Space · Monday, 20 July 2026

01 · Briefing · what happened

Each Starlink satellite now swerves 40 times a year — and nobody has to count them

Space 6 min 80 sources

Low Earth orbit is getting crowded fast, and the dodging that keeps it safe is invisible to regulators. Plus a Starship abort, a $17 billion launch contract, and a first attempt at spotting nuclear weapons in space.

Key takeaways

Each Starlink satellite now makes more than 40 collision-avoidance moves a year, yet no rule requires operators to tell regulators how many their constellation will need.
Crowding multiplies risk faster than it adds satellites: double the spacecraft in an orbital shell and you roughly quadruple the pairs that must be watched.
Money is pouring into orbit — $8.1 billion into satellite firms in six months, and a US launch contract ceiling tripled to $17 billion — largely on defence demand.

Orbit’s busiest job is the one nobody logs

Between June 2025 and May 2026, the average Starlink satellite performed more than 40 collision-avoidance manoeuvres — small engine burns to move out of something’s way [3]. That is roughly one swerve every nine days, per satellite, across a constellation that grew from about 6,000 spacecraft in 2024 to more than 10,000 by June 2026 [3]. Over the same stretch, the total number of working satellites in orbit rose from roughly 10,000 to about 16,000 [3].

The dodging works. Starlink’s system moves automatically whenever the computed chance of a collision passes 3 in 10 million, and a manoeuvre cuts the risk to around one in a million [3]. But Hugh Lewis, an astronautics professor at the University of Birmingham, points out what happens when you do that a million times. “You end up with an aggregate risk across your entire constellation that you can’t get rid of,” he told Space.com [3]. SpaceX is on course to pass a million lifetime manoeuvres as early as June 2027, and by 2030 could be making more than a million every year [3].

The mechanism to hold onto is the pairing problem. Crowding does not add risk one satellite at a time. Tommaso Sgobba of the International Association for the Advancement of Space Safety puts it plainly: double the satellites in an orbital shell and you roughly quadruple the number of pairs that have to be watched [3]. Risk grows faster than the population does.

There is a second problem underneath. Predicting where a satellite will actually be is hard, because air drag at these altitudes shifts with space weather, which changes constantly [3]. So operators cannot always tell a real threat from statistical noise. Sgobba’s phrase for it: satellites are “frequently dodging ghosts, burning fuel and shortening their operational lives in the process” [3].

None of this is a surprise, and that is his argument. He wants the expected number of manoeuvres a constellation will need each year disclosed to regulators before a licence is granted — not discovered afterwards through near-miss headlines [3]. Right now no such requirement exists [3]. Meanwhile the US is still standing up TraCSS, its civil space-traffic coordination system, which came up repeatedly at a mission-authorisation hearing this week [61]. Amazon’s LEO constellation and China’s Qianfan are both deploying into overlapping altitudes [3].

Why this reaches you: the shells being filled are the same ones carrying weather data, broadband, ship and aircraft tracking, and — increasingly — navigation.

Rockets: an abort, a hop, and an engine on the stand

SpaceX’s Starship Flight 13 aborted at the last second on Thursday, with the countdown clock already at zero at the Starbase pad in south Texas [34]. More than 11.5 million pounds of methane and liquid oxygen had been loaded [34]. The flight computers called the abort during the Super Heavy booster’s engine start sequence — “some of the engines didn’t start,” Elon Musk wrote [34]. Two Raptor engines are being replaced; Musk said a launch is probable early the following week [34][46]. The booster carries 33 engines, each producing more than half a million pounds of thrust, ignited in a staggered sequence [34].

Japan quietly took a first step toward reusability. On July 11, JAXA — Japan’s space agency — flew and landed its RV-X prototype at the Noshiro test site: a 40-second hop, 33 feet up, 50 feet sideways, touching down softly on the far side of the pad [76]. Tiny by any measure, but landing a rocket under power is something only a handful of vehicles have ever done [76]. RV-X is a precursor to a reusable first stage built with Mitsubishi Heavy Industries [76].

Rocket Lab fired up Archimedes, the engine for its larger Neutron launcher, on the test stand [13]. And SpaceX flew a previously-used booster for the 600th time, carrying more Starlink satellites to orbit [7] — a useful marker of how routine reuse has become in the six years since it was a stunt.

The money is arriving, mostly from defence

Investment in satellite companies hit $8.1 billion in the first half of 2026, already beating every previous full-year total tracked by Space Capital [52]. Finnish radar-imaging operator Iceye led the quarter after raising a $1.2 billion round to expand production against NATO backlogs [52].

Governments are pushing in the same direction. The US Space Force more than tripled the ceiling on its National Security Space Launch Phase 3 Lane 1 contract — from $5.6 billion to $17 billion — in a July 17 notice, ahead of a sharp rise in military satellite missions [16]. Lane 1 is the commercial-style lane; seven providers compete for individual task orders, including newer entrants like Stoke Space, Impulse Space and Relativity alongside SpaceX, ULA, Blue Origin and Rocket Lab [16]. Separately, SpaceX launched 21 satellites for the Space Development Agency’s military data network [14], and L3Harris and Sierra Space were tapped to build 36 more for US missile tracking [33]. European defence spending is lifting that continent’s space economy on the same logic [42].

Also worth noting: navigation satellites are moving back down to low orbit. Xona Space Systems plans 258 spacecraft offering signals roughly 100 times stronger than GPS — better in cities, under trees, indoors, and harder to jam at a moment when GPS interference is spreading across aviation and shipping [25]. First six production satellites are due to launch in October [25].

Looking outward

Chile’s Vera C. Rubin Observatory began its decade-long survey — repeatedly photographing the whole southern sky to catch anything that moves or changes [75]. It is the closest thing astronomy has to filming the universe rather than photographing it.

Astronomers found the first of an estimated 10,000 “missing” black holes in Omega Centauri, a dense cluster of stars. They spotted it by watching a star orbit something massive and invisible, using twenty years of Hubble data from 2003 to 2023, then JWST to refine the measurement [17].

Closer to home, a JPL study concluded that a puzzling near-Earth object catalogued as an asteroid is actually a comet, identified by tracking its motion precisely and imaging it with observatories built for faint targets [4]. NASA’s Perseverance rover, meanwhile, has been reading the record of ancient impacts written into Martian rock [50], and the Psyche spacecraft returned data and time-lapse video from its May 15 Mars flyby — a gravity assist that both sped it up and tilted its path toward the metal-rich asteroid it reaches in 2029 [78].

The under-covered one: how you would find a nuclear weapon in orbit

The 1967 Outer Space Treaty forbids nuclear weapons in orbit, and 118 UN member states have ratified it [71]. There has never been a way to check. Today’s satellites simply cannot detect a warhead in space [71].

A study published in Nature Astronomy last week — from Areg Danagoulian, a nuclear physicist at MIT, funded in part by the US National Nuclear Security Administration — proposes the first peer-reviewed technique for spotting one [71]. It would use radiation trapped in Earth’s inner magnetosphere, the region where the planet’s magnetic field holds charged particles, as a probe for fissionable material [71]. This matters because in 2024 US intelligence officials alleged a Russian radar satellite parked in a radiation-heavy orbit was a testbed for an orbital nuclear anti-satellite weapon [71].

A proposal is not a deployed capability. But a treaty nobody can verify is a promise on paper, and this is the first serious sketch of how to check it.

02 · Lesson · why it matters

What gets absorbed never gets counted

A strain that something quietly handles produces no failures — and failure is the only evidence most systems accept that a problem exists.

Two true numbers that point opposite ways

Here are the two facts about low Earth orbit this week. Each working Starlink satellite swerved more than 40 times last year to avoid hitting something. And there were no collisions.

Read the second number alone and orbit looks fine. Read them together and you see what is actually happening: a rising load, met by a machine that handles it silently, and a safety record produced entirely by that machine’s success.

The dodging is not a sign that the situation is under control. The dodging is the situation. It is the work being done to keep the crowding from turning into an event.

The absorber

Call the thing in the middle an absorber. It sits between a growing pressure and the failure that pressure would otherwise cause, and it converts one into the other’s absence.

Starlink’s collision-avoidance software is a good one. It watches for a close approach, calculates the odds, and fires a small burn whenever the chance of a hit crosses about three in ten million. No human has to notice. No alarm reaches anyone outside the company. The satellite moves, the moment passes, and nothing happens.

Nothing happening is exactly the problem. An absorber does not just prevent the failure. It prevents the record of the failure. Every swerve that keeps two satellites apart also erases the evidence that they were ever close.

Institutions run on incidents

Almost every system for deciding what to fix runs on things that went wrong. Accident reports. Outage post-mortems. Injury statistics. Complaints. This is not stupidity — it is how you filter real problems from imagined ones. If nobody got hurt, the case for spending money is weak.

But that filter has a blind spot shaped exactly like an absorber. A pressure that is being successfully absorbed generates no incidents, so it generates no evidence, so it never wins an argument for attention. The better the absorber works, the more invisible the load becomes, and the less anyone can justify acting on it.

So the absorber buys time and spends the reason to use it in the same motion.

Meanwhile the load keeps climbing. And it does not climb gently. Tommaso Sgobba, of the association that works on space safety, names the arithmetic: pack more satellites into one shell of orbit and you do not add risk one satellite at a time. You add pairs. Double the satellites and you roughly quadruple the pairs that must be watched. The population grows in a line; the thing being managed grows in a curve.

The rule that was never written

Now look at the shape underneath. The absorber’s number — how many swerves a constellation will need each year — is not secret. It is calculable before launch. An operator can work it out from the size of the constellation and the density of the shell.

Nobody is required to hand it over. There is no rule saying a company must tell a regulator, before it is licensed, how many avoidance manoeuvres its satellites will make, or whether they carry the fuel and the automation to make them all.

That absence looks like a natural fact — nobody decided orbit would work this way — but it is a choice with a shape. It means the only party that knows the true load is the party doing the absorbing. It means orbital crowding arrives at regulators as headlines about near misses, after the fact, instead of as a number, in advance.

And it means whoever fills an altitude first shapes what everyone after them can do. The safest arrangement is for constellations not to overlap. But an orbital shell claimed early is a shell others must plan around. That is a real advantage, and it accrues to speed.

None of that makes the absorber a villain. The software is genuinely excellent, and every one of those swerves is a collision that did not happen. An arrangement can serve whoever built it and protect everyone underneath it at the same time. Both things are true here.

You are living downstream of several of these

The shells being filled are not empty sky. They carry weather data, ship and aircraft tracking, broadband to places cable never reached, and now navigation signals moving down from higher orbits because they work better closer in. A cascade of debris in a busy shell would not read as a space story. It would read as a forecast that got worse and a phone that got less sure where you are.

You are living inside smaller absorbers too, and you already know how they feel from the inside. The colleague who quietly covers the gap left by an unfilled post — so the post stays unfilled. The old bridge that gets patched each year, and passes each inspection, and therefore never rises up the list. The body that keeps going because you push through, right up until it does not.

In every case the same thing happens. Someone or something takes the strain, the strain stops being visible, and the absence of trouble is read as proof there is none.

From the inside, coping looks like health

The hardest part is that nobody in the system is being foolish. The regulator waiting for evidence is being rigorous. The operator running the software is preventing collisions. The person covering the gap is being decent.

Each seat has a clear view of its own part and no view of the total. What the system produces — a clean safety record — is the one signal every seat can see, and it is the signal least able to tell them what is coming.

The number exists. It has always existed. It just has to be asked for while there is still nothing to report.

03 · Lab · your turn

The Clean Record

Rehearse deciding year after year on evidence that only records failures, while the real load stays uncounted.

04 · Hope · carry this

Every one of those forty swerves a year is a collision that didn't happen, done quietly by people nobody will ever think to thank. And the number that would show how long it can hold was never hidden — only never asked for, and asking is a cheap thing to start doing.