TraviaTechPie Review

The Story

For years, “how does Starlink actually make money” was a guessing game. Analysts squinted at satellite counts and made up numbers. Then SpaceX filed to go public in May 2026, and the guessing stopped. The S-1 put the real figures on the table — and they tell a much more interesting story than the launch footage ever did.

Here’s the headline number: the connectivity business, which is basically Starlink, pulled in $11.4 billion in revenue in 2025. That’s about 61% of SpaceX’s $18.7 billion total. And it was the only segment turning a real profit — roughly $4.4 billion in operating income — while the company as a whole still posted a $4.9 billion net loss for the year, dragged down by Starship development and the AI segment the S-1 folds in (roughly $3.2 billion). So the mental model most people carry around — SpaceX is a rocket company — is backwards. It’s an internet company that happens to own the world’s best rocket. Let’s follow the money.

### The subscription flywheel

At its core, Starlink is a boring, beautiful business: monthly recurring revenue. You buy a dish, you pay every month, and if SpaceX plays it right, you keep paying for years.

The subscriber curve is the part that jumps out. Starlink went from 2.3 million subscribers at the end of 2023, to 4.4 million in 2024, to 8.9 million in 2025, and hit 10.3 million by March 2026. By June 2026 it crossed 12 million subscribers across 164 countries. Whatever the exact figure by the time you read this — subscriber counts move fast, so treat any single number as “roughly X as of this date” — the shape is a rocket of its own.

But here’s the twist that makes this interesting. As subscribers exploded, the money per subscriber went the other way. Average revenue per user — ARPU, the monthly amount an average customer pays — fell from about $99/month in 2023, to $91 in 2024, to $81 in 2025, and down to $66/month by March 2026. That’s not a business in trouble. That’s a deliberate trade. SpaceX has been cutting prices and pushing cheaper regional plans to grab market share in places where $99 was a non-starter. Call it the “land grab discount”: give up margin per user now, lock in tens of millions of subscribers, worry about squeezing them later.

### The dish that lost money on purpose

Now the part almost nobody outside the industry understands: the hardware.

That pizza-box dish (SpaceX calls it “Dishy”) is a phased-array antenna — a flat panel packed with tiny elements that steer the beam electronically to track satellites racing overhead, with no moving parts. That technology used to live only on military hardware costing a fortune. Early on, each terminal reportedly cost well over $2,000 to build, with analysts estimating it fell to around $1,300 by roughly 2021. And SpaceX sold them for a few hundred dollars. On purpose. At a loss.

Why torch money on every box? Because the dish isn’t the product — it’s the toll booth. Every terminal sold at a loss is a customer who now owes you $66-plus every month, potentially for years. The upfront hardware hit is a customer acquisition cost, and the subscription pays it back. Classic razor-and-blades, except the razor is a spacecraft antenna.

The good news for SpaceX is that this loss has been shrinking. In its S-1, the company disclosed that it cut terminal manufacturing cost by about 59% in 2025, and that same year it shifted its US offering to a $0-hardware rental subscription model — you rent the dish instead of buying it, folding the hardware cost into the monthly bill. That matters because the hardware subsidy had been one of the biggest drags on the whole model. Cheaper dishes mean each new subscriber costs less to sign up, which matters enormously when you’re adding millions a year.

### Where the fat margins actually live

Consumer broadband gets the headlines, but it’s not where the richest money is. The real leverage is in the customers who can’t get a cable running to them: ships, planes, oil rigs, disaster zones, and militaries.

The post-IPO Q2 2026 results make this vivid (these are quarterly earnings figures, separate from the full-year 2025 numbers in the S-1). Consumer revenue that quarter was about $2.49 billion, up 44% year over year — healthy. But enterprise and government revenue was about $1.81 billion, up 108%. The B2B and government side is growing twice as fast, and those customers pay maritime, aviation, and enterprise rates that dwarf the $66 home plan. A cruise line or an airline outfitting a fleet isn’t price-shopping against a $40 residential plan; it’s paying for connectivity that simply has no alternative mid-ocean.

Then there’s the government layer, which is quietly enormous. The US government and military are among SpaceX’s biggest customers, and a chunk of that runs through “Starshield” — a hardened, government-only cousin of Starlink built for national-security work. In 2026, SpaceX won some of its biggest defense awards yet: about $2.29 billion for a US Space Force data network and about $4.16 billion for a satellite network to track airborne targets. These are lumpy, multi-year contracts, not monthly subscriptions — but they’re high-margin, sticky, and very hard for a competitor to displace.

### The cost side nobody screenshots

A subscription machine is only as good as the cost of feeding it. And Starlink’s costs are relentless in a way most software businesses never face: the network physically falls out of the sky.

Starlink satellites are designed for a working life of only about five years, and many get deorbited even sooner. That’s not a flaw — low orbit means low latency and easy disposal — but it means the constellation is a treadmill. To keep service running, SpaceX has to replace on the order of a thousand satellites a year, every year, forever. This is “replenishment capex,” and it’s the structural cost that makes satellite internet so different from a fiber network you dig once and run for decades.

Here’s where the whole thing hangs together, though — and it’s the same thread as the launch-cost story. SpaceX builds its own satellites (analysts estimate V2 Mini units have come down to roughly $250,000–$400,000 each, from around $500,000 earlier) and launches them on its own reused Falcon 9 rockets. A single Falcon 9 flight lists for tens of millions of dollars — around $74 million as of 2026, and the price has been drifting upward — and lofts about 20–24 satellites at once. Because SpaceX pays itself internal cost for launch instead of a competitor’s sticker price, the marginal cost of adding capacity is a fraction of what any rival faces. That vertical integration — owning the rocket — is the entire moat. If you want the mechanics of why launching your own hardware is so cheap, that’s the economics of reusable rockets.

Add it up and Starlink is brutally capital-intensive up front — satellites, ground stations, spectrum licenses, the whole constellation — but once the network exists, each new subscriber is close to pure margin. That’s how you get to a $4.4 billion operating profit: the fixed cost of the sky is enormous, and the variable cost of one more customer is tiny.

### The competition finally shows up

For most of its life Starlink had the LEO broadband market essentially to itself, which was a warning sign as much as a bragging right. That’s changing on two fronts.

The Western challenger is Amazon, with what’s now branded “Leo” (formerly Project Kuiper). It has a few hundred satellites up as of 2026 against a planned 3,236, and it’s leaning on Amazon’s cloud and retail muscle. Europe’s Eutelsat OneWeb runs a 647-satellite constellation but sells only through telecom partners — no consumer plans — so it’s playing a different, B2B-only game. The one to actually watch is China: the Guowang constellation plans about 13,000 satellites and Qianfan up to 15,000, together roughly 28,000 planned, with a few hundred launched so far and deployment accelerating. Those aren’t really aimed at competing for a Kansas farmhouse; they’re about serving China and its partner nations, which could split the world into Western and Chinese satellite blocs and cap how much of the planet any single operator can address.

The other front is “Direct-to-Cell” — beaming text, and eventually data, straight to an ordinary phone with no dish at all, through carrier partners like T-Mobile. SpaceX has hundreds of these specialized satellites up. It won’t replace your home connection, but it turns the dead zones on every hiking trail and shipping lane into coverage, and it opens a market of billions of existing phones rather than millions of dish-buyers.

The Takeaway

Strip away the spectacle and Starlink is a subscription business with an unusually literal cost of goods: it has to keep throwing hardware into space to replace the hardware that’s burning up. What makes the model work isn’t magic. It’s that the two things which would normally make satellite internet impossibly expensive — building the satellites and launching them — are both done in-house, at cost, on reused rockets. Every rival has to buy at least one of those on the open market. That’s the whole ballgame.

The number I keep circling back to is that ARPU falling from $99 to $66 while subscribers quadrupled. On paper that looks like a company getting weaker per customer. My read is the opposite: it’s a company that has decided scale is the moat, not price. Once you’ve got tens of millions of paying dishes and the only global constellation that’s actually profitable, you can raise the premium tiers, mine the enterprise and government seams, and let the cheap consumer plans do the land-grabbing. The May 2026 S-1 basically confirmed the plan is working — connectivity is the one part of SpaceX printing money while everything else spends it.

What ties this to the reusable-rocket story we looked at is that they’re really the same insight from two angles. Reusability collapsed the cost of getting to orbit. Starlink is what you build once getting to orbit is cheap: a business that only pencils out if launch is a rounding error, spraying a thousand satellites a year into the sky and billing the ground below. The rocket was never the product. It was the on-ramp. And the more consequential question for the next few years isn’t whether SpaceX can launch — it obviously can — but whether Amazon, or China, or anyone can get their own launch costs down far enough to fight a war of replenishment they can afford to keep losing money in. That’s the space beat I want to keep pulling on here.

This article is for informational purposes only and is not investment advice.

Photo: Evgeny Opanasenko / Unsplash