DevLazar
Note
084
Series
The Wire · Infra
Date
Nov 15, 2026
Read
7 min · 1,394 words
Rev
1
Drawn by
L. Stanković · Niš
Lab notes / The Wire

Starship Flight 15: The Ship Comes Home From Orbit, the Booster Does Not

SpaceX caught a Starship upper stage returning from orbit on November 14, a first, after it deployed 26 Starlink V3 satellites. The booster was lost. What it means for bandwidth supply and the cost of orbit.

Lazar StankovićEngineer, DevLazar · Niš, Serbia

SpaceX launched Starship's fifteenth flight from Starbase, Texas, at 4:47 p.m. CST on Friday, November 13 (22:47 UTC), put Ship 42 into orbit, released 26 Starlink V3 satellites, and on Saturday caught the ship with the launch tower's arms after roughly 22 and a half hours aloft. By Teslarati's account, nobody had brought an orbital upper stage back to its own pad and caught it before. The Super Heavy booster, meanwhile, aborted its catch and was destroyed over the Gulf of Mexico. Half a reusable rocket came home, and for once it was the harder half.

What was new, in order

This was not Starship's first orbit. Flight 14 on September 28 deployed the first 26 V3 Starlinks, but an engine that cut out early shrank a planned 10-hour flight to about three. Flight 15 repeated the payload and attempted everything else: the full coast, the return to launch site, and the first catch of a V3 booster.

The stack was a V3 Super Heavy on 33 Raptor 3 engines, good for more than 18 million pounds of thrust. The ship reached low Earth orbit at roughly 275 km about 26 minutes after liftoff, and the satellites went out over 42 minutes. Three carried cameras to photograph the ship's heat shield, an inspection rig you can only build when the payload is your own.

The pictures mattered. SpaceX said they showed the ship "in good shape for reentry," so controllers passed on a contingency splashdown. After about 14 orbits the ship was caught on Saturday, November 14, a first that Space.com called "a landmark achievement."

Booster 22 was due back at the tower about 6.5 minutes after launch. Instead it aborted the catch automatically and diverted offshore, where it exploded before reaching the water. SpaceX's Dan Huot first blamed guidance, then corrected himself: "It was not a GNC issue that caused us not to come back." Musk, via Teslarati, blames slightly underperforming engines that left it short of propellant, promises steadier Raptor V3.2 engines and an earlier engine cutoff, and says the timeline is unchanged. SpaceX has recovered three V1 boosters; no V3 has come back yet.

Ten Falcon 9s in one payload bay

SpaceX says the 26-satellite batch carries capacity equal to 10 Starlink V2 launches on Falcon 9, the same comparison ABI Research drew after Flight 14. The per-satellite specs explain it. Starlink rates each V3 at 1 Tbps of downlink and 160 Gbps of uplink, roughly 10 and 22 times a V2, plus six 400 Gbps laser links. Twenty-six of them is about 26 Tbps of nameplate downlink, by ABI's arithmetic on SpaceX's rating.

A Falcon 9 carried V2 Minis 20 or 30 at a time. The satellite count per launch barely moved; the bandwidth per launch rose an order of magnitude. Starlink has said a Starship launch should eventually add about 20 times a Falcon 9's worth, and Musk wants batches of 60 once flights are routine.

Starlink capacity per launch in Falcon 9 V2 launch equivalents: Falcon 9 1x, Starship Flight 15 10x, planned Starship launch 20x
Chart: Starlink capacity added per launch, in Falcon 9 V2 launch equivalents. Source: Space.com; Converge Digest; ABI Research.

Line rate is not throughput

Read the terabits the way you read a NIC's line rate. ABI models the constellation at 4.42 Pbps nameplate but 942 Tbps usable by end of 2027, about 21.3%, and even that, by ABI's own account, "rather optimistically" assumes the equivalent of one Starship launch a week. It also puts Starlink above 91% of satellite broadband by end of 2026, with more than 15.6 million subscribers.

Latency is the quiet gap: nothing we read gave a V3 latency figure, so laser routes beating fibre remain a claim until someone publishes measurements.

Mass to orbit has no public per-kilogram price either. What a caught ship changes is the denominator: hardware flown again rather than rebuilt. Gizmodo names reflying a used upper stage as the next hurdle. Musk targets a weekly cadence in the first half of 2027 and 10,000 launches a year by 2030, against a development bill of more than $15 billion. In September he called Starship "2 to 3 years away from hourly flights." Targets are targets.

Rivals, regulators and the Moon

Amazon Leo's first Vulcan mission, LV-01, carries 40 satellites. Amazon's page, updated November 14 with an adjusted launch date, now targets November 15; Vulcan is to fly more than 1,500 first-generation Leo satellites over 38 missions, with service promised "later this year." Forty satellites against twenty-six is the wrong comparison, and Amazon's per-satellite capacity is not in anything we read.

On Friday morning SpaceX was still awaiting an FAA licence change for the ship's return over land. The return happened; we have not seen the FAA's document.

SpaceX holds $2.9 billion in NASA contracts for Starship lunar landers, and an Earth-orbit demonstration with Orion and Artemis III astronauts is slated for summer 2027. A lunar ship needs several tanker flights to refill in orbit, Teslarati notes, so fast reuse is a schedule item. So is compute: Musk talks of a 100,000-satellite constellation and an orbital AI data network ten times larger. All of it runs through the same tower arms.

What we could not verify

  • SpaceX's flight page and any FAA statement would not load for us; the timeline rests on the coverage linked above. An official account could move details, not the outcome.
  • The capacity figure. Space.com renders SpaceX's statement as "a terabit" for the batch; Starlink rates each V3 at 1 Tbps. Only the per-satellite reading fits the ten-launch comparison; otherwise that comparison is off by an order of magnitude.
  • Capacity numbers are company-rated nameplate, ABI's usable share is a model, and the 20x figure does not say how many V3s per launch it assumes.
  • The booster's end and cause. Spaceflight Now saw it explode before reaching the water; Space.com describes a splashdown. The cause is preliminary, and Musk's account reached us through Teslarati alone.
  • Amazon's date conflicts on its own page (body November 15, metadata November 14).

The 30,000-foot read

Strip out the fireworks and Flight 15 is a story about denominators. The satellites carried ten launches' worth of capacity in one, and the ship that carried them came back to be used again. Those two numbers set the price of a bit delivered from orbit, and both moved on the same weekend. The booster, a stage SpaceX recovered three times in its older form, is the part that failed; that reads as tuning, not physics.

What no launch controls is the gap between nameplate and delivered. A terabit over the Pacific serves no one, and the constellation still has to prove its latency and usable capacity in public. Watch three things: a reflown ship, a caught V3 booster, and a 60-satellite manifest. When all three happen on one flight, the economics of satellite bandwidth stop being a forecast.

Sources: Spaceflight Now (launch time, Booster 22 and Ship 42, booster abort and explosion, Huot correction); Space.com (V3 booster and Raptor 3 thrust, orbit insertion, deployment timing, heat shield cameras, 10-launch capacity comparison, booster recovery history, Artemis III demo, 10,000 launches a year, constellation and AI network sizes); Teslarati (ship catch on November 14, 22.5 hours and 14 orbits, heat shield quote, Musk's booster explanation and fixes, FAA licence change, weekly cadence target, tanker flights); Space.com, catch report (ship catch as a first, "a landmark achievement"); Aerospace America (Flight 14 details, $2.9 billion NASA contracts); Reuters via Devdiscourse (Falcon 9 V2 Mini batch sizes, 60-satellite goal, $15 billion development cost, hourly-flights quote); Converge Digest (Starlink-stated V3 specifications, 20x per-launch claim); ABI Research (26 Tbps arithmetic, nameplate versus usable model, market share and subscribers); Amazon (LV-01 satellites, date, Vulcan manifest, service rollout). CAVEATS: SpaceX's Flight 15 page and FAA materials could not be opened, so the flight account relies on secondary coverage; the catch rests on Teslarati and Space.com, since a CNBC report did not load for us; all V3 capacity figures are company-stated nameplate ratings, and the 26 Tbps total is ABI's arithmetic on them; Space.com's "a terabit" batch figure conflicts with the per-satellite rating, and Aerospace America quoted SpaceX's comparison in different words again; Musk's booster explanation is single-sourced; ABI's usable-capacity and market-share figures are analyst estimates; no image marker is included because no official SpaceX or FAA page could be viewed in this session. Correction to early reports: Flight 15 was not Starship's first orbital flight or first V3 deployment (Flight 14 was, on September 28), and the booster catch was aborted in flight by the vehicle rather than skipped beforehand.

Lazar Stanković
Engineer and writer, running a one-man shop in Niš, Serbia. I build systems that survive their launch and write down what the work teaches.
Related notes
Note 081 · The Wire · 7 min
SpaceX Plans $74 Billion of AI Campuses Just as Texas and Arizona Turn Wary
SpaceX says it will spend more than $74 billion on AI supercomputing campuses at Starbase and in Cochise County, Arizona. The money is announced; the permits, the power and the politics are not settled.
Note 079 · The Wire · 7 min
'Don't Ask Again' Remembers Your Password: Claude Code Approvals Are Leaking Secrets to GitHub
Apiiro found 1,379 secrets in public copies of Claude Code's local settings file, written there by the 'don't ask again' button. How it happens, who counted, what Anthropic has said, and how to clean up.
Note 077 · The Wire · 6 min
Microsoft 365's Outage Passes 15 Hours, and the Health Checks Are the Story
A Microsoft 365 outage that began November 12 ran past 15 hours. Microsoft cites servers failing health checks and has rolled back a network change, with no sustained fix yet. The lessons are about health checks and fallbacks.
← Note 083 · Nvidia Became the AI Boom's Financier. Now Its CFO Holds the Checkbook
The Wire · Weekly · Free
Notes like this one, every week.

AI infrastructure, shipping on a deadline, and the craft of building things that last. No noise.