Will the reentry of a Starlink satellite cause damage to a structure on earth before 2030?

TL;DR

I estimate a 3% chance that a Starlink reentry will be credibly confirmed to damage a human-made structure on Earth before 2030.

Context

Starlink is already a large reentry system, not just a launch system. Jonathan McDowell’s point-in-time Starlink table, last updated on 2026-05-11 10:04 UTC, lists 11,979 Starlink satellites launched, 1,604 down, 10,375 still in orbit, 10,359 working, 281 in disposal, and 16 failed/decaying; this is a full-program count from the 2018 Tintin prototypes through May 2026, not a recent sample (McDowell Starlink statistics).

The question is much narrower than “will Starlink debris reach the ground?” It needs a Starlink reentry fragment, a surface impact, material damage to a human-made structure, and credible public attribution before 2030-01-01 UTC. SpaceX says current Starlinks are designed for demise and targeted ocean reentry, but SpaceX has also acknowledged one Starlink fragment survived reentry in 2024 (Starlink demisability note).

Evidence

The historical backbone points to a low base rate. A 2025 Frontiers paper using Space-Track/TLE data gives the full published annual Starlink reentry history through 2024 as 2 in 2020, 78 in 2021, 99 in 2022, 88 in 2023, and 316 in 2024, for 583 Starlink reentries over 2020–2024; its detailed analysis covered 523 Starlinks with adequate TLE histories, so the time-series sample is hundreds, not thousands (Frontiers, 2025). McDowell’s later inventory raises the cumulative full-program count to 1,604 down by 2026-05-11, with zero confirmed Starlink structure-damage events found in public reporting as of this forecast (McDowell Starlink statistics). Absence of evidence is not proof of safety, because ocean and remote-land impacts are usually invisible. But a public, structure-damaging Starlink impact would be much easier to notice than harmless debris in a field.

The strongest positive evidence is the Saskatchewan fragment. SpaceX says that after the 2024-07-11 Falcon 9 Group 9-3 anomaly, all Starlinks from that mission reentered from a very low 135 km perigee; on 2024-08-20, a 2.5 kg aluminum piece was found on a farm in Saskatchewan, Canada, and SpaceX engineers traced it to a specific Starlink direct-to-cell satellite part, a modem enclosure lid for a backhaul antenna (Starlink demisability note). SpaceX also says that part had been predicted to fully demise by both NASA and ESA tools and was the only known Starlink fragment not to have done so as of the note (Starlink demisability note). That did not damage a qualifying structure and occurred before this question’s window. It still proves the main physical premise.

The strongest negative evidence is targeting and low impact energy. SpaceX says its targeted reentry method can place the potential debris ellipse over open ocean, away from populated islands and heavy air and maritime routes, with entry-point tracking to about 10% of an orbit ground track, or roughly 10 minutes (Starlink demisability note). SpaceX says current V2 Mini and future V3 satellites have enough control authority for targeted ocean reentry, that the V2-mini-optimized design has modeled human casualty risk below 1 in 100 million per reentry, and that the design target is under 3 joules impact energy at component level (Starlink demisability note). I discount vendor claims, but they match the observed record: many Starlink reentries, one known ground fragment, no confirmed Starlink structure damage.

The broader reference class says structure damage from reentry debris is real. On 2024-03-08, a 0.7 kg metal support from an ISS battery cargo pallet hit a home in Naples, Florida, after NASA had expected the 5,800 lb pallet hardware to burn up; NASA later confirmed the object’s origin, and AP reported that it tore through the roof and damaged flooring (AP/NASA report). That is a useful analogue for attribution and damage mechanics. It is not a direct Starlink base rate, because the ISS pallet was much larger and not a demisability-optimized Starlink satellite.

There are also new weak signals on the risk side. A Starlink satellite suffered an anomaly on 2025-12-17 at about 418 km, lost communications, released trackable low-relative-velocity objects, and was expected by Starlink to reenter and fully demise within weeks (Space.com, 2025-12-18). Another Starlink, satellite 34343, had an anomaly on 2026-03-29 at about 560 km; SpaceX said it lost communications, and LeoLabs reported a fragment-creation event with tens of nearby objects, likely from an internal energetic source rather than a collision (Scientific American, 2026-03-31). These events do not themselves imply ground damage. They do raise my tail-risk estimate for future uncontrolled or off-nominal reentries.

For exposure, the target area is small. An older global constructed-impervious-surface estimate found 579,703 km², or 0.43% of world land area and about 0.11% of Earth’s surface, covering buildings, roads, parking lots, driveways, sidewalks, and similar surfaces (ORNL impervious-surface study). A broader 2020 Landsat-derived estimate gives 4.3 million km² of global built-up lands, about 0.84% of Earth’s surface, but that includes settlement/infrastructure categories that will not always be a qualifying damageable structure under this question (Frontiers Remote Sensing, 2022). I use an effective qualifying-hit fraction around 0.1%–0.3% for a random kilogram-class fragment, before adjusting for impact energy, ocean targeting, and public attribution.

My quantitative model is a Poisson hazard model. I project about 5,100 future Starlink reentries from 2026-05-12 through 2029-12-31, with an 80% range of about 3,000–8,000; this is my inference from McDowell’s 10,375 in-orbit count, 281 disposal-underway count, 3,345 Gen1 satellites still in orbit, and the five-year design-life turnover discussed in public Starlink reporting (McDowell Starlink statistics; Space.com Starlink facts). I split that into roughly 4,300 controlled or semi-controlled ocean-targeted reentries at 1.2×10^-6 confirmed-damage probability each, 800 failed or less-controlled reentries at 7×10^-6 each, and a separate tail-risk term of 0.018 expected events for another G9-3-like deployment anomaly, V3/Starship transition surprise, or repeat internal-fragmentation failure mode.

That gives:

$$\lambda = 4300(1.2\times10^{-6}) + 800(7\times10^{-6}) + 0.018 = 0.02876$$ $$P(YES)=1-e^{-0.02876}=2.8\%.$$

I round this in prose to 3%. The result is above a pure “Starlinks fully burn up” model, but well below a naive “thousands of satellites means one must hit something” model.

What's non-obvious

The main mistake is treating demisability as binary. SpaceX’s own evidence says some V2 Mini mass can survive as tiny, low-energy fragments, and the Saskatchewan part shows that a kilogram-class Starlink component can survive when attitude, breakup altitude, and modeling assumptions are wrong (Starlink demisability note). That keeps the probability out of the near-zero range.

The other mistake is scaling directly from satellite count. Even a dangerous surviving fragment usually lands in ocean, open land, forest, desert, or farmland. A Starlink fragment must also strike a qualifying structure, cause material damage, be recovered or otherwise tied to the event, and be credibly attributed specifically to Starlink. Targeted ocean reentry changes the base rate more than the raw number of reentries does.

Limitations

The largest gap is proprietary engineering data. Public sources do not give a component-by-component survival and impact-energy distribution for every Starlink generation, nor do they let outside analysts verify SpaceX’s below-1-in-100-million casualty-risk claim for V2-mini-optimized satellites (Starlink demisability note).

The second gap is survivorship and reporting bias. The known Starlink fragment count is not the true fragment count. Most debris that lands in ocean or remote terrain is never found, while debris that damages a roof in a developed country is likely to be recovered and analyzed.

The third gap is future fleet change. V3 satellites, Starship deployment, constellation lowering, higher launch cadence, and repeated on-orbit anomalies could change both the number of reentries and the off-nominal fraction before 2030. I treat those as tail risk rather than the central case because the current observed record is still 1,604 Starlinks down and no confirmed Starlink structure-damage event (McDowell Starlink statistics).