Placing data centers in space has gone from science fiction to buy-in. SpaceX, Meta, and Google are all pursuing this concept, and for good reasons: Orbit provides near-continuous solar power and natural space cooling, bypassing the power, permitting, and cooling constraints that currently hamper the construction of data centers on Earth. In April, Meta announced that she had already done so Reserved gigawatts Future orbital solar capacity. And since AI’s appetite for power exceeds what terrestrial grids can comfortably provide, the business logic is hard to ignore.
But the debate is about the wrong question. We’re still debating whether orbital data centers are technically feasible and how expensive it is to launch them — a debate that still rages on even among leaders in the field. In February, OpenAI’s Sam Altman approached orbital data centers “Ridiculous right now.” Citing high failure rates and cost. The most important question is what happens when something fails, and where the real vulnerability lies when that happens. It’s easy to assume that risks live in orbit. Mostly, it doesn’t happen. It lives on Earth, in a few stations that connect satellites to Earth. Today, these facilities support every orbital computing system, yet they receive only a small portion of the protection afforded by their strategic importance.
On Earth, a hardware failure is a technician entering the server room. In orbit, a routine glitch or small meteorite strike can sideline an asset until the next launch window, turning what could be routine maintenance into months of loss of capability. This is not hypothetical. In March, the SpaceX Starlink satellite was launched He suffered from a mysterious orbital anomalyThis is the kind of error that on Earth is a maintenance ticket and in orbit it can take an asset down forever. The technology for deploying servers in space is advancing faster than the technology for serving them once they arrive. Until in-orbit maintenance becomes practical, every failure represents a loss, and this reframes the entire business case. Average time to repair is no longer measured in hours. It is measured in launch windows.
This reality should rearrange the design of the system. If failures cannot be prevented the way they occur in reality, they should be expected instead. Redundancy and gentle degradation stop being nice-to-haves and become the whole architecture, where workloads are distributed across multiple satellites and orbital planes so that the loss of a single platform reduces capacity rather than disrupts operations.
The trap is co-location: when multiple tenants share a single orbital platform, a single physical impact becomes a shared outage that cannot be mitigated. The economy is merciless. Radiation, thermal cycling, and collision exposure turn normal failures into total losses, resulting in insurance and redundancy costs much higher than most energy saving projections account for. A company running critical workloads on one or two satellites is left with a single, catastrophic point of failure that no launch cost spreadsheet can adequately capture.
But even the in-orbit situation is only half the problem. Each orbital data center is still grounded, in ground stations, and downlink facilities Fiber networks that carry data to its last mile for users. A satellite is as flexible as the terrestrial infrastructure that connects it to the network, etc amenities Disabling them is much easier than disabling hardened ground data centers. Concentrate global computing behind a small number of ground stations, and you have created physical choke points that are attractive to anyone seeking to cause disruption, whether through jamming, signal interference, or physical intrusion. Computing power may be located in orbit. Easily accessible weak points remain firmly on the ground.
This is the part of the conversation that gets the least attention when it deserves the most attention. As orbital computing standards, this Earth amenities They become the new crown jewels of the digital economy. The uncomfortable truth is that the resilience of a space data center will depend not so much on the sophistication of its satellites as on the security of an ordinary building on Earth.
Securing this infrastructure is not primarily a cybersecurity issue, and it is a mistake to treat it as one. Satellites, ground stations and terrestrial networks operate as a single interconnected system, meaning that settlement at any location can cascade across an entire architecture. A compromised ground station is both a physical security incident and a cyber event.
Protecting these sites therefore means abandoning the habit of discrete security measures – a camera here, an access reader there, a visitor log in the clipboard – and moving towards integrated security architectures in which physical access, video, alarms and identity systems work together. Organizations will need analytics that can identify anomalies before they become incidents, rather than simply documenting what has already happened. Like these amenities They become single points of failure for orbital computing, and this coherence becomes the difference between a contained incident and a nationwide outage.
None of this is an argument against putting computing in space. The energy issue is real, and the companies pursuing it are serious. It’s an argument for pricing risk honestly before we get there, and for investing in service, redundancy and ground segment security with the same ambition as we do at launch. Industry has an answer to the AI energy problem. It has not yet been able to develop an equally convincing answer to the resilience problem it faces. The organizations that succeed in orbit are the ones that will solve both.
