Orbital data centers have moved from science fiction to real hardware in orbit. SpaceX has filed to launch up to a million data center satellites, Starcloud has already put an NVIDIA H100 in space, and Google is developing its own orbital compute project. The debate around all of this has focused, understandably, on whether it can work: launch costs, solar power, cooling, latency. But there is one question almost no one is asking, and it is the question an infrastructure disposition specialist notices first. What happens to the hardware when it dies? On Earth, a retired GPU cluster is an asset to be recovered. In orbit, it is something else entirely. This is a look at the orbital compute boom of 2026, and the end-of-life problem it quietly creates.
TL;DR
Orbital data centers are a real and fast-developing area, with major players and real hardware already in orbit. The essentials:
- This is no longer hypothetical. Starcloud placed the first NVIDIA H100 GPU in orbit in November 2025 and trained a small language model in space. SpaceX has filed with the FCC for up to one million orbital data center satellites, and Google is pursuing an orbital compute project of its own.
- The appeal is real. Orbit offers abundant, uninterrupted solar power (some estimates as low as fractions of a cent per kWh), no cooling water, and an escape from the land, power, and permitting constraints increasingly limiting terrestrial data centers.
- The economics are marginal today and depend on launch cost. Independent analyses put orbital compute at roughly three times the cost per watt of terrestrial equivalents today, with cost parity dependent on launch prices falling toward $200 per kilogram, likely in the 2030s.
- The hardware has a short life in a harsh environment. In low Earth orbit, satellites are designed to operate for around five years, then de-orbit, and the AI hardware inside refreshes even faster on the terrestrial three-to-four-year cycle.
- The end-of-life path is demise, not recovery. Retired orbital hardware is not brought back; it de-orbits and burns up in the atmosphere. Recovering it for reuse or materials is, in the words of industry reporting, technically impossible and economically unfeasible.
The overlooked implication: every terrestrial data center recovers a large share of its hardware value at end of life through asset recovery and materials reclamation. An orbital data center recovers none of it. The residual value that terrestrial operators capture at every refresh is, in orbit, written to zero and burned up on re-entry. That is not an argument against orbital compute, but it is a real cost that belongs in the conversation, and almost no one is putting it there.
The Orbital Data Center Race Is Real
For years, data centers in space were a thought experiment. In 2026 they are a genuine, well-funded race with hardware already flying.
The Players
The field has moved quickly and now includes some of the most serious names in technology. Starcloud, a Y Combinator-backed startup, reached a milestone in November 2025 by placing the first NVIDIA H100 GPU in orbit aboard its Starcloud-1 satellite, and went on to train a small language model in space and run inference on other models in orbit. The company has raised significant funding and is developing larger, more powerful orbital compute spacecraft designed to launch aboard SpaceX’s Starship.
SpaceX itself is the highest-profile entrant. The company filed an application with the FCC in early 2026 for up to one million orbital data center satellites, with the vision of building enormous solar-powered compute capacity in orbit, and it has publicly framed orbital data centers as a way to bypass the strained terrestrial power grid. SpaceX brings a decisive structural advantage to the idea: it operates the launch capability the entire concept depends on, and its Starlink program has already demonstrated the ability to build, launch, and operate satellites at a scale no other operator approaches.
Google has entered as well, through Project Suncatcher, an orbital compute initiative exploring networks of solar-powered satellites carrying its Tensor Processing Unit AI chips. And several other companies are pursuing orbital infrastructure, making this a genuine field rather than a single company’s ambition.
The Milestone That Changed the Conversation
The reason the debate shifted in 2026 is that the concept stopped being theoretical. Starcloud’s placement of a working H100 in orbit, and the training of an actual model in space, moved the question, in the words of one industry analysis, from “will orbital data centers happen?” to “which wave of deployment will reach cost parity with the ground?” That is a meaningful shift. The engineering is being demonstrated, incrementally, in real orbit. What remains uncertain is the economics and the timeline, not the basic feasibility.
Why Put a Data Center in Space at All
The case for orbital compute is stronger than it first sounds, and it is worth stating fairly, because the advantages are what make the whole idea worth taking seriously.
Abundant, Uninterrupted Solar Power
The single biggest driver is energy. In the right orbit, a satellite is bathed in near-constant sunlight, unfiltered by atmosphere and uninterrupted by night. Solar power in orbit can be extraordinarily cheap and effectively continuous, with some proponents citing energy costs as low as fractions of a cent per kWh. For an industry whose defining constraint has become power (as terrestrial grids strain under AI demand and electricity costs rise), the appeal of limitless orbital sunshine is obvious.
No Cooling Water and No Land
Orbital data centers sidestep two of the terrestrial industry’s most contested resource problems at once. There is no cooling water requirement of the kind straining water-stressed regions on Earth, and there is no land to acquire, no local community to negotiate with, and no zoning or permitting fight. Heat is radiated away rather than carried off by water or air. In an era when terrestrial data centers face mounting practical, political, and public opposition, the idea of building where none of those constraints apply is genuinely attractive.
Escaping the Terrestrial Bottleneck
The deeper appeal, and the one industry observers keep returning to, is that orbit offers an escape from the constraints that increasingly limit terrestrial growth. Grid interconnection queues stretch years. Communities resist new campuses. Power and water are contested. As one commentator put it, for the companies pursuing this, the engineering challenge of space may eventually look smaller than the social and logistical challenge of building on the ground. That is a striking reframing, and it captures why serious companies are investing real money in the idea.
Latency for Specific Uses
For certain workloads, especially processing data generated in space (satellite imagery, Earth observation, and space-based sensing), doing the compute in orbit avoids the cost and delay of sending raw data down to the ground and results back up. Starcloud-1 has already demonstrated this by processing satellite imagery in orbit. This is a narrower advantage than the energy case, but it is a real one for the right applications.
The Honest Economics
For all the genuine appeal, an honest look at the numbers shows orbital compute is not yet cost-competitive, and its future depends heavily on one variable.
Independent analyses converge on a similar picture. Today, terrestrial data centers remain clearly cheaper than orbital ones. One widely cited engineering analysis put a one-gigawatt orbital data center at roughly $42 billion, almost three times the cost of its ground-based equivalent, driven by the up-front cost of building and launching the satellites. Skeptics at established space firms have calculated orbital compute costing roughly three times more per watt than terrestrial equivalents. These are not dismissals of the idea; they are estimates of how far it currently is from parity.
The single variable that decides the future is launch cost. Getting mass to orbit is the dominant cost in any space business, and orbital data centers need it to fall dramatically. Falcon 9 today delivers payload to orbit at a cost measured in thousands of dollars per kilogram. The business case for orbital compute generally requires that to fall toward roughly $200 per kilogram, an improvement that proponents, including Google’s Project Suncatcher analysis, associate with next-generation launch capability and expect in the 2030s. This is precisely where SpaceX’s structural role matters: the entire economic case for orbital data centers rests on the continued reduction in launch cost that SpaceX has spent two decades driving, and on Starship reaching high-cadence operation. If launch costs fall as SpaceX intends, the orbital case improves markedly; the timeline for the whole field is effectively tied to that trajectory.
The reasonable read, shared across proponents and skeptics alike, is that orbital data centers are marginal today and plausibly viable in the 2030s if launch costs fall as hoped. This is a long-horizon infrastructure bet, not a 2026 deployment. That is not a knock on the idea; foundational infrastructure often looks marginal right up until an enabling cost curve crosses a threshold.
The Question Nobody Is Asking: What Happens When It Dies?
Here is where the conversation has a genuine blind spot, and it is the one an infrastructure disposition specialist sees immediately. Every analysis of orbital data centers focuses on getting hardware up and running it. Almost none addresses what happens when that hardware reaches the end of its life. And in orbit, the end of life is fundamentally different from the end of life on the ground.
Orbital Hardware Has a Short Life
Nothing about being in space extends hardware life; if anything, the environment is harsher. Satellites in low Earth orbit are typically designed to operate for around five years before de-orbiting, and the AI hardware inside an orbital data center faces the same compressed refresh pressure as its terrestrial counterpart, arguably more. GPUs and AI accelerators refresh on a three-to-four-year cycle as newer, far more efficient generations arrive, and the economics of a launched asset only work if it is running the most capable hardware. The radiation environment of space also degrades electronics over time. So an orbital data center is not a launch-once-and-forget proposition; it is a continuously refreshing fleet, with each generation of hardware reaching obsolescence on a timeline measured in a handful of years.
The End-of-Life Path Is Demise, Not Recovery
This is the crux. When a terrestrial data center retires a GPU fleet, that hardware comes out of the rack and enters a recovery process: it is remarketed, reused, or dismantled for materials, and a large share of its value is captured. When an orbital data center retires its hardware, none of that happens. The satellites de-orbit and burn up in the atmosphere.
To be clear, this de-orbit process is, in the leading operators’ hands, handled responsibly. SpaceX in particular has made satellite disposal a discipline: it maintains a post-mission disposal reliability rate exceeding 99%, well above the FCC’s requirement, designs its satellites to fully burn up on re-entry to minimize debris, and proactively de-orbits aging or degraded satellites before they can become hazards. The debris and safety dimension of orbital hardware disposal is a solved problem in the hands of a responsible operator, and SpaceX has set the industry benchmark for doing it well.
But solving the debris problem is not the same as recovering the hardware. Retrieving retired satellites from orbit for recycling or reuse is, as industry reporting bluntly puts it, technically impossible and economically unfeasible. There is no truck to send to low Earth orbit. There is no dismantling line, no refiner, no secondary market. The hardware’s end-of-life path is atmospheric demise: it is deliberately destroyed. Whatever residual value the GPUs, accelerators, and structured hardware still held at retirement is not recovered. It is burned up on re-entry.
The Recovery Value That Is Written to Zero
This is where the two worlds diverge most sharply, and where the terrestrial disposition perspective adds something the orbital debate is missing. On Earth, retired data center hardware retains substantial residual value. Recent-generation servers recover a meaningful share of their original value on the secondary market, and specialist resale of retired AI hardware like H100 and A100 fleets can capture a significant portion of new-price value when handled quickly and knowledgeably. Even hardware past its resale life carries a materials-recovery floor in its metals. Terrestrial operators recover a real fraction of their hardware investment at every refresh, and doing so is a standard, expected part of the infrastructure lifecycle.
An orbital data center captures none of that. Every refresh cycle, the retiring hardware is de-orbited and destroyed rather than recovered. The residual value that a terrestrial operator would harvest at end of life is, in orbit, structurally unrecoverable. Over the life of a large orbital fleet refreshing every few years, that is a substantial and recurring cost, the value of all the hardware that, on the ground, would have been recovered and, in orbit, is instead burned up. It rarely appears in the orbital business case, but it is real, and it works entirely against the orbital model relative to the terrestrial one.
None of this means orbital data centers will not happen or should not be pursued. It means the full lifecycle cost of orbital compute includes a recovery-value write-off that terrestrial compute does not, and an honest comparison of the two should account for it. The end-of-life question is not a footnote; it is a genuine line item that the current debate has almost entirely omitted.
What This Means for Terrestrial Infrastructure
The orbital end-of-life problem is, in a sense, a mirror that reflects something back about terrestrial infrastructure: the recoverability that orbit lacks is a real and underappreciated feature of building on the ground.
For terrestrial operators, the contrast is a reminder that hardware recovery is a genuine economic asset, not an afterthought. Every retired GPU cluster, every decommissioned rack of networking and compute, carries residual value that can be recovered through expert remarketing and materials reclamation, value that an orbital operator would simply lose. In a period when the whole industry is focused on the cost of building and powering AI infrastructure, the recoverable value at the other end of the lifecycle is easy to overlook. The orbital case, by removing recoverability entirely, throws its terrestrial value into relief.
The practical implication for operators building on the ground is straightforward: capture the recovery value your infrastructure generates, because it is one of the real advantages of terrestrial compute. That means treating end-of-life hardware as an asset to be recovered rather than a cost to be disposed of, working with disposition partners who can maximize remarketing and reuse before materials recovery, and building refresh planning that accounts for the value returned at each cycle. Orbital data centers, whatever their eventual future, cannot do any of this. Terrestrial operators can, and in an era of compressed refresh cycles and expensive hardware, that recoverable value is a meaningful part of the total economics.
Frequently Asked Questions
Are there really data centers in space?
Yes, in early form. Starcloud placed the first NVIDIA H100 GPU in orbit in November 2025 aboard its Starcloud-1 satellite, trained a small language model in space, and has run inference on other models in orbit. It is not yet a full-scale data center, but working AI hardware is genuinely operating in space. SpaceX has filed with the FCC for up to one million orbital data center satellites, and Google is pursuing an orbital compute project called Project Suncatcher. So orbital data centers have moved from concept to real, if early-stage, hardware, with several serious companies now in the race, though large-scale deployment remains years away.
Why would anyone build a data center in space?
The main draw is energy. In orbit, solar power is abundant, continuous, and unfiltered by atmosphere, with some estimates putting energy costs at fractions of a cent per kWh. Orbital data centers also avoid two contested terrestrial resource problems: they need no cooling water and no land, sidestepping the water-stress and community-opposition issues facing ground-based facilities. And they escape the grid interconnection queues, permitting fights, and local resistance increasingly constraining terrestrial growth. For some workloads, like processing satellite imagery, doing the compute in orbit also avoids sending data up and down. As terrestrial constraints tighten, the appeal of building where they do not apply grows.
How much does an orbital data center cost?
Today, considerably more than a terrestrial equivalent. One widely cited engineering analysis put a one-gigawatt orbital data center at roughly $42 billion, almost three times its ground-based equivalent, and other analyses estimate orbital compute costing about three times more per watt than terrestrial. The dominant cost is launching mass to orbit. The business case generally requires launch costs to fall toward roughly $200 per kilogram, down from thousands today, an improvement associated with next-generation launch capability and expected in the 2030s. So orbital data centers are marginal today and plausibly viable in the 2030s if launch costs fall as proponents like SpaceX intend.
What happens to a data center satellite at the end of its life?
It de-orbits and burns up in the atmosphere. Satellites in low Earth orbit are designed to operate for about five years, then lower their orbit and re-enter, disintegrating on the way down. Responsible operators handle this carefully: SpaceX maintains a disposal reliability rate above 99% and designs its satellites to fully burn up to minimize debris. Crucially, the hardware is not brought back. Retrieving retired satellites from orbit for recycling or reuse is considered technically impossible and economically unfeasible, so the AI hardware inside an orbital data center is destroyed at end of life rather than recovered, unlike terrestrial hardware, which is remarketed, reused, or dismantled for materials.
Can orbital data center hardware be recycled or recovered?
Not in any practical sense. Once hardware is in orbit, retrieving it for recycling or reuse is technically impossible and economically unfeasible; there is no way to bring a retired satellite back to a dismantling facility. Instead, orbital hardware is de-orbited and burned up in the atmosphere at end of life. This is a fundamental difference from terrestrial data centers, where retired GPUs, servers, and networking equipment are recovered, with a large share of their value captured through remarketing, reuse, and materials reclamation. The residual value that terrestrial operators recover at every refresh is, for orbital hardware, structurally unrecoverable and effectively written to zero.
Does the end-of-life problem mean orbital data centers will not work?
No. It means the full lifecycle cost of orbital compute includes something terrestrial compute does not: a recovery-value write-off at every refresh. Orbital data centers may still make sense if launch costs fall far enough and the energy and constraint advantages prove decisive, and serious companies are betting they will. The point is not that orbital compute is doomed, but that honest comparisons between orbital and terrestrial infrastructure should account for the recoverable hardware value that orbit loses and the ground retains. It is a real cost that the current debate has largely omitted, not a fatal flaw.
How is SpaceX involved in orbital data centers?
SpaceX is the highest-profile entrant and, structurally, the most important. It filed with the FCC in early 2026 for up to one million orbital data center satellites and has framed orbital compute as a way to bypass Earth’s strained power grids. More fundamentally, SpaceX operates the launch capability the entire concept depends on: the economics of orbital data centers rest on launch costs continuing to fall, which is exactly what SpaceX has driven for two decades and aims to accelerate with Starship. SpaceX also sets the industry standard for responsible satellite disposal, with a de-orbit reliability rate exceeding 99%, so it is central both to whether orbital compute becomes viable and to how responsibly its hardware is retired.
How does orbital compute compare to terrestrial data centers on sustainability?
It is a genuine tradeoff, not a clear win for either. Orbital compute avoids terrestrial water use, land use, and grid strain, real sustainability advantages given the pressure those put on communities. But it introduces its own end-of-life dynamic: hardware is destroyed by atmospheric re-entry rather than recovered, so the materials and residual value that terrestrial recycling captures are lost, and the atmospheric effects of large-scale satellite re-entry are an area of active regulatory and scientific attention. Terrestrial data centers, by contrast, can recover and reclaim their hardware at end of life. A full sustainability comparison has to weigh orbit’s resource advantages against its loss of recoverability and its atmospheric disposal.
The Bottom Line
Data centers in space have crossed from science fiction into genuine, well-funded engineering. Starcloud has flown a working H100 and trained a model in orbit, SpaceX has filed to launch up to a million data center satellites and brings the launch capability the whole concept depends on, and Google is developing its own orbital compute project. The advantages are real: abundant continuous solar power, no cooling water, no land, and an escape from the grid and permitting constraints squeezing terrestrial growth. The economics are marginal today, roughly three times terrestrial cost, and depend on launch prices falling toward $200 per kilogram in the 2030s, a trajectory tied directly to SpaceX’s continued progress. This is a serious long-horizon infrastructure bet by serious people.
But the debate has a blind spot, and it is the one an infrastructure disposition specialist notices first. Orbital hardware, like terrestrial hardware, reaches end of life in a handful of years, and when it does, it is not recovered. It de-orbits and burns up. Responsible operators like SpaceX handle that disposal safely, with a debris-minimizing discipline that sets the industry standard, but disposal is not recovery. Retrieving orbital hardware for reuse or materials is technically impossible and economically unfeasible, so the residual value that every terrestrial data center captures at each refresh is, in orbit, written to zero and burned up on re-entry. That is not an argument against orbital compute, which may well have its moment as launch costs fall. It is a real and recurring cost that the current conversation has almost entirely omitted, and a reminder that the recoverability of terrestrial hardware, easy to overlook amid the focus on building and powering AI, is a genuine and underappreciated advantage of building on the ground.
How ROC Telecom Fits In
ROC Telecom is an R2v3, RIOS, NIST 800-88, and ITAR-compliant ITAD specialist, and the orbital end-of-life problem is, in a sense, the clearest possible illustration of what terrestrial asset recovery is worth. Every advantage orbital compute offers, orbit cannot recover its hardware, while terrestrial operators can. That recoverable value is exactly what ROC helps operators capture:
- Data center decommissioning with 48-hour rapid-response mobilization, treating retired hardware as an asset to be recovered rather than a cost to be disposed of
- GPU and AI infrastructure asset recovery for high-density hardware like H100 and A100 fleets, with speed-to-remarketing that captures the residual value orbital operators simply lose
- Specialist asset recovery across routing, switching, optical transport, and compute, with direct buyer relationships
- NIST 800-88 data destruction with per-asset serialized Certificates of Destruction and full chain-of-custody documentation
- R2v3 Appendix E materials recovery with in-house dismantling and direct-to-refiner processing, reclaiming the materials that orbital hardware burns up
- Mass-balance recovery reporting for the ESG and sustainability disclosures terrestrial operators can make and orbital ones cannot
15+ years of ITAD experience, $25M+ in client capital recovered, 45M+ pounds diverted from landfill.
Request a Free ITAD Program Assessment
Tell us about your infrastructure and what you are building, refreshing, or retiring. A specialist will reach out to discuss decommissioning timelines, recovery value, and certified disposition. No commitment, no spam.
Prefer to talk directly? Call 585-406-1249 or email info@roctelecom.com.
"*" indicates required fields
Related reading:
- Data Center Asset Recovery: The 2026 Strategic Guide
- How GPU Decommissioning Differs from Standard Server Retirement
- Why AI Is Shortening Your Hardware Lifecycles (and What to Do About It)
- How Much Electricity Do Data Centers Use? The 2026 Numbers
- How Much Water Does a Data Center Use? The 2026 Numbers
- The Environmental Impact of Responsible Circuit Board Recycling
- Nuclear Power and Data Centers: The 2026 Energy Overview