It is tempting to think of satellite internet as an orbital problem. Constellations must be designed, funded, built, launched, and licensed. Those are genuinely hard challenges — and they get most of the industry’s attention. But the single most consequential engineering problem in LEO broadband has never been in orbit. It has always been on the ground, mounted to a rooftop: the user terminal.

A phased-array flat-panel antenna is not a simple piece of hardware. Traditional satellite dishes use a mechanical gimbal — a motor that physically rotates the antenna to track the satellite as it moves across the sky. That works fine for geostationary satellites, which appear stationary relative to the ground. It fails completely for LEO. A Starlink satellite crossing the sky at roughly 27,000 kilometers per hour may only be above the horizon for a few minutes. The terminal must hand off seamlessly from one satellite to the next, continuously steering a beam electronically, without any moving parts.

That is what a phased array does. It is a grid of hundreds or thousands of individual antenna elements, each fed by its own radio-frequency chip. By adjusting the phase of the signal sent to each element by tiny fractions of a wavelength, the system can point the aggregate beam in any direction — and shift that direction almost instantaneously. The technology works, elegantly. Making it work at consumer price points, reliably, in the millions, is a different problem entirely.

How SpaceX Approached the Cost Wall

SpaceX’s founding insight for Starlink terminals was that the only path to affordable hardware was vertical integration and custom silicon. A phased array built from commercial-off-the-shelf RF chips — the kind used in radar or defense applications — would carry an unworkable bill of materials. SpaceX designed its own application-specific integrated circuits (ASICs) to bring RF front-end costs down, collapsing functions that previously required separate components onto a single chip.

That is a standard chip economics story: high up-front engineering cost, very low marginal cost at volume. But it required betting on volume before that volume existed. The chip shortage of 2020–2023 made this particularly acute — terminal production was constrained not by demand or launch cadence but by semiconductor supply. By early 2023, the company was reportedly producing thousands of user terminals per week alongside a manufacturing pace of six v2 mini satellites per day.

The subsidy question runs quietly through all of Starlink’s pricing history. The retail hardware price has never been positioned to reflect full manufacturing cost. Selling dishes below cost to acquire subscribers — and recovering margin through monthly service fees — is a classic hardware-as-loss-leader strategy. The bet is that customer lifetime value, expressed through ARPU, holds up long enough to recoup the terminal investment per subscriber. It is the same logic that underlies smartphone carrier subsidies, except the hardware is far more specialized and the serviceable market is geographically constrained in ways a phone never is.

Reading ARPU as a Terminal Story

Starlink’s subscriber growth has been remarkable: from roughly 10,000 active lines in early 2021 to more than 12 million by June 2026. But the ARPU trend running alongside that growth tells a subtler story. Average revenue per user declined from $99 in late 2023 to $91 by end of 2024, and further to $66 as of March 2026 — a roughly one-third compression in per-user revenue in under three years.

Some of that compression is deliberate market expansion. Starlink has moved aggressively into lower-income markets and reduced prices in regions where $100/month fees represent a significant fraction of household income. Some reflects competitive and promotional pressure. But the structural driver is terminal economics: as manufacturing costs fall with scale and accumulated ASIC generations, SpaceX has had room to reduce hardware prices and total customer acquisition costs, and to pass some savings through to service fees to drive faster penetration.

The financial outcome of this approach has been stark. After years of losses, Starlink recorded its first profitable year in 2024 ($72.7 million net), then dramatically accelerated to $11.4 billion in revenue and $4.4 billion in operating income in 2025. That is not the trajectory of a business still wrestling with unsolved terminal economics. It is the trajectory of a company that climbed the manufacturing curve early enough that hardware cost is now a competitive lever, not a constraint.

Kuiper’s Terminal Strategy — and the Challenge It Faces

Amazon’s Project Kuiper, now branded Amazon Leo, enters a market where SpaceX has already completed multiple generations of hardware iteration. Kuiper has announced three terminal tiers: the Leo Nano (roughly 7 inches square, targeting up to 100 Mbps, weighing 2.2 pounds), the Leo Pro (11 inches square, up to 400 Mbps), and the Leo Ultra (a larger 20×30-inch format targeting up to 1 Gbps download and 400 Mbps upload). As of mid-2026, service is in early beta and no retail hardware pricing has been announced publicly.

What Amazon did disclose, back in December 2020, was a cost ambition: a Ka-band phased-array terminal at “less than 20% of the cost of traditional state-of-the-art flat-panel antennas.” That framing — relative to prior-generation enterprise and government hardware — says more about where the technology started than where consumer pricing needs to land today. The meaningful benchmark for Kuiper is whatever Starlink is currently charging for its standard kit, not legacy systems.

Amazon’s structural advantages are real: semiconductor design capability through Annapurna Labs, deep relationships with Asian contract manufacturers, and the ability to bundle or subsidize hardware through the Amazon ecosystem in ways few rivals can match. Whether those advantages translate into terminal economics that can compete with SpaceX’s accumulated experience remains to be proven in the market rather than announced.

The scale problem compounds the economics challenge. The FCC had required half of Kuiper’s 3,236-satellite constellation — 1,618 satellites — to be operational by July 30, 2026. The agency waived that deadline in June 2026, instead demoting spectral priority for late-launched satellites. As of this writing, 396 production satellites are in orbit across 15 missions. Meaningful service density requires far more. And terminal economics only work at volume when there are enough satellites overhead to justify selling dishes to subscribers.

The Ground-Level Moat

In the LEO broadband industry, orbital assets drive the headlines — satellite counts, launch manifests, spectrum filings at the ITU. But the durable competitive moat is on the ground. A phased-array production line operating at millions of units per year, built on several generations of proprietary ASIC design, is not something a new entrant replicates quickly or cheaply. It is capital-intensive, knowledge-intensive, and dependent on supply-chain relationships built over years.

Operators focused on enterprise and government markets — Telesat Lightspeed, OneWeb in its current form — largely sidestep the consumer terminal cost problem. Their customers pay for purpose-built hardware, and the economics are justified by the service tiers and contracts involved. But any operator that wants residential subscribers in price-sensitive markets must eventually climb the same manufacturing curve Starlink has spent a decade on.

That curve is not some abstract concept. It showed up directly in Starlink’s income statement, in its ARPU trajectory, and in the chip shortage that briefly capped subscriber growth. The hardware problem is not fully solved — each new service tier, market, and form factor opens a new iteration of it. But for one company, the hardest part is increasingly in the rearview mirror.

Sources