Dead Satellites Released 20 Tons of Aluminum Above the Ozone Layer in 2025. SpaceX Just Filed for a Million More Satellites.
No agency has ever done an environmental review of what burning satellites do to the atmosphere. Updating the published science with 2025 data, we count roughly 20 tons of aluminum oxide injected above the ozone layer last year, about eight times the natural background, riding a 30-year fuse down to the altitudes where ozone lives.
Five hundred forty-eight. That is how many satellites would burn up in Earth’s atmosphere every single day if SpaceX builds the full constellation it asked the Federal Communications Commission to approve in January, a fleet of up to one million spacecraft meant to serve as solar-powered data centers for artificial intelligence. Each would last about five years, so do the division: a million satellites over five years is 200,000 dead satellites a year, raining down as vaporized metal, even as SpaceX told the FCC these orbital data centers would “significantly reduc[e] the environmental impact associated with terrestrial data centers.” Notice what the filing never mentions: it says nothing about what 200,000 annual reentries would do to the sky itself, so we ran the numbers the filing skipped.
A dying satellite does not vanish: at 86 kilometers up, air friction heats it past 2,200 Kelvin and the aluminum skin ablates, reacting with oxygen as it peels away. A team led by José Ferreira at the University of Southern California ran the first atomic-scale simulation of this process, published in Geophysical Research Letters in 2024, where one 250-kilogram satellite, about the size of an early Starlink, yields roughly 30 kilograms of aluminum oxide nanoparticles plus another 51 kilograms of unoxidized aluminum clusters, since only about a third of the aluminum fully oxidizes. These particles measure 0.4 to 4.2 nanometers across and need up to 30 years to settle from the mesosphere down to 40 kilometers, the altitude where ozone chemistry happens, three full decades of falling.
Daniel Murphy went looking for them, and in 2023 the NOAA scientist flew a NASA WB-57 through the stratosphere over Alaska with a laser mass spectrometer in its nose. Murphy’s PNAS paper reported the fingerprints: about 10 percent of large sulfuric acid particles contained aluminum and 20-plus other elements in spacecraft-alloy ratios, with niobium and hafnium, superalloy metals absent from natural dust, clinching a human source. Murphy’s conclusion was blunt, because the mass of several metals from spacecraft reentry now exceeds the cosmic dust influx of those metals, and planned constellation growth could put reentry metals in half of all stratospheric sulfuric acid particles, a full half.
Then the totals started climbing, fast. Atmospheric chemist Connor Barker reported at the European Geosciences Union meeting in April 2025 that aluminum and nitrogen oxide emissions from reentries nearly doubled between 2020 and 2022, from 3,300 to 5,600 metric tons. By 2022, reentry aluminum oxide was exceeding the natural meteoric supply by a factor of seven. Sit with that ratio. For every ton of aluminum oxide that meteors have always delivered to the upper atmosphere, dead satellites were already delivering seven, and 2022 was before the great Starlink retirement wave even began.
Counting 2025: A Ledger
Nobody has updated the ledger since Ferreira’s paper, so we did, using numbers ESA published in its Annual Space Environment Report, issued September 8, 2026: 486.7 metric tons of material reentered in 2025, a new record and 46 percent more than 2022. Ferreira’s published method converts reentry mass to aluminum oxide at a fixed yield, 16.6 tons of oxide from 308.9 tons of satellites in 2022. Applying the same yield to 2025’s 442.8 tons of low-orbit reentries gives between 15 and 24 tons of aluminum oxide, the range reflecting whether rocket bodies oxidize exactly like satellites, so call it 20 tons. Roughly eight times the natural background, and about ten times what satellites produced in 2016. Starlink alone accounts for most of it. Jonathan McDowell’s tracking shows the constellation incinerating one to five satellites a day through 2025, roughly 900 for the year, good for about 14 of those 20 tons.
| Year | Low-orbit reentry mass | Aluminum oxide from satellites | Multiple of natural background |
|---|---|---|---|
| 2016 | 121.8 t | 2.1 t | ~1x |
| 2022 | 308.9 t | 16.6 t | ~7x |
| 2025 (est.) | 442.8 t | ~20 t (15–24) | ~8x |
| 2030 scenario | ~2,400 t | ~129 t | ~54x |
That 2030 row is our own projection; the table shows the arithmetic. Now the million-satellite filing: 200,000 reentries a year at 500 kilograms each is roughly 5,400 tons of aluminum oxide annually, more than two thousand times natural. To be fair, nobody expects the full million to fly. But even one percent of the filing, 10,000 satellites, implies 2,000 reentries a year and 54 tons of oxide, 23 times the natural background. That application remains under FCC review, with the comment period closed since March.
A Review That Never Happened
Here is the embarrassing part. No American agency has ever conducted an environmental review of what satellite reentries do to the atmosphere, not one, and the reason is a categorical exclusion buried in FCC procedure: the agency excludes all of its licensing decisions from the National Environmental Policy Act, and the D.C. Circuit upheld that exclusion for Starlink in 2024. FAA’s own report on large-constellation reentry risk is entirely about debris hitting people and airplanes, and nobody owns the chemistry of the vaporized 95 percent, which is why Public Employees for Environmental Responsibility filed formal comments in July demanding a NEPA review of the million-satellite proposal. SpaceX’s response promised a “phased deployment approach to monitor actual atmospheric effects.” Read that promise carefully: monitoring an experiment is not the same as reviewing it before it starts.
And the irony: a 2022 FCC rule ordering five-year deorbits, meant to protect orbit, means fewer collisions but more aluminum in the sky, sooner. One regulator’s cleanup is another regulator’s pollution, and the second regulator does not exist.
A Thirty-Year Fuse
Cruelest of all is the timing. Ferreira’s particles need up to three decades to drift down to ozone altitudes, which means the aluminum injected in 2025 arrives around 2055. WMO puts ozone recovery to 1980 levels at 2040 for most of the planet in its 2022 assessment, 2045 for the Arctic, and 2066 for the Antarctic. Overlay the two timelines: that pollution wave we are launching now reaches the ozone layer precisely during the decades the Montreal Protocol spent forty years buying. We fixed the ozone layer by banning chemicals; we may be unfixing it by burning metal, thirty years delayed, with no treaty and no review.
Why This Might Be Wrong
Stated fairly, the ozone damage is chemistry, not yet measured harm. Nobody has modeled actual ozone loss from reentry aluminum; the 2 percent activation figure comes from laboratory surfaces. Ferreira is careful to say “potential” throughout. Consider scale, because it cuts both ways: the CFC era put chlorine into the stratosphere measured in parts per billion, from megatons of chemicals, while satellites contribute tens of tons of oxide a year. The Montreal Protocol is visibly working, with small ozone holes in 2024 and 2025. Leaving dead satellites in orbit is the alternative, and it threatens the Kessler cascades that could deny everyone the use of low orbit. SpaceX’s filing is a ceiling used for negotiating flexibility, not a construction plan, and treating the maximum as a forecast is a category error. Aluminum is not a chlorofluorocarbon. It does not linger as a gas for a century. Murphy himself says the actual effects remain unknown, and unknown is not catastrophic. Also true: every one of these defenses was available to the CFC industry in 1974, the year Rowland and Molina published, when the damage was also only chemistry.
What This Does Not Prove
Honest accounting, because an extrapolation is not a measurement. Our 2025 number assumes the 2022 fleet’s composition still describes today’s, and that 15-to-24-ton range exists precisely because rocket bodies may or may not oxidize like satellites. That 900-satellite Starlink count is modeled from McDowell’s stated daily rates, not a direct tally of his catalog, while Barker’s seven-times-natural figure comes from a conference presentation, not a peer-reviewed paper, and we could not verify a published version. That 30-year settling time is a model result sensitive to particle-size assumptions, and nanoparticle coagulation in the mesosphere is poorly understood. None of this weakens the core finding; it bounds it. Certainty about the direction, uncertainty about the exact tonnage, and a growth rate near tenfold per decade.
What You Can Do
If this bothers you, the venue is specific. Constellation licensing happens at the FCC under a categorical NEPA exclusion that has never been revisited for the megaconstellation era, which makes PEER’s July comments, now part of the record on the million-satellite application, more important than they look. Public comments on FCC satellite dockets are the only lever that exists, because the exclusion means nobody is required to ask for them, so ask one question: why has the FCC never studied the atmospheric effects of the industry it licenses? That 2024 D.C. Circuit ruling is the wall, though walls get revisited when the facts change. For the technically inclined, watch three things: Murphy’s follow-up flights, the WMO’s 2026 ozone assessment, and the FCC’s pending decision. For operators: start the review at the design phase, and treat aluminum as a choice with an atmospheric price. And the next time someone calls a satellite “fully demisable,” remember Saskatchewan. In March 2025 a 2.5-kilogram chunk of a Starlink fell on a Canadian farm, the piece SpaceX said could not exist.
Bottom Line
The 20th century taught us the sky is not an infinite sink: first CFCs, then carbon. This century is running the same experiment with vaporized spacecraft, except this time there is no treaty, no review, and no measurement program worthy of the name. Twenty tons a year: small against the CFC era, large against nature. That number has grown roughly tenfold in nine years. Those particles we add today arrive at the ozone layer in 2055, right on schedule for the recovery we spent forty years earning. Ask not whether burning hundreds of satellites a day changes the stratosphere. Ask whether anyone will be measuring when it does.