Why a rocket company paid $60 billion
for a code editor

SpaceX just paid $60 billion for a coding tool. All stock. No cash. The target was Anysphere, the company behind Cursor. And the reason a rocket company bought a code editor tells you everything about where software is heading.

Here is the proof. Cursor reportedly hit over $1 billion in annualized revenue inside three years of founding. By May 2026, CNBC's reading of Ramp spending data put Cursor at roughly 26% of AI coding tool spend, with Anthropic's tools at about half that, near 13%. That is not a tool anymore. That is infrastructure. I think SpaceX saw that clearly, and the price tag proves it.

The Owned-Layer Principle

Here is the framework: when a capability becomes load-bearing, you stop renting it and start owning it.

Renting works fine when something is just a feature. You pay per seat, you swap vendors, you move on. But once a thing sits under your entire operation, renting turns into risk. Every price change is your problem. Every API shift is your downtime.

SpaceX runs rockets, Starlink, satellites, and reportedly autonomy software systems. All of it is software now. So when AI coding agents started writing, reviewing, and merging that software, the agent stopped being a tool. It became the layer the whole engineering org stands on.

The Owned-Layer Principle says this: you buy what you cannot afford to lose access to. Cursor was that. So they bought it.

Why a Rocket Company Bought a Code Editor

Step back and look at the structure of the bet, not the headline number.

SpaceX is one of the most vertically integrated companies alive. That same compute reportedly trained Cursor's Composer model and xAI's own Grok.

So the stack already existed in pieces. Hardware at the bottom. Frontier models in the middle. And now an agentic application layer on top, where humans and agents ship code together. The Cursor deal just welded the top of the stack to the bottom.

This is counterpositioning in plain sight. Most companies consume AI through someone else's API. OpenAI's Codex. Anthropic's Claude Code. Each of those is a competitor's platform that can change pricing, change limits, or change priorities overnight.

We saw the warning shot elsewhere. Google moved to limit Meta's Gemini access over capacity constraints. Read that slowly. One frontier lab throttled another giant because compute is scarce. When the supplier can starve you, supply is not a convenience. It is leverage held against you.

The asymmetric logic here is simple. The downside of NOT owning it is depending on rivals for the layer that writes your rockets. One risk is financial. The other is existential.

I will own the honest part. A $60 billion price for a single coding company is narrow. If developer taste shifts away from Cursor's interface, the asset cracks. It is unclear whether one editor can stay dominant for a decade in a market this fast.

But the strategic frame changes the math. SpaceX is not buying a P&L. It is buying a permanent seat at the agentic layer, embedded across its whole engineering body. Salary buys furniture. This kind of equity buys the factory floor.

There is a contrast worth holding here. Amateurs ask, "What does this tool cost per seat?" Leaders ask, "What happens to us if this layer disappears?" Those are different questions, and they lead to different checkbooks.

That is the real prize. Cursor stops being the app sitting on other people's models. With Colossus behind it, it gets its own training pipeline. The tenant becomes the landlord.

2031

Three signals inside the same shift

Pull the lens back five years.

By 2031, I think the build-versus-buy question for serious technical companies will look almost unrecognizable. The old pattern from 2018 to 2023 was clean: consume AI via API, integrate it into your existing tools, keep moving. That pattern assumed the model layer was a stable utility, like electricity.

It is not a utility. It is contested ground. When Google can throttle Meta over capacity, the model layer behaves more like oil reserves than the power grid. Scarce, strategic, and worth fighting over. That single fact rewrites the calculus.

Here is the contrast that will define the next half-decade. Renters optimize for cost. Owners optimize for control. In a scarce-compute world, control compounds and cost does not. The company that owns its agentic layer ships faster every quarter, because nobody can throttle it. The company that rents ships at the mercy of its supplier's roadmap.

That does not mean everyone should buy a Cursor. Most companies cannot, and should not. The lesson is not "spend $60 billion." The lesson is to know which layer is load-bearing for you, and treat that one differently from all the others.

Think of it as a compounding flywheel. Own the layer, ship faster, learn faster, ship faster again. Rent the layer, and your flywheel turns at your vendor's speed. Over five years, those two curves separate violently.

My read on this: the SpaceX move will get cited as a precedent far more often than it gets copied. Few have the balance sheet. But the mental shift, treating coding agents as infrastructure rather than software, spreads to every technical org regardless of size. The data is mixed on how fast traditional enterprises follow, but the direction looks one-way.

What to Build This Weekend

You do not have $60 billion. You do not need it. The principle scales down to a laptop and a weekend.

Your job this weekend is to find your own load-bearing layer. Not buy it. Just identify it. The thing that, if it vanished, would stop you cold.

First, list every AI tool and API your projects depend on. Write them on one page. Be honest about which ones you could swap in an hour, and which ones would break everything.

Second, pick the scariest one. The one you cannot easily replace. That is your owned-layer candidate. Now ask: do I understand it well enough to rebuild a basic version if I had to?

Third, build a tiny escape hatch. If you depend on one model API, wire in a second provider behind a simple switch. This is called a fallback. It means if your main vendor throttles you, your app keeps running on the backup.

A quick definition, since I used it. An agentic coding tool is software where the AI does not just suggest code, it plans, edits, tests, and ships across multiple steps. Cursor is the leading example. Claude Code and Codex are the main rivals.

Get your reps in here. Open Cursor, or any agentic editor, and let it build one small thing end to end. A script. A landing page. Watch where it shines and where it stumbles. Things will break. That is the point of learning in public.

You will not own infrastructure this weekend. But you will start thinking like someone who knows the difference between a tool and a layer. That shift is worth more than any single app you build.

Take a deep breath and take it step by step. The giants are buying their layers. You can at least learn to see yours.