It's the first time in the history of the world.
$60 billion to buy a ticket.
On October 10, more than 2,000 people attended the X Takeover event, where Musk was remotely connected. The host asked him directly: Why would you buy Cursor?
In just 61 seconds, Musk's response quickly spread across the web.
He said that if you want to create a super-intelligent like this, you have to program -- and as far as he knows, there are only three monkeys in the world that have enough programming training data.
OpenAI, Anthropic, Cursor. Google is barely half that.
In this list, there is no xAI, no Grok, and no Musk himself.
A man with a rocket in his hand, a supercomputer named Colossus, a man named Tesla, admitted in front of more than 2,000 spectators:
On the most critical route to super-intelligence, I didn't have a ticket myself.
So he bought the Cursor.
Why is programming data so valuable?Cursor is an editor that programmers write code in every day.
AI gives you a suggestion that you can take with tabs, delete directly, or reuse a few lines.
Each such operation is tantamount to criticizing AI for a working pair of punches, a wrong fork, and a half-pair of the correct answers.
It is reported that Cursor has used trillions of tokens in its training to interact with real developer data.
This data, which can't be crawled online or created by a lab, can only be generated by thousands of real programmers, one line at a time, one day at a time.
In the same interview, Musk himself explains why programming is so special: the logic of the software is fixed and can run unit testing.
The code says yes or no, and you run around and find out that the answer doesn't depend on who subjectively scores.
Programming is one of the easiest subjects for AI to practice, and the larger the database, the more criticism, and the faster it can be practiced.
And the quality of the reviews is critical.
He reviewed the new version of Grok's reinforcement training record and found that AI was sometimes right, but wrong, because the title itself was a problem.
He said he hated that feeling in exchange for himself.
The problem is that AI is so much better, and it's even more distorted. And every time a real programmer tries to change this code, I accept it, and I refuse it, and here I change the two lines, which is the most reliable criticism, because it comes from real work, and it's judged wrong by real people.
Whoever holds the biggest, cleanest book of puzzles is one step closer to ASI.
He's got it all in his hand, and it's kind of distorted.Back to the bottom of Musk's house: there's massive text on X, Tesla has road data, SpaceX has engineering data, and Colossus has the world's top GPUs.
This is a piece of code written by real people who are partially lacking in it.
The IDC said in its analysis in early September that Musk had admitted in his affidavit that xAI had used OpenAI to train Grok.
It's not enough to copy the answers of others, but it's not that far away, and Musk's heart is clearer than anyone else's.
Computers can make money, and models can take time to follow. But this kind of programming data has to be done by the user line by line.
Cursor is a company that was founded in 2022 and in just a few years, only three things in the world have been put in the mouths of the masks.
In April, when SpaceX and Cursor announced a partnership, SpaceX was given the option of either paying $10 billion to continue the partnership or spending about $60 billion to buy the entire space.
SpaceX chose to buy the whole thing.
At $60 billion, the most expensive acquisition of a startup in history, Musk himself justified that it was not buying a product, but a ticket.
There are four things that are super-intelligent, and the most surprising is the fourth.At the event, Musk simply revealed the recipe for creating super-intelligence:
Data, talent, computing skills, and a good sales team.
The first three are all consistent, and the fourth is a really surprising sales team.
What's super-intelligence got to do with sales?
The relationship is huge. Thousands of corporate partnerships mean that Cursor has been installed in the computers of these corporate programmers.
Superintelligence wants to get into the real world of work, and that's the way it works. The more doors open, the more people use them, the more new programming data, the better the models practice, and the more companies it can sell to.
It was built and used to build a fleet.
In fact, Musk thinks he's better than anyone else: in a real company, programmers write code every day, turn back suggestions, the theme comes from real people, and it's judged by real people. Every time a company opens its doors, AI sells more than one real workbook.
Superintelligence is not just a technical issue, it's also a business issue.
On July 8, SpaceXAI and Cursor jointly released Grok 4.5, the main programming language and Agent, which is trained on tens of thousands of GB300 units.
Musk says it's Opus-level, faster, more efficient, and cheaper.
However, the actual score was 64.7 percent for SWE-Bench Pro, 80.3 percent for Claude Fable 5, and 83.3 percent for Terminal-Bench 2.1.
Tickets have been bought and people are sitting in the back row.
On October 4, Musk confirmed on X that SpaceXAI had changed its name to SpaceXSI for Super Intelligence, or Super Intelligence.
OpenAI has a Codex and a huge amount of programming dialogue in ChatGPT.
Anthropic has Claude Code, and the use of programming scenes is amazing.
The first two masks, models, programming products and programming data, are already in his hands.
Interestingly, Musk's evaluation of his opponents was not low. In the interview, he talked about music videos made by AI, specifically praising Claude, saying "to give them a lot of praise".
You think you're writing code.In fact, you're teaching superintelligence to code.Musk's interview is equivalent to rewriting the scoring rules of the ASI competition.
In the past few years, everyone has been counting cards. Who has more GPUs, who is near the end.
But now, one of the largest computing conglomerates in the world, pulling out the highest price ever paid for a startup to buy something that can't be traded for.
Human judgments.
Computers can make money, and the data can only come from humans. Every record in the programming data is a trade-off from a real programmer, this one is accepted, that one is withdrawn, that one is re-submitted.
So the tuition fees for super-smart things, not the giants burning electricity, are pressed out by programmers all over the world.