把自己绑上SpaceX战车，Cursor的600亿美元崛起之路
Original title: Inside Cursor's wild rise
Original author: Shubhangi Goel and Charles Rollet, Business Insider
Original translation: Peggy, BlockBeats
Editor's note: This article tells the story of Cursor CEO Michael Truell and the rapid rise of this AI coding unicorn.
In 2019, Truell, then an 18-year-old MIT student, completed a programming test in less than 10 minutes that was originally expected to take an hour. A few years later, he co-founded Anysphere with several MIT classmates and launched Cursor, aiming to redefine how developers write code. By the end of 2025, Cursor was used by millions of developers, and its revenue grew 10 times in less than a year, surpassing $1 billion.
But Cursor's story is more than just a Silicon Valley narrative of a "genius programmer's entrepreneurial success." The more noteworthy part of the article lies in its revelation of the structural predicament faced by AI application companies: When a company is built upon frontier models, it can grow rapidly by leveraging the model's capabilities, but it can also be quickly squeezed when the model supplier decides to enter the arena itself. This is precisely the relationship between Cursor and Anthropic. Cursor was once highly dependent on Anthropic's models. After Anthropic launched Claude Code, the two transformed from partners into potential competitors. Consequently, Cursor began to push for its own self-developed model, Composer.
At the same time, Cursor's high growth has also been accompanied by controversy. The article mentions that Cursor's recruitment process is extremely rigorous, with candidates asked to participate in multi-day, or even weeks-long, unpaid "working trials." Internally, there has also been a long-standing fear of over-reliance on a single AI model supplier. These details make Cursor's success seem more complex: It is both one of the most representative application-layer companies in the AI coding wave and a startup trying to find a balance between rapid expansion, an intense culture, and model dependency.
What truly moves the story into a new phase is Truell's tie-up with Elon Musk's SpaceX. To support its self-developed model, Cursor needed expensive and scarce computing power. Meanwhile, SpaceX/xAI needed to improve Grok's coding capabilities. On the surface, the cooperation involves a complementary exchange of computing power in return for data and model capabilities, but behind it lies a potential $60 billion acquisition arrangement. If the deal ultimately goes through, Cursor could become a key coding infrastructure within Musk's AI ecosystem. If it remains independent, it must also prove that an AI application company can grow into a true generational company amidst the constraints of frontier model giants.
The core question of this article is: Will Cursor become the gateway for the next generation of software companies, or just a piece of the puzzle in the computing power wars of AI giants?
The following is the original translation:
Michael Truell: From Prodigy Programmer to Cursor CEO
In 2019, 18-year-old MIT student Michael Truell sat in the cafe of the Computer History Museum, staring at a programming test problem. It was a problem that should have taken about an hour to complete, but he finished it in less than 10 minutes.
"He completely crushed that problem," recalled tech investor Ali Partovi. Partovi runs a program specifically seeking out the world's best programmers during their undergraduate years. With plenty of time remaining, Partovi asked Truell to give him a programming problem instead. Partovi, a programmer himself who co-founded Code.org, took much longer. When he finished, the paper was a mess; by contrast, the teenager's lines of code were neat and clear.
Now, at 25, Truell is the CEO of Cursor. This AI coding startup has reached a potential $60 billion acquisition agreement with Elon Musk's SpaceX. This slender, frizzy-haired redhead is described by colleagues as quiet and friendly. Unlike some young founders who love to post their latest revenue numbers or workout results, he prefers to immerse himself in coding, almost like a form of meditation, for long stretches. Inside Cursor, everyone knows he did not pay himself a salary in the company's early years.
However, beneath his humble exterior, Truell has long harbored ambitions as grand as anyone in Silicon Valley. He has told employees he wants Cursor to become a "generational company." As a teenager, he developed a popular programming game themed around conquering the universe. Fresh out of MIT and starting his company, he and several college classmates challenged Microsoft in the code editor space and ultimately won. At Cursor, he fosters a work culture of extremely high intensity: to find the perfect fit, the company puts candidates through complex, unpaid "working trials" that can sometimes last for weeks.
Becoming one of the fastest-growing startups in the tech industry has not been easy. Cursor has constantly dealt with a nuanced and tense relationship with Anthropic. Anthropic was Cursor's primary AI model supplier until this frontier AI lab launched its own extremely popular coding tool. After Claude posed an existential threat to the company, Truell declared an internal state of emergency. Subsequently, he tied Cursor's fate increasingly closely to Musk's just-publicly-listed SpaceX. SpaceX is desperate to win the AI race and controls billions of dollars' worth of computing power.
Cursor declined to comment for this article. Anthropic and SpaceX did not respond to requests for comment.
Truell now faces his biggest test yet: can the partnership with Musk succeed? Regardless of the outcome, the Cursor CEO is already planning to ensure his company secures its place in the history of computing.
Truell grew up in New York, the son of journalists. He was a gifted programmer from a young age and began promoting coding early. At 15, while still a student at the elite private school Horace Mann, he co-developed a programming game called Halite. The game taught foundational programming concepts by having players conquer territory on a grid. The project attracted thousands of users, mostly high school and college students who had never coded before, and earned him a $10,000 prize from a top mathematics society.
Upon entering MIT, he double-majored in computer science and mathematics and began conceptualizing startup ideas. Claire Shorall, who helped run a startup boot camp Truell attended during his undergraduate years, said she was impressed by his curiosity and humility. At the time, he was making cold calls to doctors across the US to validate an early startup idea. Truell asked Shorall to sit next to him and critique his communication skills as they made calls on a landline phone. That project, an attempted competitor to Zocdoc, ultimately didn't succeed, but Shorall could already see that Truell possessed more than just raw programming ability.
"I gave him some advice — but it was clear he already had that capability," she said.
After graduating in 2022, Truell co-founded Anysphere with MIT classmates Sualeh Asif, Arvid Lunnemark, and Aman Sanger. At the time, it was a code editing platform. In less than 12 months, they reached $1 million in recurring revenue by creating a better alternative to Microsoft's open-source code editor, VS Code.
"In the coming years, our mission is to make coding an order of magnitude faster, while also making it more fun and more creative," Truell told TechCrunch at the time.
The Controversies Behind Hypergrowth: Unpaid Trials, Extreme Hiring, and Model Dependency
To achieve this mission, Cursor officially launched in March 2023 and grew rapidly. It quickly gained popularity among developers and businesses eager to dramatically boost their output. In 2024, Cursor disclosed it had over 40,000 customers and set the ambitious goal of building a "magical" tool that could one day write all the world's software.
"Something beautiful is happening to code," the company wrote in a blog post at the time.
By the end of 2025, Cursor was adopted by millions of developers. The company announced its revenue had grown 10 times in less than a year, exceeding $1 billion.
Cursor's growth was incredibly intense, and this intensity was reflected in its hiring process. Four former employees said Truell was deeply involved in recruitment. He would often search for top engineers on GitHub and X, then invite candidates to Cursor's expansive, campus-like headquarters in San Francisco for multi-day "working trials."
During these trials, candidates did almost everything regular employees would do: have lunch with the team, sit at a desk using company computers, and complete projects based on a frozen version of Cursor's codebase.
"It really does give us an enormous amount of signal on whether someone has the raw technical ability to be successful in our environment," Truell said on a podcast in November last year.
However, some have criticized these trials for being unpaid. A person who claimed to have interviewed at Cursor posted on Reddit, denouncing the process as "exploitative and unethical."
A former employee recalled receiving a late-night email asking them to be at the Cursor office by 9 a.m. the next day to complete a series of programming projects. In another instance, the former employee stated that Cursor put a management-level candidate through a month-long working trial. During this period, the person met almost every member of the team, but the company eventually decided not to hire them.
"After the month was up, their attitude was: 'We can probably find someone even better than this candidate,'" the former employee said. They believed this illustrated both Cursor's incredibly high bar for new joiners and the effectiveness of this screening mechanism.
Despite its phenomenal growth, Cursor's executives long worried that the company had become overly attached to and dependent on a single AI supplier. Employees often used one word to describe the relationship between Cursor and Anthropic: strange.
The two companies were highly interdependent. Cursor relied heavily on Anthropic's AI model to power its coding tool. At the same time, Anthropic benefited immensely from Cursor's explosive growth. According to an employee familiar with the numbers, Cursor accounted for roughly 40% to 50% of Anthropic's revenue at an early stage.
"Both sides were somewhat aware that they needed each other. We brought a lot of revenue to Anthropic," said another employee. "But at the same time, Anthropic also had its own competing product."
Before launching its heavyweight code editor, Claude Code, Anthropic executives privately assured Cursor management that the product was more of a research project than a major commercial push. A person familiar with the matter said communications regarding this had occurred. However, Claude Code quickly became popular among developers. By February 2026, its annualized revenue had grown to $2.5 billion, about $500 million more than Cursor's annualized revenue at the time. This figure was first reported by Bloomberg. Developers also began posting that they were cancelling Cursor in favor of Claude Code.
Prior to this, concerns among Cursor executives about the company's dependence on Anthropic were already high. One reason was that Anthropic had previously cut service to Windsurf, a competing AI coding startup, during its acquisition talks with OpenAI.
On January 5th, Truell held what one employee called an "emergency meeting" with the entire company, announcing that Cursor needed to build its own AI model. Two employees said the message was very clear: We must ensure we aren't left behind. The company would cancel all non-essential meetings, and staff might be temporarily reassigned to work with different teams that week. "We must stay flexible and adapt quickly," was the message.
Following the meeting, Cursor began a lengthy pricing analysis comparing Claude Code and OpenAI's Codex, while also holding meetings to reassure its largest clients. Executives concluded that Cursor must double down on its self-developed model to reduce dependence on frontier model labs and gain more control over pricing.
Although Cursor declined to comment for this article, Truell described the relationship with Anthropic in a recent interview as a "deep partnership," adding, "we are very grateful for it."
Cursor's Biggest Gamble: Breaking Away from Anthropic, Hitching Up with Musk
Subsequently, Cursor launched Composer, its own set of coding-oriented models. Composer is built upon an open-source model from the Chinese AI lab Moonshot. It has already begun gaining traction among developers. Cursor stated that in the Composer 2.5 model released in May, over 85% of it came from Cursor's own work – meaning the underlying Moonshot model constitutes only a small portion of the final product.
"Composer has received extremely positive feedback," said Cursor engineer Lucas Garza. This is mainly attributed to its low price and high speed, especially against the backdrop of rising AI costs and pressure on engineering budgets at tech companies.
Cursor's newest tools are also generating fresh buzz. On a hot afternoon in June, Cafe Cursor, a pop-up coffee shop run by Cursor in San Francisco's touristy North Beach district, might have been the busiest cafe on the block. The shop handed out free lattes and $50 credits to enthusiastic entrepreneurs. Many praised Cursor for boosting their productivity.
Tech professionals lounging at Cafe Cursor, a pop-up coffee shop owned by Cursor, earlier this month. Charles Rollet/Business Insider
Aneesh Dharani, founder of an AI flashcard startup, said that despite lacking a software engineering background, Cursor helped him actually build his product. Another founder, Devon Lim, mentioned he used Cursor to replace an outsourced engineer who had suddenly "gone dark" and stopped working on his sales startup.
But building and running a top-tier AI model is extremely expensive, and Cursor itself doesn't have enough chips to do it entirely independently. So, this spring, Truell and his company found another founder with "interstellar-level ambitions" to fill this gap: Elon Musk.
On April 21st, in his typically concise style, Truell announced a new partnership on X.
"Excited to partner with the SpaceX team to scale Composer. This is a major step towards making it the best place for AI programming," he wrote.
On the surface, the deal benefits both sides. Cursor gains access to SpaceX's vast computing power resources, including Colossus – a supercomputer powered by hundreds of thousands of top-tier Nvidia AI chips. In return, SpaceX's Grok gets a boost in the AI coding race. A contractor for xAI previously told Business Insider that Grok is not the "best at coding."
What Truell didn't mention in that X post was that a much larger development was already underway: he had agreed that SpaceX could potentially acquire Cursor for $60 billion later this year.
This news surprised many Cursor employees, as Truell had previously always talked about building Cursor for the long haul. One former employee said that whenever someone mentioned an acquisition, Truell would say, "This is a huge risk, or rather a huge bet, that we're taking."
The structure of this deal is also unusual. According to SpaceX's S-1 filing last month, if either party decides not to proceed with the deal, SpaceX will pay Cursor a $1.5 billion termination fee and provide an additional $8.5 billion worth of free computing power.
Ali Partovi, one of Cursor's earliest investors, is not privy to the internal details of the deal. He said that while many founders claim they will never sell their company, in reality, they fall on a spectrum. Partovi believes Truell leans closer to the end favoring independence.
"His ambition, confidence, and drive would push him more towards staying independent," Partovi said.
For now, Cursor remains independent and continues its rapid growth. According to Forbes, its revenue doubled in three months to reach $4 billion.
Some early progress has already surfaced. Musk posted on X that a recent version of Grok had significantly improved after being trained on a "massive amount" of Cursor data. Both Grok and Composer are climbing the much-watched AI model rankings – i.e., benchmarks – though they haven't reached the top yet.
For Musk, the goal is clear: his AI will be "very good" regardless.
"Whether it will become the very best remains to be seen, but I will never give up," he wrote on X. "Never."
For Cursor, the ultimate goal is less clear, as the structure of the deal with SpaceX remains quite open-ended.
Truell said in a recent interview that Cursor now employs 700 people and serves 60% of the Fortune 500. He also said the company can now be compared to many of the world's largest public software companies.
"It's a bit crazy," he said, "and we are very aware of how special this is – how historically unprecedented it is."
