SpaceX's IPO Bets Everything On One Number: The 95% Collapse In Cost-To-Orbit

A $1.77 trillion valuation that careful models can't justify makes perfect sense once you see the cost of putting internet bandwidth into orbit fall from $6.55 to 30 cents per megabit per second, and why the same curve unlocks data centers in space.

There's a single, deeply unsexy number sitting underneath the largest IPO in the history of human capital markets, and almost nobody buying the stock is talking about it. Not the subscriber count. Not the revenue. The cost of putting one unit of internet speed into orbit, measured in dollars per megabit per second. That number just fell by 95% in one generation of hardware, and once you understand why, the gap between what the spreadsheets say SpaceX is worth and what the market is paying stops looking insane and starts looking cheap.

Key Takeaways

• SpaceX priced its IPO in June 2026 at $135 a share, valuing the company around $1.77 trillion and raising roughly $85 billion. That's more than two and a half times Saudi Aramco's $29 billion in 2019, making it the biggest IPO ever.
• Morningstar's discounted cash flow model came out to about $780 billion, less than half the IPO price. At a $2 trillion valuation you're paying 107 times revenue, when a hot tech company trades at maybe 30.
• Falcon 9 delivered bandwidth to orbit at $6.55 per megabit per second. Starship, at nearly the same launch cost, drops that to about 30 cents. One Starship launch replaces 23 Falcon 9 flights.
• Starlink already runs the business. In Q1 2026, connectivity was $3.26 billion of $4.69 billion in revenue, roughly 69%, at a 63% EBITDA margin. That's a software margin on a hardware product.
• Ten million subscribers, doubled in a year, heading for 17 million by the end of 2026 - and prices went UP by as much as $10 a month in May. You don't raise prices when customers have alternatives.
• Oppenheimer slapped an outperform rating on it with a $190 target, or $2.5 trillion, higher than the IPO. BlackRock reportedly wanted at least $5 billion of the deal.
• SpaceX unveiled the AI-1 satellite in June 2026: a 150-kilowatt rack of compute with a 70-meter wingspan, and filed to build up to a million of them in Bastrop, Texas.
• Everything hinges on one question - does Starship get under $250 per kilogram and stay there? Answer yes and the whole bull case follows. Answer no and treat the valuation exactly as the bears do.

---

The Gap Nobody Can Explain With A Spreadsheet

Let me steel-man the bears, because their case is genuinely strong. Morningstar is a serious, buttoned-up research shop, and they ran the textbook exercise: project the cash the business throws off for the next decade or two, discount it back to today's dollars. That's a discounted cash flow, the most conservative, most defensible way to price any company. Their answer was about $780 billion.

The IPO priced SpaceX at $1.77 trillion. So the careful math says the market is paying roughly double what the numbers justify. Put another way, at a $2 trillion valuation you're paying 107 times annual revenue. A healthy fast-growing tech company trades at 10 or 15 times sales. A really hot one, 30. SpaceX is at 107.

So the cynical read writes itself. A hype machine. The greatest bag-hold ever constructed, a bubble with a rocket painted on the side. If you bought this purely on the financials as they exist today, you paid a price no spreadsheet on Earth currently supports. I want to be honest about that, because the answer to it isn't a vibe. It's a number the DCF literally cannot see, because it hasn't fully happened yet.

The One Number That Governs Everything

People still think of SpaceX as the rocket company. At this point it's a telecom company that happens to own the cheapest rockets in the world, plus an AI company bolted on. Most of the cash comes from Starlink, the satellite internet service beaming connectivity down to people who can't get it any other way. The rockets largely exist to feed the satellites.

So the number that decides the whole thing is the cost to deploy bandwidth. Megabits per second is just internet speed - the figure your provider advertises when they sell you a "500 megabits per second" plan. Dollars per megabit per second to orbit asks: how much does it cost SpaceX to launch one unit of that speed into space, where it can beam it back down? It's the unit cost of building the internet from the sky, and it's SpaceX's single biggest expense.

Track that number and you can ignore almost everything else. Forget subscribers for a second. Forget even the revenue. The cost to deploy is the master variable.

From $6.55 To 30 Cents

Start with the old world. Falcon 9 is a reusable rocket that lands itself on a drone ship in the ocean, and it already cut the cost of reaching space by something like 85% versus everything before it. It carries about 17 tons to orbit at an internal cost near $1,000 per kilogram, so call it $17 million to fill it up and launch. That batch of satellites adds up to about 2,600 gigabits per second of capacity. Do the division and you land at roughly $6.55 per megabit per second.

And $6.55 was already the cheapest number in human history. Everyone in the industry looked at it and assumed it was the floor, about as good as physics would allow.

Then comes Starship. It's the biggest flying object humans have ever built, a fully reusable monster designed to carry nearly 100 tons - almost six times Falcon 9's payload - at under $185 per kilogram, a fifth of the cost. A single launch runs about $18.5 million, barely more than Falcon 9. But you don't launch the old satellites. You launch Starlink V3, which carries roughly four times the bandwidth per kilogram of mass. Two multipliers stacking on top of each other: six times the mass, four times the bandwidth per unit of mass. One Starship deploys about 60 V3 satellites and 61,000 gigabits per second of capacity. Run the division and you get about 30 cents per megabit per second. From $6.55 to 30 cents. A 95% collapse.

Why A 95% Cost Collapse Rewrites The Map

Imagine you ran an airline and jet fuel fell 95% overnight. Or a steel mill and iron ore dropped 95%. Two things happen, and they're both enormous. Your margin explodes, because you keep charging customers the same $80 a month while it costs you a fraction to serve them. And things that were flatly uneconomical at the old cost suddenly become trivial. Markets that made no sense at $6.55 make perfect sense at 30 cents.

This is the oldest pattern in technology, and it has a name: Wright's Law, the experience curve. Every time cumulative production doubles, cost falls by a predictable percentage. The Model T went from a rich man's toy to something a factory worker could buy, and it didn't just sell more cars, it reshaped American cities. Solar panels fell about 99% over a few decades and became the cheapest electricity ever generated.

The cleanest rhyme is the shipping container. In 1956, Malcom McLean stacked 58 metal boxes on a converted oil tanker, and the cost of loading cargo fell from $5.86 a ton to about 16 cents a ton. That one collapse didn't make shipping cheaper - it created globalization, made it economical to build a thing on one side of the planet and sell it on the other. That's what $6.55 to 30 cents does. It doesn't improve satellite internet. It makes a whole category of things economical that weren't.

Starlink Is Already Winning, And Already Printing Cash

For the bull case to bridge $780 billion to $1.7 trillion, three things have to be true. Two of them already are.

The first is that Starlink becomes the default winner of internet. When your deployment cost falls 95%, you can drop prices into markets nobody else can afford to serve and flood capacity into markets you own, so service never degrades as you add millions of users. Look at the field. Amazon's Kuiper is in enterprise beta only. AST SpaceMobile has around 45 satellites targeting intermittent US coverage. Fiber wins on latency - 11 to 14 milliseconds versus Starlink's 25 to 60, which matters if you're a hardcore gamer. But you can't run fiber to a cargo ship in the Pacific, a farm in rural Montana, a village in the Congo, or a plane at 38,000 feet. Two-thirds of Earth's surface is unfiberable, and with direct-to-cell beaming straight to the phone in your pocket, Starlink is the only company that can light all of it up. The proof it already knows this: it raised prices by up to $10 a month in May 2026. You don't do that when customers have options.

The second condition is that those margins are real and fund everything else. Starlink runs about a 63% adjusted EBITDA margin - meaning for every dollar of subscription revenue, roughly 63 cents is raw cash profit before interest, taxes, and equipment depreciation. The research firm Quilty forecasts Starlink at around $20 billion in revenue for 2026, about $14 billion in EBITDA, and $8.1 billion in free cash flow, the actual spendable money left after building new satellites. This isn't a money-losing land grab. It's a cash cow, and it's on the S1 filing. That cash funds Starship, the next satellites, and the science fiction.

The Trillion-Dollar Payoff Is Data Centers In Orbit

The third condition is the one bears can't model, and it's the one that actually justifies the number: orbital AI compute. In June 2026, right before the IPO, SpaceX unveiled the AI-1 satellite. It flies the equivalent of an entire rack of AI computers at 150 kilowatts of peak compute, about the same as one of NVIDIA's top-end server racks, with a 70-meter wingspan wider than a Boeing 747. SpaceX has filed to eventually build up to a million of them in a new factory in Bastrop, Texas, with volume by the end of 2027.

Why put a data center in space? Because the two things strangling AI on Earth right now, power and cooling, are nearly free up there. You get about 1,361 watts per square meter of continuous solar with no night and no clouds, no terrestrial grid to wait years of permits for. You radiate waste heat straight into the cold of space, so no water, no cooling towers, no reservoirs drained to keep a town's data center cold. And AI-1 is derived from the Starlink V3 platform - same laser links, same solar tech, same factory muscle. The rockets feed the satellites, the satellites feed the internet, and the platform feeds the data centers. It's one machine.

This isn't only SpaceX talking. A startup called StarCloud flew an actual NVIDIA H100 chip in orbit back in November 2025, has a follow-up with 100 times the power coming before the end of 2026, and is valued over a billion dollars. Google is chasing its own project. When three rivals who hate agreeing on anything independently decide the same crazy idea is worth pursuing, that's a signal worth taking seriously.

The Whole Bet Compresses To One Sentence

Bloomberg warned on the orbital data center economics. TechCrunch called them brutal. The bears are right that at today's launch cost, the math doesn't math - AI chips go obsolete every two to three years, so you'd have to keep shipping hundreds of thousands of tons of new silicon to orbit forever, and that's hundreds of billions a year just in launch. At today's cost, it's a non-starter.

But look at the break-even. Orbital compute pencils out somewhere around $500 to $1,000 per kilogram to launch. Starship is targeting under $185 to $250. The exact same cost collapse that took satellite internet from $6.55 to 30 cents is what drags orbital compute from insane to inevitable. It's the identical problem - cost per kilogram is the master variable for both businesses. Starlink is the proof of concept that the curve is real and already paying out. The data centers are the trillion-dollar payoff the curve unlocks next.

So here's how I'd think about it, and this is not investment advice. Two of the three conditions are demonstrably true today - Starlink is winning, and the margins fund everything. The third is a physics and engineering bet on the most capable rocket team in human history. Ask yourself one question and only one: do you believe Starship gets under $250 a kilogram and stays there? If yes, the rest follows logically and the IPO looks cheap. If no or not sure, treat the $1.7 trillion exactly as the bears do and size accordingly. Either way, whether or not you ever own a share, you live downstream of that curve. Cheaper bandwidth brings billions of new people online. Cheaper launch turns space from a government-sized project into a place where actual businesses get built. That's where a lot of the next decade's wealth gets created.