The numbers that have driven the orbital computing conversation are not imaginary. Data centres already account for roughly 415 TWh of global electricity consumption, and the International Energy Agency projects that figure could double by the end of the decade as AI model training and inference workloads scale. Planning permission for new terrestrial facilities is becoming contentious across the United States and Europe. Water use is scrutinised. Grid capacity is straining.
Against that background, the logic of moving the most power-hungry compute into orbit, where solar panels receive uninterrupted, atmosphere-free sunlight around 36 percent more intense than at ground level, has a basis at least in theoretical physics.
The same argument applies to telecoms. Megaconstellations of low-Earth orbit satellites have extended broadband connectivity to regions that terrestrial infrastructure has never reached. Starlink’s ten million subscribers (as of February 2026) are real customers with real connectivity. The case for orbital infrastructure is there.
In November 2025, Google unveiled Project Suncatcher, an 81-satellite sun-synchronous constellation for solar-powered AI compute, with a demonstration mission planned for 2027. On January 30 2026, SpaceX filed an application with the Federal Communications Commission for an orbital data centre constellation of up to one million satellites, operating between 500 and 2,000 kilometres altitude.
This article isn’t going to go into questions which have already been rehearsed on the feasibility or otherwise of running data centres in space. We don’t have the engineering chops and, in any case, what is currently unfeasible as far as we know may change with technology breakthroughs that haven’t made the news yet (although having said that, for those interested this video is quite an entertaining science-based hit piece).
What is unfeasible, or at least wildly optimistic, is the pace and scale of growth which is now being proposed.
In addition to the data centres outlined above, the announced pipeline of new satellites is staggering. SpaceX has authorisation for up to 42,000 Starlink satellites quite apart from its additional million. China’s Guowang constellation plans 13,000 satellites; the Qianfan / Thousand Sails programme targets 15,000 by 2030. Amazon’s Project Kuiper plans approximately 3,200 satellites. Blue Origin has announced its TeraWave high-throughput network. If even a modest fraction of these commitments is fulfilled, the orbital population would increase by an order of magnitude within a decade.
Note that these are not all engineering programmes right now by any means. They are regulatory filings and press announcements. We’ll come back to that.
The authoritative live source for orbital population data is Harvard astrophysicist Jonathan McDowell’s Space Statistics tracker. As of late January 2026, McDowell counted 14,518 active payloads in Earth orbit. Of those, 9,555 belong to Starlink alone, representing approximately 65 percent of all operational satellites. When defunct spacecraft, spent rocket bodies, and tracked debris fragments are included, the total number of catalogued objects in orbit exceeds 47,000.
Trouble Getting it Up
The first question to ask is whether the rockets exist to launch all of these. The answer, presently, is no. Amazon’s situation is illustrative: the company is already seeking an FCC deadline extension on its Kuiper commitments, citing a shortage of launch vehicles.
SpaceX’s Falcon 9, which has enabled the Starlink build-out, launches roughly once a week. Starship, the vehicle SpaceX cites in its million-satellite filing as the enabler of unprecedented tonnage to orbit, has yet to complete a fully operational mission. New launch facilities take years to permit and construct, and their proliferation has its own environmental footprint.
If it seems like launch capacity renders the announcements hokum, there is a good reason for it. The International Telecommunication Union’s ‘first come, first served’ rule for orbital spectrum and slot allocation creates a perverse incentive. People should file early and file large, because the ITU will reduce your authorised constellation size if you do not deploy within seven years. Ambitious filings are, then, a regulatory strategy rather than a construction commitment.
The Debris Problem
The ESA Space Environment Report 2025 counted over 39,000 catalogued objects in Earth orbit at end of 2024: nearly 14,000 payloads, 2,000 rocket bodies, and 23,000 debris fragments. ESA’s MASTER debris modelling tool estimates there are approximately 1.2 million objects between one and ten centimetres, none of which can be individually tracked. A one-centimetre fragment travelling at orbital velocity carries enough kinetic energy to disable a satellite.
Debris accumulates through several mechanisms: satellite “fragmentation events” (explosions caused by residual fuel or batteries), anti-satellite weapons tests (Russia’s 2021 strike on Cosmos-1408 generated over 1,800 trackable fragments); and collisions. In 2024, a Chinese Long March 6A rocket broke apart in low Earth orbit producing hundreds of trackable fragments, a debris cloud that will remain in orbit for years and poses an ongoing risk to constellations operating below 800 kilometres. The ESA report records an average of 10.5 “non-deliberate fragmentation events” per year; 2024 produced over 3,000 newly catalogued fragments.
A December 2025 preprint by Sarah Thiele and colleagues, ‘An Orbital House of Cards: Frequent Megaconstellation Close Conjunctions’ introduces the concept of the CRASH [Collision Realization and Significant Harm] Clock, which measures how long it would take for a catastrophic collision to occur if satellite operators suddenly lost the ability to perform avoidance manoeuvres.
In 2018, the CRASH Clock was calculated to be between 121 and 164 days. As of mid-2025, the figure had collapsed to approximately 5.5 days owing to the growth of both satellites and debris. A 24-hour loss of avoidance capability now carries a 30 percent probability of a catastrophic collision.
Shell Shock
The operational reality of megaconstellation management is already extraordinary. Across all LEO megaconstellations, a close approach, defined as two objects passing within one kilometre of each other, occurs every 22 seconds.
For Starlink alone, that interval is 11 minutes. Each Starlink satellite performs an average of 41 avoidance manoeuvres per year, which translates across the 9,000-satellite constellation to approximately one manoeuvre every two minutes, continuously. Historically, that manoeuvre rate has been doubling every six months.
Sun-Synchronous Orbit [SSO], the orbital band preferred for near-constant solar power generation, and therefore the first choice for orbital data centre constellations, is already the single most congested highway in low Earth orbit. A satellite in SSO, according to the same analysis, currently encounters debris larger than a grain of sand approximately every five seconds.
The manoeuvre burden for any substantially larger SSO constellation will become horrific.
Collision probability in a given orbital shell scales roughly with the square of the number of objects present.¹ Even assuming SpaceX got 10% of their proposed million AI datacentres into orbit, that would represent approximately an eightfold increase in total active objects, implying a roughly 64-fold increase in close approach frequency. As a result, satellites in those shells would require near-continuous autonomous avoidance manoeuvring, burning fuel at a rate that would significantly reduce operational lifetimes and generate additional deorbit debris.
The Starlink Barrier: Risk in Transit
Any rocket launching to an orbit above approximately 550 kilometres must transit the dense band that Starlink has occupied at that altitude. As astronomer Samantha Lawler told IEEE Spectrum: “The way Starlink has occupied 550 km and filled it to very high density means anybody who wants to use a higher-altitude orbit has to get through that really dense shell. China’s megaconstellations are all at higher altitudes, so they have to go through Starlink… Really, everybody has to go through them, including ISS, including astronauts.”
The barrier is permeable. Rockets can and do pass through it, and the relative dwell time of a transiting vehicle in that band is brief. The risk is not zero though, and it compounds with the growing frequency of launches. A near-miss between a Starlink satellite and a Chinese rocket in late 2025 prompted SpaceX to lower more than 4,000 satellites from 550 to approximately 350 kilometres, which compresses the active shell downward but doesn’t eliminate the transit problem for any mission aimed at medium or high Earth orbit.
The question is not whether launches can currently pass through a Starlink-density shell, but what level of transit risk is acceptable as that shell becomes denser. If constellation sizes increase by the factors proposed, the risk profile for routine launches changes by an order of magnitude. A launch failure attributable to collision with constellation debris would have profound regulatory and insurance consequences for the entire industry, as well as generating a significant amount of debris a critical altitude.
Crashing Out
The underlying problem, described by NASA scientist Donald Kessler in 1978 and increasingly confirmed by observational data, is that beyond a critical density threshold, orbital collisions become self-sustaining. A single collision between two substantial objects generates thousands of fragments; those fragments increase the probability of further collisions; those collisions generate further fragments. In 2009, Kessler himself wrote that modelling results indicated the debris environment in low Earth orbit had already become unstable.
The authors of the CRASH Clock paper are careful to point out that a major collision is “more akin to the Exxon Valdez oil spill disaster than a Hollywood-style immediate end of operations in orbit.” The cascade process, if triggered, would unfold over decades rather than days, but its effects would be persistent and potentially irreversible for specific orbital shells. The ESA Space Environment Report 2025 states plainly that “there is a scientific consensus that even without any additional launches, the number of space debris would keep growing” through fragmentation of the current satellite population.
If that weren’t enough, solar storms throw another potential spanner into the works. In May 2024, the Gannon Storm – the strongest geomagnetic event in two decades – caused over half of all satellites in low Earth orbit to use fuel on repositioning manoeuvres, and orbital uncertainties expanded to kilometres. As Lawler noted in the IEEE Spectrum interview, when uncertainty is measured in kilometres and objects are travelling at seven kilometres per second, “that’s terrifying.” If a solar storm of that size were to disable satellite communications and navigation simultaneously, the CRASH Clock would begin running immediately.
The Center for Security and Emerging Technology’s 2025 mapping of space debris notes that over 83 percent of all tracked objects reside within low Earth orbit, and that a small number of fragmentation events are responsible for most of the current congestion. The implication is that one sufficiently large collision in a heavily occupied band could transform the risk for all operators in that shell for years.
With a CRASH clock currently running at days and multiples of extra satellites being proposed, that seems like a recipe for problems.
Announcement Versus Reality: A Grounded View
Against this background, it’s thoroughly reasonable to be sceptical of the largest recent satellite constellation announcements.
SpaceX’s filing for one million data centre satellites was submitted to the FCC on January 30, 2026. On the same day, Elon Musk announced SpaceX’s acquisition of xAI — the artificial intelligence company he controls, which has faced questions about funding sustainability and compute access. The one-million-satellite figure eclipsed coverage of the merger. Could this be a case of ‘flooding the zone’ from the Steve Bannon playbook?
Amazon’s position is rather different. Kuiper made their announcement, despite already facing a shortage of rockets. Amazon does not own a launch provider and is dependent on a market where demand significantly outstrips supply. We might take this announcement… well, yes, with a pinch of salt, but also as a buying signal to the market, encouraging some competition there.
It would be an overstatement to characterise all large orbital announcements as deliberate distractions from earthly difficulties. Some fraction of the ambitions expressed in these filings will be realised. Google published a blog in November which concluded that orbital data centres might become cost-competitive with terrestrial facilities if launch costs fall to roughly $200 per kilogram, which might arrive around 2035 if Starship scales to 180 launches per year. That is a possible scenario, albeit not imminent and still subject to argument.
The more immediate reality is captured in the ESA’s new Space Environment Health Index, introduced in its 2025 report. If current behaviour continues, “the risk level passes beyond the point of sustainability.” An analysis published in October 2025 draws an uncomfortable parallel with the early stages of climate change awareness. Namely, this is a situation where individual operators have short-term incentives to ignore long-term collective costs, and where the international regulatory framework lags behind deployment.
Smoke and Orbital Mirrors
But the ITU filing incentive structure, the gulf between announced and deployed satellite numbers, and the pattern of headline-generating announcements coinciding with moments of corporate or reputational pressure all warrant scrutiny.
The debate about space-based data centres and the next generation of communications constellations has become focussed on whether the technology is physically possible, and that’s something which changes with technology advancements. It’s stirred up conversation and attention, which is great for accessing investments.
If we frame this as a debate about physical constraints, governance and honest accounting of risk, the picture changes. Responsible assessment of any future constellation announcement should ask: Does the launch capacity exist? Does the regulatory and debris management framework exist? Does the proposed orbital band safely have capacity for the additional density?
If any of those answers are ‘no’ then people are putting time and money into something fundamentally unstable, and that might change the calculation both for investors and for people soaking up these grand announcements.
