Power and Heat Are the Hard Limits for Orbital AI Data Centers
Makenzie Lystrup, a principal consultant at Peridot Services and former director of NASA’s Goddard Space Flight Center, argues that orbital data centers should not be treated as one idea. In a Bloomberg Technology interview, she says near-term edge computing in orbit is plausible, while hyperscale AI infrastructure of the kind associated with SpaceX faces much harder constraints: power systems, heat rejection, radiation-tolerant hardware, networking, reliability and maintenance. Her central point is that the challenge is not merely launching servers into space, but operating them as space-qualified infrastructure.
Orbital compute is three problems with very different burdens
Makenzie Lystrup separates “space-based data centers” into three grades of ambition. The distinction matters because only the most extreme version approaches the hyperscale AI vision associated with SpaceX.
The near-term category is on-orbit edge computing. Satellites collect imagery, radar, weather, maritime, climate, or defense data, then send raw data to an orbital compute node. That node filters the data, runs AI or machine-learning inference, or performs other processing before transmitting a smaller or higher-value product back to Earth. Lystrup calls this “real” and “near term.” She points to Axiom as saying it deployed a prototype unit on the International Space Station and has an orbital data center node.
The second category is resilient or sovereign storage and computing: infrastructure physically separated from terrestrial threats such as disasters, political borders, or ground-based attacks. Lystrup says this has plausible use cases in defense, continuity of government, and critical infrastructure, though it remains “not entirely explored.”
The third category is hyperscale AI work in orbit. That is the “moonshot.” It may be enormous, but Lystrup says the problem is not reducible to launch cost or orbital downlink nodes. The system would have to solve power, heat rejection, radiation-tolerant computing, high-bandwidth optical networking, orbital reliability, and other operating demands at the same time.
It is not merely about the launch cost or even the orbital downlink nodes.
Ed Ludlow frames SpaceX’s extreme version as hyperscale orbital compute running massive AI workloads and inference. Bloomberg showed SpaceX-attributed animation of a satellite deploying large solar panels, followed by close-up animation of server racks and computer hardware assembling in space. Lystrup’s point is that the harder unresolved question is whether that hardware can operate as space-qualified infrastructure, not merely whether it can be launched or assembled.
In orbit, the data center is a heat engine before it is a compute platform
Lystrup’s sharpest technical objection concerns heat. Ludlow notes the common science-fiction intuition of space as a vast cold environment. Lystrup says that intuition is misleading for orbital data centers.
A terrestrial data center can reject heat through chillers, air, water or liquid cooling, and evaporative systems. In orbit, there is no convective air heat sink. Waste heat has to move through the spacecraft and then be radiated away.
In orbit, a data center is a heat engine first and foremost and a compute platform second.
That formulation turns thermal design into a core economic constraint. Radiator area, radiator emissivity, spacecraft attitude control, contamination, and degradation all become “first-order economic variables.” The costs and limits are not only in processors, launch, or bandwidth. They are also in the surface area and durability required to radiate heat, in the orientation needed to make that radiation effective, and in the way orbital conditions degrade the system over time.
For hyperscale AI workloads and inference, the spacecraft architecture would have to reject the resulting heat without the mechanisms terrestrial operators rely on. In Lystrup’s account, the absence of convective cooling makes orbital compute a fundamentally different infrastructure problem even before networking, radiation tolerance, or maintenance enter the discussion.
More sunlight does not automatically mean usable power infrastructure
The power argument for orbital data centers often starts with a simple premise: in the right orbit, sunlight is stronger and more continuous than it is on Earth. Lystrup acknowledges that premise. She says Google’s Project Sun Catcher argues that solar panels in the right orbit can be up to eight times more productive than on Earth.
Her answer turns on the difference between access to energy and a working energy system. “Just because you have space energy access doesn’t mean that you have space energy infrastructure,” she says.
Reliable orbital power requires generation, storage, distribution, thermal rejection, and redundancy. Those systems have to function under close-formation orbital dynamics, radiation effects, thermal-management constraints, and ground-communications requirements. They also have to be flight qualified, reliably operated in space, and maintained.
Maintenance is the cost Lystrup says is currently under-discussed. The issue is not just launching power hardware; it is keeping the power system working in orbit, with maintenance treated as part of the operating model rather than an afterthought.
The solar story is therefore incomplete rather than false. Stronger and more continuous sunlight may improve the energy input side of the equation, but the orbital system still has to store, route, cool, protect, and repair that power infrastructure.
SpaceX would have to prove infrastructure, not just ambition
Asked whether SpaceX can do it, Lystrup does not dismiss the possibility. SpaceX is vertically integrated and has demonstrated many technical challenges before, she says, so she “wouldn’t count them out.”
The burden she describes is broader than launch. Hyperscale orbital AI would require coordinated performance across spacecraft design, power generation and storage, heat rejection, radiation-tolerant computing, high-bandwidth optical networking, redundancy, maintenance, and reliable operations. Success would mean making those systems work together in orbit as infrastructure.
That is a higher standard than an orbital compute prototype in the lower-end categories Lystrup describes. Edge computing can be useful when it filters data close to where that data is collected. Sovereign orbital storage and compute can have value because they are physically separated from terrestrial risks. Hyperscale AI work in orbit is different: it concentrates the hardest energy, thermal, networking, and reliability demands into one operating system.
Lystrup’s position is not that orbital data centers are fantasy. It is that “orbital data center” names several levels of ambition, and the hyperscale AI version is where energy and heat become decisive constraints.