By Jerome Paye, CEO TAU Systems
Orbital computing is no longer a thought experiment. But the hardware behind it has two problems that nobody has solved… yet.
Elon Musk thinks orbital data centers will eventually handle more AI computing than every facility on Earth combined (SpaceNews). Jeff Bezos is less bullish on the timeline (CNBC) but agrees the destination is real. Blue Origin has applied to launch its TeraWave satellite constellation. SpaceX is reportedly planning to use Starship to get hardware to orbit at a scale beyond anything ever attempted.
The appeal is obvious. Abundant solar power without clouds or nightfall and none of the land disputes and grid bottlenecks that are slowing data center build-out here on Earth. AWS chief Matt Garman has publicly dismissed the idea (Fortune), and he has a point worth taking seriously: you cannot just put a server rack in a rocket and call it done.
Two problems sit beneath all of this. First, today’s AI chips are too heavy and power-hungry to launch at any meaningful scale. Second, even when you get them to orbit, space radiation would degrade or destroy them in a fraction of the time they would last on the ground. TAU Systems works on both of these challenges, using the same underlying technology.
Why the chips themselves have to change
Launch cost is measured in thousands of dollars per kilogram. A single AI server rack can weigh more than a metric ton and draw over a hundred kilowatts of power. Multiply that by the rack count needed for a data center with any real capacity, and the launch bill alone rules out the project before it starts. Add to that the infrastructure neededto cool these power-hungry chips in the vacuum of space, and the problem becomes clear. Smaller, lighter, more power-efficient chips are not a nice-to-have. They are a precondition for achieving on-orbit compute at a scale meaningful to the AI race.
Today’s most advanced chips are printed using EUV lithography, which fires high-power lasers at droplets of molten tin to generate 13.5 nm light. It works, but it is reaching its limits. Each generation beyond this point needs a fundamentally different light source, and the industry knows it. That is why labs and start-ups are now looking seriously at laser-plasma acceleration: firing intense laser pulses through plasma to accelerate electrons across a few millimetres instead of the hundreds of metres a conventional accelerator needs. Done right, this could drive free-electron lasers at wavelengths well below 13.5 nm, opening the door to chips smaller and more efficient than anything on the market today.
This is the technology TAU Systems is commercializing. Our compact laser-powered accelerators and LPA-driven free-electron lasers fit inside a standard commercial facility rather than a kilometre-long tunnel. As the semiconductor industry looks past High-NA EUV for its next roadmap step, compact laser-powered sources are one of the few credible candidates. The chips that eventually come out of this approach are the chips orbital data centers will actually be able to afford to launch.
Why radiation is the other half of the problem
Smaller chips solve the weight problem. They do nothing for the radiation problem, and in some ways make it worse: as transistor nodes shrink, they become more susceptible to upsets.
Cosmic rays and trapped radiation belt particles cause two kinds of damage in orbit. A single high-energy particle can flip a bit, trigger a latch-up, or in the worst case burn out a component outright; this is a Single Event Effect, and it happens with no warning. Slower, cumulative exposure degrades the semiconductor material over time until performance drops off and the part eventually fails. AI training workloads are particularly exposed to the first kind, because thousands of chips run in tight lockstep and one undetected flip can cascade through the whole system.
Radiation-hardened chips exist, but they typically lag commercial silicon by a generation or more, cost far more, and are made in small batches. None of that is compatible with running AI workloads that need cutting-edge performance. What space computing needs is a way to test the latest commercial chips against real radiation conditions, fast and repeatedly, so the ones that hold up can be qualified and used.
That testing is currently a bottleneck in its own right. Heavy-ion beam time at the world’s conventional accelerator facilities is in short supply, and wait times of several months are normal. For a development program trying to move fast, that delay is its own kind of failure.
TAU Labs and the testing bottleneck
TAU Labs in Carlsbad, California is the world’s first commercial laser-powered accelerator center. It generates high-energy particle beams that replicate real-world radiation environments, and offers Single Event Effects testing to space, defence, semiconductor, and industrial customers. Because the platform is compact and laser-driven rather than a large conventional accelerator, it can offer the kind of flexibility and turnaround that test programs need but rarely get.
In April 2026, TAU Systems joined RADNEXT 2030, the CERN-coordinated European network giving researchers and industry access to radiation testing facilities. TAU Labs is the only US-based facility in the network, and European researchers and companies can now book time with us, supported by Horizon Europe funding. A compact laser-powered accelerator sitting in a network built around large conventional facilities says something about where this field is heading.
For anyone building hardware for orbital computing, that access will matter a great deal. Radiation qualification sits squarely on the critical path between a working chip and a chip that can actually be launched at scale. To meet the need, the industry needs to vastly expand test capacity to eliminate the bottleneck that results in the months-long waits customers face today.
One technology, two roles
TAU Systems sits at an unusual point in this picture. The same laser-powered acceleration physics underpins both halves of the problem: it is the basis for the lithography that could eventually produce smaller, lighter chips, and it is the technology running TAU Labs today, where those chips and their predecessors get qualified for the radiation environment they will face in orbit.
Whether the orbital data center arrives in three years, as Musk suggests, or considerably later, as Bezos and Garman would have it, the hardware problems underneath it have to be solved either way. That work is already happening at TAU Labs.