carry heat away. Ultimately, all those megawatts of waste heat have to leave through thermal radiation.
But space data centers have another radiation problem.
This time, we’re talking about the kind of radiation most people think of when they hear the word.
And instead of trying to get it out, we’re desperately trying to keep it from getting in.
Earth Is a Wonderful Place for a Computer
We tend to forget just how protected we are down here.
Earth’s atmosphere and magnetic field shield the surface—and the computers sitting on it—from much of the high-energy particle radiation found in space. Move those computers into orbit, and the environment changes dramatically.
Space contains energetic protons and heavy ions originating from the Sun, galactic cosmic rays, and particles trapped by Earth’s magnetic field.
Those particles don’t care that the object in their path is a $30,000 GPU.
When an energetic particle passes through a semiconductor, it can deposit electrical charge inside the device. Sometimes that changes a bit from a 0 to a 1, or vice versa.
That’s called a single-event upset.
A bit flip might sound trivial. Modern computers deal with errors all the time. But radiation-induced events can do considerably more than corrupt a single piece of data. They can disrupt processor operation, cause system crashes, trigger latch-ups, and in some circumstances permanently damage a device.
NASA has been dealing with this problem for decades. The difference is that NASA generally isn’t trying to put a hyperscale AI data center in orbit.
A GPU Is Not a Spacecraft Computer
Spacecraft electronics are designed around the reality of radiation.
Engineers can use radiation-hardened components, error-correcting memory, redundant processors, watchdog systems, shielding, and software capable of detecting and recovering from faults.
There’s a tradeoff, however.
The processors at the bleeding edge of computing performance are commercial chips designed for terrestrial data centers. Radiation-hardened processors have traditionally lagged well behind state-of-the-art commercial hardware in raw computational performance. For a spacecraft guidance computer, that’s often acceptable. For an AI data center, computational performance is the entire reason the machine exists.
The economics of AI depend on packing astonishing numbers of transistors into cutting-edge GPUs and running enormous numbers of those processors simultaneously. Now imagine taking tens of thousands of those extraordinarily complex devices out of their comfortable terrestrial environment and placing them where energetic particles can pass through their circuitry.
You’ve created a reliability problem on a scale spacecraft designers normally don’t have to contemplate.
Radiation Doesn’t Just Flip Bits
There’s also a slower problem: radiation damage accumulates.
Engineers call one important measure total ionizing dose. Over time, repeated exposure can alter the electrical properties of semiconductor materials, increase leakage currents, change transistor behavior, and eventually cause components to fail.
Other radiation effects can physically displace atoms within materials, gradually degrading electronic devices.
So there are really two different clocks running. One counts sudden events: a high-energy particle hits exactly the wrong place and something immediately goes wrong. The other counts accumulated damage as years of radiation exposure slowly degrade the hardware.
That’s an uncomfortable combination for something we expect to operate continuously.
Just Add Shielding?
The obvious solution is shielding. Put enough material around the computers and you can reduce their radiation exposure.
Unfortunately, in space, mass is expensive. Every kilogram of shielding has to be launched. More shielding means more mass, which means greater launch requirements and higher costs. And high-energy cosmic rays are particularly difficult to stop completely.
There are engineering solutions. Shielding can reduce exposure. Redundancy can tolerate failures. Error-correcting systems can catch corrupted data. Workloads can be distributed so individual failures don’t bring down the system.
None of this makes space computing impossible. It does make the phrase “let’s put the data centers in space” conceal an extraordinary amount of engineering.
Space Keeps Sending Us the Bill
That’s what fascinates me about the current enthusiasm for orbital data centers.
At first glance, the idea sounds almost obvious. AI needs enormous amounts of energy. Space offers abundant sunlight. Put the computers next to the energy source and eliminate the terrestrial constraints.
Then physics starts sending invoices.
In Part I, the invoice was thermodynamics: every watt you put into the computers eventually becomes heat, and getting rid of that heat in a vacuum is surprisingly difficult. In Part II, the invoice is radiation: the very environment where you want to put those cutting-edge computers is actively hostile to the microscopic electronics that make them valuable. And we haven’t even gotten to servicing failed machines, replacing obsolete hardware, launch economics, communications, or orbital debris.
Maybe engineers will solve all of these problems. Human beings have an impressive record of solving difficult engineering challenges. But putting data centers in space isn’t simply a matter of moving the cloud a little closer to the actual clouds.
It means taking some of the most sophisticated—and delicate—machines humanity has ever built and placing them in an environment where every error is harder to detect, diagnose, and fix.
Jayson L. Adams is a technology entrepreneur, artist, and the award-winning author of The Quantum Mirror, Ares, and Infernum.
His novels blend high-stakes science fiction, psychological tension, and character-driven suspense. Explore the books at www.jaysonadams.com.