Starlink runs the two highest-volume manufacturing operations in the space industry — roughly 70 satellites a week in Redmond, Washington, and about 15,000 user terminals a day in Bastrop, Texas. Both figures came out of SpaceX’s IPO filing, which turned years of estimates into disclosure. And the real constraint on the whole chain is not a factory at all. It is a rocket.
👉 The insight that reframes this topic: Starlink is not a space company’s side project with a supply chain problem. It is a consumer electronics manufacturer that happens to put its products in orbit. Judge it against Foxconn and Dell, not against Boeing and Airbus.
- Satellites: about 70 per week from Redmond — roughly 3,640 a year — confirmed in the S-1 filing. Every Starlink satellite, including V3, is built there.
- Terminals: around 15,000 a day at Bastrop, Texas, on a site exceeding one million square feet that houses one of the largest printed circuit board operations in the US.
- Vertical integration: extreme. Phased arrays, satellite buses, thrusters, laser links and terminals are largely in-house; semiconductors and raw materials are not.
- Cost curve: terminal production costs reportedly halved through automation and design simplification at Bastrop.
- The real bottleneck: Starship. The V3 satellite is too large for Falcon 9, so satellite production capacity is ahead of the ability to launch it.
What the IPO Filing Changed
For years, Starlink production figures were estimates. SpaceX’s S-1 filing put hard numbers on it: the Redmond facility produced an average of approximately 70 Starlink satellites per week from December 2025 to April 2026, or about 3,640 per year at full rate (GeekWire on the S-1 production disclosure).
For scale, Amazon’s rival constellation was described the same week as producing "tens of satellites a week" from its Kirkland facility, up from one a month a year earlier. The gap is roughly an order of magnitude, and Redmond is credibly one of the highest-volume spacecraft factories in history (Redmond facility profile).
The constellation itself passed 12,178 satellites launched and 9,213 operational as of 30 May 2026, with reporting in 2026 putting the network near 10,200 satellites and more than 10 million customers. Total network capacity is approaching 450 Tbps with median peak latency around 25.7 ms for US customers (DataCenterDynamics on network capacity).
💡 Source-quality note: satellite counts vary between 9,200 and 10,200 depending on whether the figure is operational, launched, or includes recently deorbited units, and on the date. Any article giving one exact number without specifying which measure it uses is not being careful.
The Two Factories That Are the Supply Chain
| Redmond, Washington | Bastrop, Texas | |
|---|---|---|
| Product | Starlink satellites (all generations, incl. V3) | User terminals and kits |
| Output | ~70 satellites/week (~3,640/year) | ~15,000 terminals/day (~4.7M/year) |
| Footprint | Expanding with 124,900+ sq ft at Redmond Ridge | Over 1 million sq ft; 1,000+ staff |
| Scope | Design, manufacture, integration, test, network ops | PCB fabrication, phased arrays, assembly |
| Stated target | 5,000–10,000 V3 satellites per year | Double output by end of 2026; 10M kits annually |
Redmond: the satellite line
SpaceX states plainly that every Starlink satellite, including V3, is built in Redmond (Starlink V3 technical page). The V3 manufacturing detail worth knowing is the solar array: SpaceX produces a continuous roll of solar blanket, cuts it into 19-metre segments and stitches four together per array, generating roughly twice the power of a V2 array while being optimized to reduce drag at low altitude.
That is a roll-to-roll process description, not a spacecraft assembly description. It is the clearest signal in any SpaceX document of how differently this factory thinks about production.
Bastrop: the terminal line
The Bastrop site reached roughly 15,000 terminals per day by mid-2025, equating to about 4.68 million units annually, with weekly output exceeding 70,000 kits. Expansion plans include doubling kit capacity in 2026 and adding a million square feet over three years, supported by a $17.3 million state grant (Bastrop factory profile).
SpaceX board member and early investor Steve Jurvetson described the line as producing 4.7 million terminals a year and noted that one subsection is the largest printed circuit board factory in America (Jurvetson on the terminal line).
✔ The PCB detail is the most underrated fact in this entire supply chain. SpaceX did not outsource circuit board fabrication to Asia like every consumer electronics firm does. It built domestic PCB capacity at national scale — which is why terminal cost reduction was possible in the first place, and why the terminal chain is unusually tariff-resistant.
The Terminal Cost Curve
Terminals were once the business’s biggest financial problem: early units reportedly cost SpaceX $1,000–$2,000 more than the $499 charged to customers. Production costs have since been cut roughly in half through high-volume automation, robotics, simplified phased-array design and in-house production (analysis of terminal cost reduction).
The strategic choice that followed is instructive: SpaceX did not pass the savings to consumers. It reinvested them into production scale and geographic expansion. In supply-chain terms, cost reduction was converted into volume rather than margin — the classic decision of a firm that believes it is in a land-grab.
- Design simplification. Electronically steered phased arrays eliminate moving parts entirely, which removes the mechanical failure modes and the assembly labour that come with them.
- Automation. High-volume robotic assembly at a scale unheard of in aerospace, closer to smartphone manufacturing than satellite ground equipment.
- In-house PCB. Vertical integration into the single highest-value component of the terminal.
- Volume itself. At 4.7 million units a year, per-unit component pricing improves independently of any engineering change.
What Goes Into a Starlink Satellite
| Subsystem | Detail | Sourcing |
|---|---|---|
| Phased array antennas | Ku-, Ka- and E-band | In-house design; RF components external |
| RF amplifiers | GaN amplifier modules | External — Filtronic among reported suppliers |
| Propulsion | Hall-effect thrusters, krypton or argon fuelled | In-house thruster; noble gas externally refined |
| Laser inter-satellite links | Optical terminals for on-orbit mesh routing | In-house |
| Solar arrays | Continuous-roll blanket, 19 m segments stitched in fours (V3) | In-house production technique |
| Avionics and processors | Radiation-tolerant compute | External semiconductors |
| Star trackers | Attitude determination | External — reported concentration in few suppliers |
| Structure | Flat-panel bus optimized for stacked launch | In-house |
Note the propulsion line. Hall-effect thrusters run on noble gases refined by a small number of producers worldwide — a genuine single-point dependency that scales linearly with constellation size. Each generation also carries laser terminals enabling ocean-crossing traffic to relay satellite-to-satellite until one sees a gateway (Starlink satellite subsystem reference).
⚠ The chokepoint most analyses miss: noble gas supply. Krypton and argon for Hall thrusters come from a handful of global refiners, and demand rises with every satellite launched. It is the same structural risk as xenon in the wider space industry — low-profile, hard to substitute, and invisible until it binds.
The Real Bottleneck: Launch, Not Manufacturing
Here is the tension at the centre of Starlink in 2026. Redmond can build roughly 3,640 satellites a year and is tooling toward 5,000–10,000 V3 units annually. But V3 is designed for Starship, not Falcon 9 — it is too large and heavy for the workhorse rocket that has carried the constellation since 2019.
SpaceX states that V3 enables roughly 20 times the capacity per Starship launch compared with a Falcon 9 launch of V2 satellites. That is a transformative ratio — and entirely contingent on Starship reaching operational cadence. With Starship’s development record including multiple launch failures, commentators have questioned whether the vehicle will be market-ready on the timeline the satellite factory implies (DataCenterDynamics on the Starship dependency).
| Constraint | Current state | Assessment |
|---|---|---|
| Satellite production | ~70/week, scaling toward 5,000–10,000 V3/year | Not binding — capacity ahead of demand |
| Terminal production | ~15,000/day, doubling planned | Not binding — reported surplus |
| Launch capacity | Falcon 9 proven; Starship still maturing | Binding — gates V3 deployment entirely |
| Spectrum and regulatory | Market-by-market licensing | Slow but tractable |
| Noble gas supply | Few global refiners | Watch item, scales with fleet |
👉 Read the surplus correctly. Reporting describes a production surplus shifting Starlink’s focus from scarcity to market share. That is true on the ground and misleading in orbit: terminals are genuinely abundant, but V3 satellites cannot reach orbit in volume until Starship does. One surplus is real, the other is stranded inventory waiting on a rocket.
How Starlink’s Chain Compares
Starlink’s S-1 named Amazon’s planned Leo constellation and Blue Origin’s TeraWave among its rivals, alongside Eutelsat OneWeb, Telesat Lightspeed and AST SpaceMobile (competitor list from the S-1). The supply-chain contrast is the whole story.
- Production rate: roughly 70 satellites a week versus tens a week for the nearest competitor.
- Launch access: SpaceX owns its rockets. Every competitor buys launch — often from SpaceX.
- Terminal manufacturing: millions of units a year in-house, including domestic PCB fabrication.
- The vulnerability: the same vertical integration means a Starship delay has no external workaround. Competitors can switch launch providers; SpaceX cannot switch to itself faster.
Checklist: How to Read Starlink Numbers
- Check whether a satellite count is launched, operational, or in-orbit-including-deorbiting. The three differ by hundreds.
- Distinguish terminal production from terminal shipment. Producing 15,000 a day is not the same as activating them.
- Separate V2 from V3 claims. Capacity-per-launch figures for V3 assume Starship, not Falcon 9.
- Treat S-1 figures as the highest-confidence source available. Everything published before May 2026 was an estimate.
- Watch launch cadence, not factory output. Manufacturing has not been the constraint since 2024.
FAQ
How many Starlink satellites does SpaceX build per week?
About 70, per the company’s IPO filing covering December 2025 to April 2026 — roughly 3,640 a year at full rate. All are built in Redmond, Washington.
Where are Starlink terminals made?
Bastrop, Texas, at a facility exceeding one million square feet producing around 15,000 terminals a day, with plans to double capacity by the end of 2026.
Who supplies Starlink components?
SpaceX builds phased arrays, thrusters, laser links and satellite buses in-house. External suppliers cover semiconductors, RF and GaN amplifier modules, star trackers, connectors and refined noble gases.
How much does a Starlink terminal cost to make?
Not officially disclosed. Early units reportedly cost $1,000–$2,000 more than the $499 retail price; production costs have since been cut roughly in half through automation and design simplification.
What is actually limiting Starlink growth?
Launch capacity. V3 satellites require Starship, which is still maturing. Satellite and terminal factories are both running ahead of what can be deployed.
How does Starlink compare to Amazon’s constellation?
On production rate, roughly an order of magnitude ahead — about 70 satellites a week versus tens a week — plus SpaceX owns its launch vehicles while competitors purchase launch services.
The Bottom Line
Starlink solved manufacturing. Two factories, one turning out spacecraft at automotive cadence and one turning out consumer electronics at smartphone cadence, with domestic PCB capacity and halved terminal costs behind them.
What it has not solved is the ride. The V3 satellite that justifies the Redmond expansion cannot fly on the rocket that built the constellation, and the rocket that can fly it is still proving itself. The supply chain is ahead of the launch chain, and only one of those can be fixed with a bigger factory.
The number to watch through the rest of 2026 is not satellites built but V3 satellites delivered to orbit. Track the parent picture in our SpaceX supply chain analysis and the compute side in our xAI supply chain breakdown.
💡 We track the whole Musk hardware ecosystem — SpaceX, Starlink, Tesla, SpaceXAI, Neuralink, The Boring Company — with confirmed and reported figures kept in separate columns. Join the community to get each breakdown as it lands.
Sources: SpaceX S-1 filing via GeekWire, Starlink.com, DataCenterDynamics, Grokipedia, Benzinga, SpaceXStock, Wikipedia launch records. Satellite counts vary by measure and date and are reported as such. Last fact-check: August 5, 2026.
NEVER MISS AN OPTIMUS UPDATE
We track every Tesla Optimus development — specs, deployment milestones, pricing and competitive moves — updated as news breaks.