In December 2022, Amazon Leo's director of production took an empty building in Kirkland, Washington, and started turning it into a satellite factory. Paul Palcisco still remembers the shape of the job: 170,000 square feet, fifteen months of conversion, and a first qualification spacecraft that proved the floor plan wrong almost immediately.
"As soon as we opened the doors in that factory, we needed to change it," Palcisco told Ars Technica last week.
On 8 October, that line produced its thousandth satellite. Amazon Leo is now the second-largest satellite manufacturer on Earth, behind SpaceX. It has 396 spacecraft in orbit, several hundred more stacked in Florida waiting on rockets that are not flying, and a broadband service it says is weeks from switching on.
All three numbers are true at the same time. The factory Amazon built to escape a rocket shortage has produced a satellite inventory it cannot launch, and that backlog has already cost the company something measurable at the FCC.
A thousand built, 396 in orbit
Amazon Leo began full-scale deployment in April 2025 on the back of an order for more than 80 launches, the largest commercial procurement of launch capacity ever placed. Fourteen of those flights are done. They carried 396 satellites, and only four have been deorbited. The first generation is planned at 3,232 spacecraft.
The first 27 off the Kirkland line took six months. The same line now turns out 27 in a week.
Production total since April 2025
The Kirkland line assembled 27 spacecraft in its first six months and now runs at roughly 27 a week, against a stated capacity of 30 a week. · Amazon Leo, 2026
Amazon built the factory alongside the satellite rather than after it, which is why the ramp went the way it did. Engineers and floor workers were in the same building from the start, iterating on test procedures that had taken months per spacecraft in the pre-Kuiper lab. Testing time fell from months to days. Amazon has stated a plant capacity of 30 satellites a week.
Satellite production has become a business worth naming. As we wrote in September, [Open Cosmos' €300M round treated satellite manufacturing as an asset class](/open-cosmos-300m-satellite-manufacturing-2026/) — and for a few companies it now is one. Amazon has told the FCC it has committed more than $10bn to putting these spacecraft in orbit.
None of this was in doubt a week ago. What is in doubt is how much of it can reach orbit.
First-generation deployment
Delivered across 14 flights; four spacecraft deorbited so far. Amazon says the count is enough to support initial service. · Amazon Leo, October 2026
Amazon Leo can build spacecraft faster than it has ever needed to. Its problem is that every heavy-lift vehicle on its manifest except one was grounded when the satellites became ready.
The FCC has already priced that gap. It waived the missed deployment deadline and took spectrum priority back from Amazon as the price.
The rockets were never the plan
In 2022, when Amazon placed its first large launch order, it deliberately kept away from SpaceX. Sixty-eight flights went to United Launch Alliance, Arianespace and Blue Origin, built to give the company heavy-lift options that did not belong to its main competitor. The vehicles those contracts named — Vulcan Centaur, Ariane 6 and New Glenn — had not flown once between them. Only ULA's Atlas V had any flight history, and Atlas V is retiring with one mission left for Amazon.
Amazon says the Leo manifest now exceeds 100 flights. The disclosed book includes 38 Vulcan, 18 Ariane 6, 24 firm New Glenn launches, nine Atlas V, and 10 Falcon 9 bought in late 2025 once the gap became obvious. Four providers. No dependency, in theory.
| Vehicle | Provider | Leo payload | Status, October 2026 |
|---|---|---|---|
| Ariane 6 | Arianespace | Three dozen | ✔ Flying |
| Vulcan Centaur | ULA | 40–45 | ◐ Grounded on recurring strap-on booster faults; return to flight targeted no earlier than 29 October |
| New Glenn | Blue Origin | 48 | ✗ Grounded since a May pad anomaly traced to the BE-4 engine; return to flight targeted for December |
| Atlas V | ULA | 29 | ✗ One mission remaining before retirement |
| Falcon 9 | SpaceX | Not disclosed | ◐ Reserve capacity bought in late 2025 |
Leo satellites per mission, and vehicle availability, as reported in October 2026.
By June 2026, only Arianespace had flown a Leo payload. Ars Technica reported Amazon executives describing hundreds of finished satellites sitting in a Florida processing facility, built and idle. The company had assumed rockets would arrive in step with the production line. Production won the race by a wide margin.
What followed was a scramble to unblock capacity that did not belong to Amazon, and it turned up a coupling nobody had priced. Blue Origin's New Glenn lost a vehicle on the pad in May. Multiple sources pointed at the BE-4 engine, which powers the New Glenn first stage and also powers Vulcan's first stage. One anomaly was therefore sitting on two of Amazon's four launch paths at once.
Vulcan's grounding is a separate and longer-running story. ULA has been chasing a recurring fault in the strap-on solid rocket boosters since spring, and Badyal told Ars that Vulcan's blocking nozzle issue has since been fixed. Two Vulcans now stand vertical in two separate vehicle integration facilities at Cape Canaveral, and ULA is targeting no earlier than 29 October for the first flight carrying Amazon Leo satellites. If that date holds, a second Amazon Leo mission follows before the year ends.
We started off with ULA, Blue Origin, Arianespace, and SpaceX. We signed deals with all four of them. We knew there was risk, but we had to get heavy lift vehicles. Without that kind of plan, we probably would be in a far worse situation.— Rajeev Badyal, program lead, Amazon Leo
Amazon has been buying its way around the bottleneck with money aimed at somebody else's pads. The company has put more than $200M into ULA's Cape Canaveral ground infrastructure: a dedicated vertical integration facility, a rail system and second mobile launch platform, plus a second ship to move boosters from Alabama. Badyal said the same infrastructure now lets ULA process two rockets at once.
The diversification strategy has one uncomfortable wrinkle. Amazon founder Jeff Bezos owns Blue Origin, which is both a contracted launch provider for Leo and a competitor building its own TeraWave communications constellation. Bloomberg reported that Blue Origin hired at least 47 former Amazon Leo engineers over the past year, many of them now working on TeraWave. The company Amazon hired to diversify its launch supply is competing for its staff.
The deadline Amazon already missed
The FCC authorised Amazon Leo in July 2020 with two conditions. Half of the 3,232 first-generation satellites had to reach orbit by 30 July 2026, and the rest by July 2029. In January 2026 Amazon asked the regulator for a two-year extension, citing a shortage of launch vehicles. It told the FCC it expected roughly 700 satellites aloft by the deadline. The number that arrived was close to 400.
The FCC granted the waiver. It also stripped spectrum priority in the Ka and Ku bands from everything Amazon launches after that date, meaning those spacecraft must be engineered to avoid interfering with Starlink — the incumbent they exist to displace. The regulator said priority returns from October 2027 if Amazon can demonstrate that the hardware is built and the launch manifests are booked.
That is why the thousandth satellite matters more than a milestone. The 1,000-unit production total and two vertical Vulcans are the argument, made in public, that Amazon has done the part it controls.
Delta, and the ceiling on the antenna
Delta Air Lines picked Amazon Leo over Starlink for in-flight connectivity, with installations on 500 aircraft beginning in 2028. The array is electronically steered, with no moving parts, rated for up to 1 Gbps down and 400 Mbps up, and installs in about a day.
Those figures are a ceiling on the aircraft, not a guarantee to each passenger. Every beam it passes through is shared with other aircraft on the same route, and a full cabin working at once divides the same capacity. Delta's brand is the collateral Amazon is deploying against that arithmetic.
Rajeev Badyal, who runs the programme, led Starlink in its early years before Elon Musk fired him roughly eight years ago for moving too cautiously. He is now the person deciding how quickly the second broadband-from-space operator reaches its first paying month.
Whether the first Vulcan flight carrying Leo satellites actually flies on or near 29 October, and whether the second one happens at all.
Whether the satellites in orbit count moves off 396 for the first time since July.
Whether the FCC treats the public evidence of built hardware and booked manifests as sufficient when the October 2027 review lands.
Whether Blue Origin's New Glenn returns in December, since that vehicle's BE-4 engine also sits under Vulcan.
Probability: 70% — the hardware exists, the airspace is allocated, and Chris Weber, the company's vice president of business, says the plan is initial service this year. The binding constraint is whether Vulcan actually flies in October.
✔ Arguments for it holding
Amazon has already put more than $10bn into the system and $200M into someone else's launch infrastructure.
Confirmation criteria: a Leo payload flying on Vulcan by early November, with the second mission dated before the year closes.
✗ Arguments against
The strap-on booster fault is described as recurring, which is a different problem from a single anomaly.
Satellites launched without spectrum priority must avoid interfering with Starlink, so the newest spacecraft are the ones operating under the tightest constraint.
Refutation criteria: a second Vulcan scrub, or 29 October passing without a new date.