The satellite internet story usually gets told in orbital mechanics and bandwidth figures. But the race to put thousands of broadband satellites into low Earth orbit is, at its core, a manufacturing problem. And right now, that manufacturing problem has a specific, underappreciated chokepoint: the RF semiconductor supply chain is positioned for 2027, while several operators have made 2026 commitments they need it to support today.

The Scale of What’s Being Built

Start with the numbers. SpaceX’s IPO filing in May 2026 revealed that its Redmond, Washington facility is producing roughly 70 Starlink satellites per week — approximately 3,640 per year at that rate. That is an industrial cadence without precedent in the history of spacecraft manufacturing. Traditional geostationary satellite programs measured output in single digits per year, per contractor, with bespoke components and multi-year build cycles. SpaceX turned satellite production into something closer to automotive assembly.

Amazon is attempting a similar ramp for its constellation (rebranded from Project Kuiper to Amazon Leo in late 2025). Its Kirkland, Washington factory is designed for up to five satellites per day at peak capacity, targeting a 3,232-satellite constellation. A $139.5 million investment in a Florida payload processing facility — capable of handling more than 100 satellites per month and running three simultaneous launch campaigns — signals that Amazon is taking the logistics of constellation-scale production seriously.

These are real industrial commitments. But they depend on a supply base that is still catching up.

What Goes Inside a LEO Satellite

Every LEO broadband satellite contains, at its core, a stack of RF hardware: power amplifiers, low-noise amplifiers, frequency converters, phased-array antenna elements, and the baseband processing that ties them together. The physics of Ka-band and Ku-band transmission demand high-performance compound semiconductors — gallium nitride on silicon carbide (GaN-on-SiC) for power amplification, and GaAs for low-noise applications — rather than the commodity silicon that fills a smartphone.

GaN-on-SiC is not a mature commodity market. The dominant RF component suppliers — MACOM and Qorvo among them — are treating SATCOM volume as a fiscal 2027 story. MACOM demonstrated a Ka-band power amplifier as recently as March 2026, but its commercial revenue ramp is not positioned for this year. The SiC substrate situation compounds this: Wolfspeed, which dominates U.S. wafer supply, has been primarily targeting electric vehicle power electronics, and saw its revenue fall in early 2026 as it worked through factory utilization challenges.

The physics of the problem makes it structurally slow to fix. Compound semiconductor capacity additions require 12 to 24 months from capital expenditure commitment to first qualified wafer. You cannot order your way out of that timeline in a single quarter.

The Vertical Integration Divide

SpaceX largely sidesteps this problem through vertical integration. Its Redmond factory produces the phased-array antenna tiles that are among the most technically demanding components in each satellite, and it has reportedly been building terminal production capacity at a rate of around 90,000 dishes per week. When you control your own component supply, you absorb shortages internally rather than competing on the open market.

Amazon’s position is more nuanced. It developed a custom “Prometheus” baseband ASIC, reducing dependence on merchant silicon for signal processing. But the question of terminal volume ramp — the user equipment that subscribers actually install — remains open, and that depends partly on the same RF foundry capacity that everyone else is chasing.

The operators most exposed are those with neither the scale nor the integration to buffer against supply chain timing. AST SpaceMobile has approximately $1.2 billion in contracted commitments from mobile network operators and has built its gateway rollout plan around third-party RF foundries that are, by every public signal, volume-ready in 2027 rather than 2026. Eutelsat OneWeb faces similar exposure, with terminal supply dependent on Kymeta and Intellian, both of whom are targeting commercial volume in 2027.

Tariffs Add Another Layer

The RF component picture is further complicated by the current tariff environment. LEO satellite manufacturers have seen build costs rise by an estimated 8 to 15 percent due to tariffs on semiconductors, transceivers, and optical payloads — particularly those sourced from Asia. GaN and SiC electronics specifically face duties in the 10 to 17 percent range in the current 2026 structure. Ground terminal hardware — which includes RF amplifiers, waveguides, and antenna components with significant Asian supply chain exposure — is also affected.

For operators running thin margins on consumer-priced services, a 10 percent cost increase in the RF stack is not trivial. For those, like Starlink, that are already competing on price (monthly revenue per subscriber fell from roughly $99 in 2023 to $66 by Q1 2026, reflecting a more international subscriber mix), cost pressure at the component level cannot simply be passed through.

The industry response has been a shift toward domestic sourcing and vertical integration where possible — both of which take years to execute, not quarters.

What This Means for 2026 and Beyond

The most likely near-term consequence is quiet schedule slippage dressed up in other language. Operators who overcommitted on 2026 subscriber targets, relative to what the foundry supply chain can support, will attribute shortfalls to demand softness or regulatory friction rather than acknowledging a parts problem. That is the rational communications strategy, even if it obscures what is actually happening.

The deeper implication is structural. The constellation era was supposed to be defined by software-driven, commodity-hardware satellites built at automotive scale. That vision is real — SpaceX has largely achieved it. But “commodity” in this context still means specialized compound semiconductors, and that segment of the electronics industry was not designed for the volumes that multiple simultaneous megaconstellations require.

The operators who win the next phase of the LEO race will not just be those with the best orbital geometry or spectrum position. They will be those who secured component supply chains before the bottleneck became obvious — or who built enough vertical integration to not need the open market at all. SpaceX learned that lesson early. Everyone else is still learning it.

Sources