SpaceX acquired the AI coding tool Cursor for $60 billion. Cursor’s annualized revenue grew from $100 million to $4 billion within two years, setting a record for B2B software growth, yet it continued to incur losses—user payments were insufficient to cover the costs of large model APIs. Cursor lacks moats such as proprietary data, network effects, or switching costs; the faster it grows, the more dependent it becomes on upstream model providers. Its programming behavior data is not exclusive—competitors like Claude Code and Copilot are also collecting similar data. This acquisition is interpreted as Musk’s bet to secure developer data and channels for Grok. In contrast, Thomson Reuters withstood the impact of the Claude Crash due to its entrenched legal database壁垒. Cursor’s case reveals a structural challenge in the AI application layer: if a product lacks assets that model companies cannot replicate, rapid growth merely paves the way for others.

Article author and source: WeChat public account "Wang Zhiyuan" (ID: Z201440)

Cursor, died from growth

On its fourth day as a public company, SpaceX's first move: spending $60 billion to acquire Cursor. As the fastest-growing SaaS company, why did it ultimately fail to sustain itself?

01

Cursor is fundamentally an AI-powered programming editor where developers write code, and it leverages large models to provide code completions, debugging, and refactoring.

Think of it as a super-smart assistant built into your coding workspace; over the past two years, it has been the fastest-growing tool among developers worldwide, and more than half of the Fortune 500 companies are using it.

I looked up its revenue curve.

In January 2025, annual revenue just surpassed $100 million. By June of this year, it reached $4 billion—less than two years from zero to $4 billion. Neither Slack nor Zoom ever achieved this pace. No B2B software company in human history has.

Then the company suffered a total loss.

Individual developer accounts have remained negative-margin to this day; to put it bluntly, the monthly membership fees paid by developers don’t even cover the cost of running models for Cursor. Enterprise subscriptions are somewhat better, only barely turning profitable this past April. Overall, the higher the revenue, the larger the deficit.

A team of fewer than 200 people manages an annualized transaction volume of $4 billion.

Sounds like an efficiency miracle, right? But look at it another way—it just shows how lightweight this product is. How lightweight? Every dollar earned doesn’t belong to the company.

Look, the growth flywheel of Cursor, in one sentence:

Users pay a monthly fee; Cursor uses this money to purchase large model APIs from Anthropic and OpenAI, packages them, and delivers them to you. With each additional user, Cursor places another order with its suppliers—revenue increases, but so do dependencies, proportionally.

Why can traditional SaaS build a moat? Because of three things, and none can be avoided.

First, data: With Salesforce, after using it for a year, your customer data and sales records are all stored on their cloud—you can't leave. What about Cursor? Your code is on your local machine; you can walk away anytime, leaving nothing behind.

Two, network effects; everyone at Slack uses it, so once one person is in, the entire team’s communication habits are locked in. Cursor is your personal tool—what editor your colleagues use doesn’t matter to you.

Third, switching costs: switching to another traditional SaaS solution might take a small company one to two weeks and a large company several months. Switching from Cursor to Claude Code takes five minutes—zero friction.

Cursor's growth is "flow-through growth"—money passes through it without staying.

Some might argue that Cursor isn't entirely without accumulation: millions of developers write code on it every day, and the data from which snippets are accepted, modified, or completely deleted holds value.

Elon Musk himself said that the new version of Grok showed significantly improved performance after using Cursor's data.

That's true. The issue is that this data isn't unique to Cursor. Claude Code is also accumulating the same type of programming behavior data daily, as are Copilot and Codex.

Moreover, in terms of scale, Claude Code's developer base is growing faster than Cursor's; what you have, your competitors also have—and possibly even more. This is known as industry public resources.

What constitutes a true data barrier? Something that only you have and others cannot access. Cursor’s programming behavior data clearly does not fall into this category.

Before SpaceX stepped in, Cursor was negotiating a standalone funding round at a $50 billion valuation, planning to raise $2 billion, but the deal did not proceed.

Someone close to the deal said that $2 billion is not enough to help Cursor break even; a company with $4 billion in annual revenue shouldn't struggle to survive on a $2 billion funding round.

Look closely at these numbers—they point to a growing structural challenge in the AI application layer: the faster the growth, the more money is burned, and the deeper the dependence on suppliers. The faster you run, the more you can’t do without others.

In short, Cursor is trapped by its own growth model.

02

The company trapped by its own growth was eventually taken over—by Musk—but his own wounds were no small matter.

In May, SpaceX's prospectus disclosed xAI's financial data, and I did the math:

In the first quarter of this year, xAI generated $818 million in revenue and incurred an operating loss of $2.47 billion, resulting in a net loss of nearly $2.5 billion for the quarter.

This isn't over—back in March, Musk himself said: "xAI's programming tools are behind Anthropic and OpenAI"; he said this in front of everyone.

Does xAI lack computing power? No. The Colossus data center in Tennessee is equipped with over 220,000 NVIDIA GPUs, placing its computing capacity among the largest in the world. What it lacks is an outlet to transform that computing power into products and, ultimately, revenue.

For example, having a mine but no miners.

I looked through SpaceX's S-1 filing and noticed something.

Anthropic pays SpaceX $1.25 billion per month to rent computing power—why? To train Claude. The contract also includes a 90-day termination clause, allowing either party to end the agreement with three months’ notice.

Including Google’s lease, SpaceX earns approximately $26 billion annually just from leasing computing power. SpaceX’s total revenue for 2025 is $18.7 billion—meaning revenue from leasing computing power alone exceeds its entire annual earnings.

An annualized revenue of $26 billion—Anthropic wants to pull out, and three months ago, with just a heads-up, it vanished. This money is nothing more than a promissory note that anyone can reclaim at any time.

Musk must buy—it’s essential for him to establish a truly independent AI revenue stream beyond his compute leases; otherwise, one day Anthropic might build its own data center and choose not to renew, causing that $26 billion to vanish overnight.

Consider this relationship.

SpaceX's most profitable AI business is renting computing power to its competitors; Anthropic uses the rented computing power to train Claude, and Claude Code is the main force eroding Cursor's market share.

SpaceX is indirectly handing weapons to its competitors while also seeking to acquire a company that has been wounded by those same weapons.

So, looking again at these $6 billion, I tend to believe that Musk wasn't buying Cursor's $400 million in annual revenue or its 200-person team.

He is buying a possibility: giving Grok direct access to a pool of developers and real programming behavior data.

He said on X that the new version of Grok showed significantly improved performance after using "a large amount of Cursor data."

This logic sounds familiar—in 2018, Microsoft spent $7.5 billion to acquire GitHub, following the same strategy: buying access to the developer ecosystem.

After the acquisition, GitHub continued to operate independently, and several years later it spawned Copilot, enabling Microsoft to secure the leading position in the AI programming market.

I calculated the prices of the two transactions:

When Microsoft acquired GitHub, GitHub's annual revenue was approximately $300 million, making the $7.5 billion price tag roughly 25 times revenue; Cursor's annualized revenue is $4 billion, making the $60 billion valuation about 15 times revenue. Looking solely at multiples, Cursor is actually cheaper than GitHub was.

The figure of 60 billion is not unreasonable in itself. The issue lies with the quality of the underlying 4 billion. GitHub’s revenue is sticky—user repositories, collaboration relationships, and commit histories are all hosted there, resulting in extremely high migration costs for users.

Cursor's $400 million is in transaction volume; with the same multiple, the asset quality is entirely different.

Microsoft did one very restrained thing back then: it didn’t turn GitHub into Microsoft’s private backyard. To this day, GitHub still hosts projects in all languages and on all platforms—including code from Microsoft’s competitors. The Linux kernel is still there.

Precisely because it didn’t shut the door completely, the developer ecosystem has continued to grow. Copilot’s later success relied on this restraint.

Can Musk demonstrate the same restraint?

Look at what happened after the xAI merger—all 11 co-founders left within two months. This outcome doesn’t inspire optimism.

SpaceX has a cutthroat competitive relationship with Anthropic and OpenAI. After Cursor was acquired, it’s unclear whether competitors will still be willing to provide it with model APIs—no one can say for sure.

I reviewed JetBrains' April developer survey, which found that 70% of engineers use two to four AI programming tools simultaneously.

Cursor handles daily tasks, Claude Code manages complex tasks, and Copilot provides quick completions. Developers have near-zero loyalty to any single tool.

Even if SpaceX successfully integrates Grok into Cursor, developers can still freely use Claude Code alongside it—you can't lock them out.

$60 billion is essentially betting on three "what-ifs": what if Grok catches up using Cursor’s data, what if developers can be locked in, and what if a moat finally emerges.

The contract includes hundreds of billions of dollars in breakup fees; signing at this level shows both parties understand that none of these "what-ifs" are certain.

03

Logically, Cursor’s story could have ended here. In fact, the matter was revisited multiple times, making it a useful point of comparison.

On February 3rd, Anthropic released a legal plugin called Claude Cowork. What’s it for? It helps company legal teams review contracts, assess compliance, and draft reports—sounds unremarkable, right?

On that day, Thomson Reuters' stock fell 16%, the largest single-day decline in the company's history; RELX fell 14%, and Wolters Kluwer fell 13%.

The entire legal tech sector was wiped out in a single day. People in the industry named this day the "Claude Crash."

Pay attention to the structure here: Anthropic is doing exactly what it did in the programming space—first providing foundational model capabilities, allowing a group of application-layer companies to emerge, and then, once the赛道 is well-established, inserting its own plugin.

In the programming niche, Cursor spent two years achieving $4 billion in ARR before being acquired. In the legal sector, Thomson Reuters lost a significant portion of its market value in a single day.

A media outlet called Artificial Lawyer published an article that day, stating that this panic was completely unfounded.

Thomson Reuters is built on a foundation of legal data—a fortress. It holds Westlaw, LexisNexis, and decades of accumulated case law, statutes, and compliance documents, all stored in its own proprietary database. These resources are simply not available on the public internet and cannot be acquired by Anthropic in the short term.

I revisited Morningstar analyst’s breakdown and found it even more interesting.

Thomson Reuters suffered the biggest decline because 45% of its profits come from its legal segment—the blade struck right at its thickest area. RELX and Wolters Kluwer, by contrast, derive only 10% to 13% of their profits from legal services, so their fundamentals were far less affected.

Both get hit, but some only bruise the skin and flesh, while others break the bones. Ultimately, the difference lies in how much you can handle.

After falling, Thomson Reuters is slowly recovering; the data moat has indeed held strong.

This is precisely the most fundamental distinction between it and Cursor.

Thomson Reuters has something that model companies cannot take: its database. What about Cursor? It has nothing. Its growth does not accumulate any assets that suppliers cannot bypass.

Putting these two cases together points to a emerging pattern.

Model companies eroding the application layer is no longer hypothetical—it’s happening simultaneously across multiple sectors. In the programming sector, the full journey has been completed: from dependence to erosion to acquisition. In the legal sector, we’re midway through the process; the market has already experienced an initial wave of panic, and companies with moats have temporarily held their ground.

Who’s next?

Every category of application built on large model APIs—writing tools, design tools, customer service tools—is within range.

The only difference is this: Does your product include something that the model company cannot create, take away, or bypass?

This isn't necessarily data.

Palantir also operates at the application layer; its moat lies in being deeply embedded into the workflows of government and military clients—security certifications, compliance credentials, and years of integrated deployments are not something a model company can disrupt with a simple plugin.

A data fortress is a barrier; an embedded barrier is another—both are essentially the same: you possess something others cannot bypass.

If you are Thomson Reuters, you can withstand the first wave; if not, you are Cursor. No matter how fast you grow, you're just paving the way for someone else.

A $60 billion transaction has been finalized. It also offers a lesson to domestic companies in this industry: What exactly are you building your growth upon?

Reference materials

[1]. SpaceX S-1 Prospectus, May 2026; [2]. SpaceX SEC 8-K Acquisition Filing, June 16, 2026; [3]. TechCrunch, Series of Reports on Cursor's Funding and Financial Status

[4]. Morningstar, xAI financial analysis, and Thomson Reuters legal sector breakdown; [5]. JetBrains 2026 Global Developer Survey; [6]. Artificial Lawyer, Claude Crash review, February 4, 2026; [7]. SiliconANGLE, The Next Web, SpaceX compute leasing coverage