Can Tech Companies Really Put Data Centers in Space? Enterprise backup software
By Gammatek ISPL, Industrial Systems & Compliance Analyst at Gammatek ISPL
Last updated: September 2026 | 14 min read
Author block: Gammatek ISPL covers infrastructure and technology trends affecting industrial and enterprise operations at Gammatek ISPL. This piece draws on publicly reported technical filings, funding disclosures, and industry analysis current as of September 2026.
Why This Matters
If you've assumed "data centers in space" is a sci-fi thought experiment, it's time to update that assumption — because it's already happening, and the company doing it just raised over a hundred million dollars to do a lot more of it. This matters to anyone in tech, infrastructure, or industrial operations for a concrete reason: AI's computing demand is colliding with a very real, very physical wall on Earth — power grid capacity, water for cooling, and land for new facilities are all running short in the regions where demand is highest. If space-based computing works even partially, it changes where the next decade of AI infrastructure gets built, who builds it, and what skills and software become valuable in the process. This isn't a distant hypothetical — the first orbital AI chip has already run a real workload.
What's Actually Happening Right Now
The clearest proof this isn't speculative: a Seattle-area startup called Starcloud — originally named Lumen Orbit — placed an NVIDIA H100 GPU into orbit aboard a small satellite in November 2025, the first time a GPU of that class had operated in space. The company has raised $200 million in total funding, including a $170 million Series A round in March 2026 that valued it at $1.1 billion, backed by investors including Benchmark and EQT Ventures.
What happened next moved fast. Starcloud used that first satellite to train a large language model in orbit and has run and queried Google's Gemma model directly in space, while also processing satellite imagery for applications like wildfire monitoring. The funding push is explicitly tied to a real bottleneck: rising AI compute demand is straining power grids and slowing new data center construction on Earth, and Starcloud's pitch is to sidestep that limit entirely by moving computing into low Earth orbit, where satellites can draw on continuous solar power.
Starcloud isn't operating alone in this space (pun unavoidable). Industry tracking now counts roughly eight companies actively competing to build computing infrastructure in orbit, ranging from early filings to funded pre-launch stages, with players spanning from SpaceX's large-scale satellite constellation filings to Google's own orbital TPU cluster research. Starcloud's own regulatory filings describe an eventual constellation of up to 88,000 satellites designed to process data directly rather than simply relay signals, envisioning a multi-gigawatt orbital computing cluster powered by an enormous solar array.
That's the headline version. The realistic version — the one worth actually planning around — is considerably more complicated.
The Case For Space: Why Anyone Would Try This
Three genuine engineering and economic arguments make this more than a stunt:
1. Uninterrupted solar power. A satellite in the right orbit can receive sunlight nearly continuously — no clouds, no nighttime, no grid congestion. For power-hungry AI training workloads, that's a meaningfully different energy equation than a terrestrial facility competing for grid capacity against homes, hospitals, and other industries in the same region.
2. No land, water, or zoning constraints. Terrestrial data centers increasingly face local opposition over water use for cooling and strain on regional power grids. Space removes that entire category of conflict — there's no neighborhood to object to a satellite.
3. Falling launch costs. The economics of getting mass into orbit have dropped sharply over the past decade due to reusable rocket technology, changing the math on what's financially viable to put in space versus what used to be prohibitively expensive.
None of these are hypothetical advantages — they're the actual stated rationale behind the funding rounds happening right now.
Comparison: Terrestrial vs. Orbital Data Centers
Terrestrial Data Center | Orbital Data Center | |
Power source | Grid electricity, increasingly capacity-constrained | Near-continuous solar, effectively unconstrained by grid capacity |
Cooling method | Water/air-based active cooling systems | Radiative cooling only — no air or water available in vacuum |
Land/siting constraints | Significant — water rights, zoning, local opposition | None — no physical neighbors |
Hardware access for repair | On-site technicians, same-day hardware swaps | Effectively none — no physical access once launched |
Deployment timeline | Months to years for permitting and construction | Constrained by launch manifests and satellite production, not permitting |
Redundancy requirements | Standard enterprise redundancy practices | Extreme — a single hardware failure may be permanent and unfixable |
Cost structure today | Well-understood, mature market pricing | Still early-stage; true cost-per-compute-hour not yet proven at scale |
The Hard Engineering Problems Nobody's Marketing Deck Emphasizes
Cooling is fundamentally different in a vacuum. On Earth, data centers cool servers using air or liquid — both rely on the presence of a medium to carry heat away. In space, there's no air and no water to convect heat into; the only way to shed heat is radiative cooling, releasing it as infrared radiation into space. This requires much larger radiator surface areas than most people assume, and it's one of the primary engineering constraints shaping how large an orbital cluster can realistically get before cooling becomes the bottleneck rather than compute power itself.
Radiation exposure degrades chips over time. Consumer and even most enterprise-grade GPUs aren't designed for the radiation environment outside Earth's atmosphere. Extended exposure causes gradual degradation and occasional bit-flip errors that terrestrial hardware never has to handle. Any orbital computing plan has to either radiation-harden the hardware (expensive, and it reduces raw performance) or accept a shorter usable hardware lifespan than an equivalent terrestrial GPU cluster.
There is no "send a technician" option. This is the constraint that most differentiates orbital computing from every other category of enterprise IT infrastructure, and it's the one with the most direct relevance to how these systems actually have to be run day to day.
The Operational Software Problem — Where This Gets Genuinely Interesting
Here's the part of this story that gets skipped in most coverage, and it's the part that actually matters if you work in enterprise IT or industrial operations: an orbital data center is, functionally, the most extreme "remote, unreachable infrastructure" problem in enterprise computing.
On Earth, if a server fails, someone walks into the data center and swaps a drive. In orbit, that's not an option for years, possibly ever, for a given satellite. That single constraint cascades into serious requirements on the software layer running these systems:
Backup and recovery has to be treated as the primary failure-prevention layer, not a secondary safety net. A terrestrial enterprise backup and recovery system exists partly as insurance against failures that on-site staff can also physically respond to. In orbit, backup and recovery software isn't a fallback — it's the only line of defense, because there's no secondary physical intervention possible. This is pushing orbital computing companies toward far more aggressive, automated, and geographically (or orbitally) distributed redundancy than even the most cautious terrestrial enterprise backup software deployment would typically require.
Network monitoring has to be autonomous and predictive, not reactive. A typical enterprise network monitoring software setup on Earth flags an issue and a technician investigates. The best enterprise network monitoring software for orbital infrastructure has to do more — it has to predict degradation before it becomes failure, because by the time a hardware fault is confirmed, there may be no way to address it at all. This is functionally the same problem industrial plants face with aging, hard-to-access equipment, just taken to its most extreme version.
Patching and updates have to work with zero room for error. Enterprise patch management software on Earth can tolerate an occasional failed update — you roll it back, or a technician intervenes. An update pushed to orbital hardware that bricks a system has no fallback. This is pushing orbital computing firms toward far more conservative, heavily tested patch management practices than most terrestrial enterprise IT teams currently operate under.
An Implementation Consideration From the Ground
This might sound like a problem unique to satellite engineers, but the underlying pattern — critical infrastructure that's genuinely hard or impossible to physically access, where monitoring has to predict failure rather than just report it — is exactly the same category of challenge industrial plants face with remote or hard-to-reach equipment: offshore facilities, remote pipeline infrastructure, equipment inside hazardous processing environments where sending a technician is expensive, slow, or dangerous.
The specific engineering is obviously different — nobody's radiation-hardening a factory floor sensor. But the operational philosophy — treat monitoring as predictive rather than reactive, because physical access is the constraint, not compute power — is the exact principle behind predictive maintenance platforms already running in manufacturing and processing plants today. Orbital data centers didn't invent this problem; they've just taken an existing industrial challenge to its most extreme version.
Is This Actually Economically Viable Yet?
Here's where healthy skepticism belongs. Launch costs have fallen, but they haven't fallen to zero, and getting a gigawatt-scale computing cluster into orbit is a fundamentally different mass and cost problem than launching a communications satellite. Starcloud's own roadmap targets a five-gigawatt orbital cluster requiring a solar array spanning four square kilometers — an enormous undertaking that has never been attempted at that scale, and the true cost-per-compute-hour once you account for launch, radiation-hardening, redundancy overhead, and eventual hardware replacement hasn't been proven out at commercial scale yet. Early satellites like Starcloud-1 have demonstrated that orbital AI computing works technically; they haven't yet demonstrated that it's cost-competitive with terrestrial data centers at meaningful scale.
The honest read: this is a real, funded, technically-proven-at-small-scale trend, not vaporware — but "proven to work" and "proven to be economically better than building another data center in Texas" are two different claims, and only the first one is settled right now.
Where This Is Actually Headed
The realistic near-term outcome isn't "all data centers move to space" — it's a hybrid model where specific, high-value, power-intensive workloads (large model training runs, for instance) get offloaded to orbital capacity as a supplement to terrestrial infrastructure, not a replacement for it. That's a meaningfully different story than the "AI companies are literally launching entire data centers into orbit" framing that tends to dominate headlines — and it's the version worth actually planning around if your work touches enterprise infrastructure decisions in the next few years.
The Compliance and Monitoring Layer This Trend Points Back To
Whether the infrastructure in question is a satellite in low Earth orbit or a processing unit on a factory floor, the same principle is becoming harder to avoid: as equipment gets more remote, more automated, and more expensive to physically access, the software layer monitoring it stops being a convenience and becomes the actual safety and compliance backbone of the operation. That's the same shift driving predictive maintenance and compliance monitoring adoption across manufacturing and industrial plants today — orbital data centers are simply the most extreme, most visible version of a pattern already reshaping how ground-based industrial facilities operate.
Comments