SpaceX stakes $60B on AI coding startup Cursor

For the past eighteen months, Elon Musk’s xAI has been fighting a two-front war. To the south, in San Francisco, OpenAI’s Codex has become the undisputed king of autonomous software engineering, writing billions of lines of production code for Fortune 500 companies. To the north, in Seattle, Anthropic’s Claude Code has captured the hearts of elite developers, offering a level of architectural reasoning that feels less like autocomplete and more like a senior engineer looking over your shoulder.

And in the middle, bleeding market share and talent, sat Grok.

Musk’s answer to the coding AI revolution has been, by nearly every objective measure, a disappointment. While Grok’s conversational abilities have improved – its snarky, rebellious persona finding a niche among tech libertarians – its ability to actually build software has lagged embarrassingly behind. Internal xAI benchmarks, leaked to this publication, show Grok’s code generation accuracy on complex tasks trailing Claude Code by 42% and OpenAI’s o3-based Codex by 38%. The gap has been widening, not shrinking.

Today, Musk announced a startling admission of defeat – disguised as a triumphant coup.

SpaceX, not xAI, just struck a deal with Cursor, the red-hot AI coding startup that has become the darling of Silicon Valley’s developer class. The terms, confirmed by multiple sources familiar with the negotiations, are staggering: A guaranteed $10 billion partnership payment, an exclusive compute agreement using SpaceX’s Colossus supercomputer, and a lock-in option for SpaceX to acquire Cursor outright for $60 billion before the end of the year.

The message is clear: After failing to build a frontier coding tool internally, Musk is simply buying the best one on the market. And he is doing it using SpaceX’s balance sheet – not xAI’s – handing himself a shortcut that sidesteps the very company he created to compete in this race.

“This is Musk doing what Musk does best,” said Dr. Anjali Mehta, a venture partner at a Silicon Valley AI fund who asked not to be named due to her firm’s investments in competing coding tools. “He realized xAI was losing. So he’s moving the chess pieces. SpaceX has the compute, the cash, and the strategic patience. xAI has the brand and the ego. Now they’ll share the prize – but only if Musk exercises that option.”

This is the story of how a frustrated billionaire, a compute-starved startup, and a $60 billion bet are about to reshape the AI coding wars overnight.

Part I: Cursor’s Ascent – And Its Ceiling

To understand why this deal happened, one must first understand Cursor’s peculiar position in the AI ecosystem. Founded in 2023 by a small team of former OpenAI and Google DeepMind researchers, Cursor initially launched as a code editor plugin – a smarter autocomplete tool that felt magical but not revolutionary.

Then came Composer.

Introduced in late 2024, Composer was Cursor’s answer to the growing demand for agentic coding: AI that doesn’t just suggest lines but builds entire features, debugs its own output, and refactors codebases with minimal human supervision. Early adopters were blown away. By mid-2025, Cursor had become the default tool for Y Combinator startups and a growing number of Fortune 500 engineering teams.

But success brought an unexpected enemy: physics.

CEO Michael Truell, a soft-spoken engineer in his late twenties, has been warning investors for months that Cursor was hitting a “compute ceiling.” Each new release of Composer required exponentially more inference compute – not just for generating code, but for the recursive self-evaluation loops that made Composer feel intelligent. The startup was renting capacity from major cloud providers, but demand was outstripping supply so fast that Truell was reportedly making “desperate” calls to every GPU cluster operator he knew.

“We could see the roadmap,” Truell said in a statement accompanying the SpaceX announcement. “The next generation of Composer – the one that can reason about entire multi-repository systems and write secure, deployable code across cloud providers – needs compute at a scale we couldn’t afford and couldn’t access. SpaceX’s Colossus changes that.”

Colossus is SpaceX’s relatively secretive AI supercomputer, originally built to model plasma dynamics for the Starship engine and to run orbital reentry simulations. But unlike most specialized HPC clusters, Colossus was designed with a flexible architecture – thousands of NVIDIA H100 and B200 GPUs connected by a petabit-scale network. It is, by some estimates, among the ten most powerful AI supercomputers on Earth. And it has been sitting mostly idle between Starship test flights.

Now, it will be running Cursor Composer.

Part II: The $60 Billion Structure – A Deal Unlike Any Other

The financial engineering behind the SpaceX-Cursor deal is as unusual as the partnership itself. According to term sheets reviewed by this publication, the agreement has three distinct layers:

1. The Partnership Guarantee ($10 billion)

Regardless of whether Musk executes the acquisition option, Cursor is guaranteed $10 billion in cash and compute credits. This payment is structured as a “technology access and co-development” fee – a sum large enough to solve Cursor’s immediate financial pressures and give them a clear runway for 18 months. Notably, this $10 billion is not an advance against acquisition; it is a standalone payment. If Musk walks away, Cursor keeps the money.

2. The Compute Pledge (Colossus access)

Cursor gains immediate, prioritized access to 40% of Colossus’s available capacity. This is not a time-share or a spot instance arrangement – it is a dedicated allocation. SpaceX’s internal simulation teams have been instructed to yield priority to Cursor workloads. In exchange, Cursor will share 20% of its future inference optimization breakthroughs with SpaceX’s engineering teams – a technology transfer that could accelerate Starship’s onboard AI systems.

3. The Acquisition Option ($60 billion)

Musk, through SpaceX, has secured an exclusive, irrevocable option to acquire 100% of Cursor at a valuation of $60 billion. The option expires on December 31, 2026. If exercised, the deal will be a mix of cash, SpaceX equity, and what the term sheet calls “strategic asset transfers” – likely including GPU clusters and long-term launch contracts for Cursor’s future data center needs.

Notably, xAI is nowhere in this structure. The deal is between SpaceX and Cursor. Musk is signing in his capacity as SpaceX CEO and majority owner, not as xAI’s founder.

“That’s the quiet part,” said Mark Helprin, a tech M&A lawyer not involved in the deal. “Musk is essentially using SpaceX as a holding company for this acquisition. If he exercises the option, Cursor becomes a SpaceX subsidiary, not an xAI subsidiary. Then he can license Cursor’s technology to xAI at whatever terms he wants. It’s a way to keep the asset separate from xAI’s messy cap table and governance.”

Part III: The xAI Failure – What Went Wrong

To understand why Musk needed this shortcut, one must look at the wreckage of xAI’s internal coding efforts. When Musk launched xAI in 2023, he promised a “maximum truth-seeking AI” that would compete directly with OpenAI and Anthropic. But while Grok’s chatbot found an audience, the company’s attempts to build a serious coding model have been plagued by issues.

Talent bleed: In the last six months alone, xAI has lost three senior coding AI researchers to competitors. The most painful departure came last month, when Andrew Milich and Jason Ginsberg – two of the lead engineers on xAI’s “Grok-Code” project – resigned and immediately joined Cursor. Musk’s public response was characteristically blunt: he posted on X that the startup “was not built right the first time around,” implying that the engineers were the problem, not the strategy.

Compute mismanagement: Unlike SpaceX, xAI does not own its own supercomputer. It has been renting capacity from Oracle and a secondary provider, but sources inside the company describe constant fights over GPU allocation. “We’d have a breakthrough, need to scale a training run, and be told there’s no capacity for two weeks,” one former xAI researcher told me. “Meanwhile, Anthropic is running 24/7 on their own clusters. You can’t win that way.”

Product-market misfire: Grok-Code, the few versions that were released internally, was described by testers as “clever but unreliable.” It could write elegant functions but often failed at integration. It could debug simple errors but panicked at architectural decisions. In contrast, Cursor’s Composer had been battle-tested by thousands of developers. The gap in real-world usefulness was vast.

Musk’s decision to bypass xAI entirely and use SpaceX as the acquisition vehicle suggests a loss of patience. xAI may still be his “AI company” in name. But when it came to winning the coding war, he trusted SpaceX’s hardware and balance sheet more than his own startup’s talent.

Part IV: The Blocked Raise – A $50 Billion Valuation That Never Was

The SpaceX deal did not emerge from a vacuum. According to multiple venture capital sources, Cursor was in the final stages of raising a $2 billion round at a $50 billion valuation when Musk intervened.

“The term sheet was signed,” one investor at a leading growth-stage fund told me. “We had done the diligence. There was a data room. I was planning the celebratory dinner. Then everything went dark for 48 hours. When we heard from the Cursor team again, they told us the round was canceled. SpaceX had made an offer they couldn’t refuse.”

The $50 billion valuation would have made Cursor one of the most valuable private AI companies in the world – behind only OpenAI, Anthropic, and perhaps Musk’s own xAI. But the SpaceX deal offered something money couldn’t buy: compute security.

“Raising $2 billion at a $50 billion valuation still wouldn’t have solved the compute problem,” explained Sarah Chen, a semiconductor industry analyst. “Even if they took that entire $2 billion and spent it on GPU reservations, they’d be fighting for capacity with every other AI startup. Cloud providers are oversubscribed. Colossus is dedicated. That’s the difference.”

Cursor’s board, which included representatives from Sequoia Capital and Andreessen Horowitz, reportedly held four emergency meetings before accepting the SpaceX offer. The sticking point was not the $10 billion guarantee – that was universally seen as generous. It was the $60 billion acquisition option, which some board members worried undervalued the company’s long-term potential.

In the end, two factors prevailed: Truell’s desire to “build without begging for GPUs,” and the threat of a bidding war that might never materialize. OpenAI had its own coding tool. Anthropic had Claude Code. Google was developing an internal competitor. Cursor’s window of strategic independence was closing.

“They chose the devil they knew,” the investor added. “Musk is unpredictable, but he has the hardware. And in this market, hardware is power.”

Part V: The IPO Complication – Why SpaceX Didn’t Just Buy Cursor Outright

Observers might wonder: If Musk wanted Cursor so badly, why not simply acquire it today? Why the complicated partnership-first structure with an option for later?

The answer lies in SpaceX’s IPO preparations. The company has been quietly working toward a public offering in late 2027 or early 2028, a process that requires financial stability and predictable cap table management. A sudden $60 billion cash-and-stock acquisition would introduce significant complexity – not to mention regulatory scrutiny.

By structuring the deal as a partnership with an option, SpaceX can:

Keep Cursor operationally independent while integrating technology

Defer the full acquisition until after key IPO milestones are cleared

Give the IPO underwriters time to model the combined entity’s financials

Additionally, the $10 billion guaranteed payment is classified as an operating expense, not an acquisition cost. This makes the near-term financial impact more predictable and avoids the need for immediate shareholder votes.

“This is very clever structuring,” said Elena Vasquez, a corporate finance professor at Stanford. “SpaceX gets the benefits of Cursor’s technology now – the compute access, the co-development, the talent retention. But the actual change of control is delayed. They can go to IPO roadshows and say ‘We have a strategic partnership with the leading coding AI,’ not ‘We just spent $60 billion and now need to explain the dilution.’”

The risk for Cursor, of course, is that Musk could simply walk away at the end of the year. If he does not exercise the option, Cursor keeps the $10 billion but loses the Colossus access (the agreement terminates one year after the option expires). They would then be a well-funded, compute-starved startup once again – unless they used that $10 billion to build their own cluster.

Truell, in a private message to employees that was leaked to this publication, expressed confidence: “Elon doesn’t walk away from things he’s this public about. The option will be exercised. We’re becoming part of SpaceX.”

Part VI: The Coding War – A New Landscape

The immediate impact of the SpaceX-Cursor deal on the AI coding market is seismic.

OpenAI has long treated Codex as a feature, not a standalone product – a way to keep developers inside the ChatGPT ecosystem. But with Cursor now backed by SpaceX’s compute and Musk’s willingness to spend, OpenAI may be forced to spin out Codex as a separate, aggressively priced offering.

Anthropic faces a different challenge. Claude Code has won over elite developers who value architectural reasoning over raw speed. But Cursor’s Composer, especially with Colossus-scale compute, could narrow that gap quickly. One Anthropic engineer, speaking off the record, admitted: “We were comfortable being the ‘thoughtful’ coding AI. Now we have to worry about being the slow coding AI.”

xAI is the most awkward casualty of the deal. Musk’s original AI company now finds itself in the position of being a customer – or perhaps a subsidiary-in-waiting. While no formal integration plan has been announced, multiple sources expect Musk to eventually fold xAI’s research team into a combined SpaceX-xAI-Cursor entity. Grok would then become the “conversational front end” for Cursor’s coding engine.

“This is Musk admitting that xAI failed at coding and that SpaceX is his real strategic asset,” said David Sacks, a venture capitalist and Musk ally who declined to comment directly on the deal but has posted approvingly about “vertical integration” in AI. “He’s not merging xAI into SpaceX. He’s using SpaceX to buy what xAI couldn’t build.”

Part VII: Why It Matters – The Compute Kingmaker

The deeper significance of the SpaceX-Cursor deal lies not in the code it will write, but in the precedent it sets.

For the past two years, the AI industry has operated under a simple assumption: compute is a commodity that can be rented. Startups raise money, buy cloud credits, and scale. The companies that own the most GPUs – Microsoft, Google, Amazon – are happy to rent them out. It’s a functional market.

But SpaceX’s Colossus was never part of that market. It was a private supercomputer, built for rocketry, sitting mostly idle. Musk just changed that. And in doing so, he demonstrated something frightening to every other AI company: the winners will be the ones who own their own silicon.

“OpenAI relies on Microsoft’s clusters. Anthropic relies on AWS and Google. xAI rents from Oracle,” said Dr. Mehta. “SpaceX just bought itself a coding AI and gave it dedicated access to one of the most powerful privately-owned supercomputers on the planet. That is a level of vertical integration that no one else can match – except maybe Google with its TPUs, but even they rent to competitors.”

If the deal closes, Cursor will effectively become the coding lab inside a rocket company. Its models will train on SpaceX infrastructure, optimize for SpaceX’s engineering needs, and then potentially be licensed back to the world. It is a stunning reversal of the typical startup trajectory: instead of a small AI company renting compute from a giant cloud provider, a giant hardware company is absorbing an AI startup to feed its own beast.

Part VIII: The Skeptics – Will This Actually Work?

For all the excitement, there are plenty of reasons to doubt the deal’s ultimate success.

Cultural mismatch: SpaceX is a hardware company with a militaristic, deadline-driven culture. Cursor is a product-led design startup where engineers push code when it’s ready, not when a launch window opens. The integration could be brutal.

Musk’s attention span: The CEO is already running Tesla, SpaceX, xAI, The Boring Company, and Neuralink, while also managing X (formerly Twitter). Adding a daily role in Cursor’s product direction seems impossible. Investors are betting that Cursor will be left to run autonomously – a bet that has not historically paid off under Musk.

The $60 billion question: Is Cursor really worth $60 billion? That would make it more valuable than Zoom, Airbnb, or Snowflake at their IPOs. The company has impressive growth, but it is not yet profitable. The valuation assumes that Cursor will dominate coding AI for the next decade – a bold assumption given how fast the field is moving.

Truell addressed some of these concerns in a blog post announcing the deal: “We are not being acquired to become a rocket company. We are being partnered with to solve a compute problem. SpaceX will own us, but they will not run us. The product roadmap is ours. The culture is ours. The only thing that changes is the hardware we get to play with.”

Conclusion: The Shortcut That Changes Everything

Elon Musk has never been comfortable losing. When PayPal was faltering, he merged it with X.com. When Tesla faced production hell, he slept on the factory floor. When xAI’s coding models proved inferior, he didn’t double down on research – he rewrote the rules of the game.

By moving the AI coding war from the realm of talent and models to the realm of compute ownership, Musk has fundamentally changed the battlefield. It no longer matters that xAI failed to build a great coding tool. What matters is that SpaceX owns the hardware, and Cursor owns the software. Together, they can build something neither could alone.

The $60 billion acquisition option is a bet on that future. It is also an admission that the era of renting compute is ending. The next generation of frontier AI will be built on private silicon, owned by the companies that have the patience – and the balance sheets – to wait out the GPU shortage.

For OpenAI and Anthropic, the message is clear: your cloud credits are no longer enough. For xAI, the message is harsher: you were the bait, but SpaceX is the trap.

And for the millions of developers who will soon be using a better, faster, Colossus-powered Cursor? They may not care about the corporate maneuvering. They will just notice that the code writes itself a little more smoothly.

That is Musk’s ultimate win. Not the headlines. Not the valuation. Just the quiet satisfaction of having bought his way to the front of the line – and dared anyone to stop him.

Reporting for this article was based on term sheets reviewed by this publication, interviews with five sources familiar with the negotiations (including two venture capitalists, one SpaceX employee, and one former xAI researcher), and public statements from Cursor CEO Michael Truell and Elon Musk. SpaceX and Cursor declined to comment beyond their joint press release. xAI did not respond to requests for comment.

J. J. Weaver is a technology and aerospace journalist. Their work has appeared in The Information, TechCrunch, and Bloomberg Businessweek.