On Tuesday of this week, SpaceX agreed to buy Cursor for $60 billion in an all-stock deal. This happened four days after SpaceX went public, which means the currency for paying for the deal is a stock price that has existed as a tradable idea for less than a week. SpaceX also told its IPO investors that it sees an addressable market for AI products worth $26 trillion. That is, give or take a few trillion dollars, the entire gross domestic product of the United States.
So if we take those facts together, the official position here is: A rocket company that also has a fledgling social media company, and the fifth best frontier model on the market, is now acquiring a code editor that it is saying is worth $60 billion… payable from not-real-money, that has not had real time to settle, measured against a market the size of the United States.
Still following me? Cause I sure as shit… am not. And I’m the one trying to write this.
At face value, these seem like extremely dumb details. It gets even dumber when you understand that back in April, SpaceX wrote the deal so that it would either buy cursor for $60 billion or pay a $10 billion breakup fee if the deal fell apart. At this exact moment, Cursor was lined up to raise about $2 billion from Andreessen Horowitz, Thrive, and NVIDIA at a $50 billion valuation. Then SpaceX walked through the door, blew up the funding round, offered $10 billion more than the next smartest venture investors in the country, and attached a consolation prize larger than most companies will ever be worth.
The question I keep asking myself is how did we collectively arrive at a place where any of this shit makes sense? The answer is self-evident: this shit does not make sense. Remember that a valuation is just a story that a group of people agree to tell each other is true until somebody else flinches. SpaceX certainly isn’t going to flinch. They just convinced a bunch of retail investors that you can spend money based purely on adrenaline, if that’s what you’d like to do.
The thing that makes this $60 billion acquisition genuinely funny… is the physical thing that is currently sitting next to me on my desk. I have an Nvidia Orin AGX. It has about a 50 watt power draw. I could fit it in my Star Wars lunchbox, and I have a Mistral coding model running on it inside an open agentic harness called Vibe. That set up gets me, in my own daily use, probably somewhere between 95% and 98% of what Cursor gives me at a hardware cost you could cover with like… a single BMW lease payment. What I’m doing isn’t some hard-to-achieve hobbyist fantasy world. It is something that a lot of people have figured out how to do because all of the software is insanely easy to install.
The cat has been out of the bag since the end of March when somebody at Anthropic accidentally shipped the entire source code for Claude Code inside a routine npm update, which was about a half a million lines of code across 1,900 files.
Whoops!
Anyway, within the next 24 hours, a million zillion people had forked it and rebuilt it and refactored it into Python. The entire secret sauce got open-sourced completely by accident, and then a whole phalanx of volunteers made a better version and kept making better versions. The open source community is magical like that.
The mass of this open source flywheel is something that makes this $60 billion valuation kind of stupid. The better these systems get at writing code, the better they get at writing the frameworks and harnesses that make code-writing systems good. They improve themselves, and a meaningful slice of that improvement lands in the open source world where I tend to operate. That means it immediately starts competing with these paid products. So if an open weight model delivers 98% of a frontier model’s coding ability, the entire commercial argument can be shrunk down to one real question: What is that 2% actually worth to you? Is it worth a subscription and vendor lock-in that amounts to a $60 billion product?
My answer to that question is no, and now I’m going to talk about football. Because this entire situation is the quarterback market problem wearing a software analogy.
For the last few years, OpenAI and Anthropic ran the rookie wage scale on all of us. The rookie wage scale, if you don’t follow the NFL, is the closest thing that NFL owners have to a salary cheat code. You draft an elite quarterback and you pay him like a backup quarterback for four years, and you use the savings to surround that quarterback with talent. Patrick Mahomes, Josh Allen, and Brock Purdy: every single one of them was a contender while still being paid a fraction of their real market value. Purdy is the most hilarious example because he was Mr. Irrelevant, the literal last pick of the NFL draft, quarterbacking a Super Bowl roster on the kind of money that gets spent on Nissan Altimas.
That’s the same thing that OpenAI and Anthropic did. They had frontier intelligence models, and they were pricing it like a rookie wage scale quarterback, a guy who was making one twentieth of what he might command in a few years. They wanted you to build your habits and your codebase and all your dependencies around their metaphorical quarterback. They knew in a few years that the quarterback would get the price that the market would actually command.
And for the first few years, we all enjoyed franchise quarterback output at practice squad prices.
Now, we can talk about the open weight models and Sam Darnold.
The first time Sam Darnold threw a pass in the NFL, he threw a pick six across the field; the dumbest thing an NFL quarterback can do. He was a turnover-prone disappointment on a bad Jets team, and I think most people wrote off Sam Darnold. I wrote off Sam Darnold, but mostly just because he was the quarterback for the Jets. That’s exactly how open weight models started. They were weird. They hallucinated. They had short context windows, and they made us all think, “Wow, these things really suck.”
Then something funny happened. Sam Darnold got better because he started working in better quarterback rooms and he learned the position. He went to Minnesota and had a Pro Bowl year where he won 14 games and had a real receiver to throw to in Justin Jefferson. Then he signed a contract in Seattle and won a Super Bowl, beating the New England Patriots. The talent was always there with Sam Darnold. He just needed time to improve and get better and put a real system around him, but now he’s an excellent quarterback. The open weight models are almost exactly the same narrative. They started off pretty weird, and if you weren’t paying attention, you didn’t know that there was this development process that was happening. Now these open weight models are really good.
Sam Darnold is the 18th highest-paid quarterback in the league by average annual contract value. That means that there are 17 quarterbacks who make more money than the man who just won the Vince Lombardi Trophy. No person who watches football and is serious is going to put Sam Darnold in the same sentence as Mahomes, Allen, or Burrow. Those are your frontier models, your Fables, your Codex. They operate at a ceiling that Sam Darnold will never achieve on his best day. But the ceiling is not the question we have to answer when it comes to what allows a model to render high-level results. The answer that we’re looking for is: What does it actually take to achieve tangible value in business results?
The frontier model alone will not deliver results. If you hand Patrick Mahomes four practice squad-level receivers and a tight end who is past his prime and planning a wedding with Taylor Swift, then you’re going to get a quarterback who throws for 225 yards a game and loses a lot of games. Sometimes you’ll see a frontier model do something heroic, but the system around it still matters.
A good open weight model inside a great harness goes 14 and 3 and wins a Super Bowl. Harness engineering is your offense line having an effective pass rush. Route concepts that make sense against the defense you’re playing… And open source harnesses are meeting the moment with open weight models.
So now I have to talk about what SpaceX actually bought with Cursor.
Cursor is a fine product. I have used it. It is competent, but it is in its own space a good WR3 or WR4. It’s useful sometimes, but it’s not something that I would want to build a franchise around, and it’s not even the best at its own position. Cursor is not a market leader, and there is nothing structurally underneath it that the open harnesses and the leaked patterns and the lighter clones have not already replicated and/or exceeded. So they’re paying $60 billion for Cursor. I’m here to tell you this doesn’t make sense.
This is effectively a third-round receiver who had a really fast 40 time, who hasn’t really played effective NFL football, and you’re paying them like they’re a top-three NFL receiver. You’re saying you can build your team around Cursor, when at its best, Cursor is a 40 reception a year receiver. This is Jerry Jones behavior. I know because I’m a Cowboys fan and I’ve had to deal with this my entire adult life.
Speculation is how businesses and football teams get hurt. The reason sports eventually correct and bubbles eventually burst is because the field settles every argument. You do not pay football players for the future they might have. You pay them for the games that they’ve already won and how they have already demonstrated their market value. You are paying a price for a number to say, “This is how much winning you can buy.” This is simply a smart pricing strategy. Cursor has not won anything. It got drafted higher than expected. It tested well at the Combine, and then it convinced a company having the best week of its life to guarantee its contract before they ever took a real snap in an NFL game.
I have full confidence that we are absolutely living in bubble pricing, because I’m a Dallas Cowboys fan. Buying Cursor for $60 billion looks a lot like Jerry Jones running a rocket company.