SpaceX Has Option to Buy Cursor for $60B — Here's What It Means

TL;DR

SpaceX disclosed yesterday that it holds the contractual right to acquire Cursor’s parent, Anysphere, for $60 billion later this year. The alternative, if SpaceX walks away from the full buy, is a $10 billion payment for the current partnership work. It’s the biggest M&A tag ever placed on an AI coding tool. The piece that actually matters, though, is the compute: Cursor training its next models on Colossus, the xAI supercluster. Cursor users won’t see anything change this week. The story to watch is whose models ship inside the editor in six months.

What SpaceX and Cursor Actually Announced

The disclosure came in a SpaceX communication to investors on April 21, picked up by TechCrunch, Bloomberg, CNBC, Axios, and Engadget. The shape of the agreement:

• Option to buy: SpaceX can acquire Anysphere (Cursor’s parent) for $60 billion, at an unspecified point later in 2026.
• Fallback: if SpaceX doesn’t exercise the option, it will pay $10 billion to Cursor for “our work together” — compensation for the compute partnership without taking the company.
• Partnership now: Cursor gets access to SpaceX’s Colossus training supercomputer, described by SpaceX as roughly 1 million H100-equivalent GPUs. xAI is already renting compute to Cursor via tens of thousands of chips, per TechCrunch.
• Talent flow: two senior Cursor engineers, Andrew Milich and Jason Ginsberg, joined xAI earlier this year and now report to Elon Musk directly.

Cursor’s own note to users puts the emphasis on compute, not ownership. The phrasing is that the company has been “bottlenecked by compute” and will train future coding models on Colossus. That reads as a roadmap commitment: Cursor has been racing to ship proprietary models (Composer 2 shipped with Kimi K2.5 underneath), and doing that at scale needs chips that don’t exist on the open market in the quantities required.

The $60B Price Tag, In Context

Cursor was in the news last week for something else: a $2 billion primary round at a roughly $50B valuation, led by a16z and Nvidia. The SpaceX number is higher. It’s a takeout price rather than a funding round, and takeout multiples on high-growth private companies typically carry a 10–30% premium over the last round, so $60B on top of a $50B round is in the right zone.

The trajectory tells a story of its own:

Twenty-four times in fifteen months is an unusual trajectory even for AI. It does track with Cursor’s public ARR curve, which crossed $1B in annualized revenue last winter and has been widely reported as running well ahead of that by Q1.

The other number worth noting is $10 billion. That figure is the floor on the deal, what SpaceX has already committed to pay even if the acquisition never happens. Ten billion dollars in pre-IPO SpaceX equity, for a compute-sharing arrangement with a code editor company. You can read that as a benchmark for what AI infrastructure access now costs at this scale.

Why Musk Wants This

Neither xAI nor Cursor has a proprietary frontier model that matches Anthropic’s Claude or OpenAI’s GPT line. Cursor ships code through other people’s model APIs, including a lot of Claude. xAI’s Grok family has improved but doesn’t win the coding benchmarks that matter to developers. This deal is an answer to both gaps at once.

For Cursor, the trade is compute for optionality. The editor has distribution (one of the largest paid developer user bases outside GitHub) but can’t build a frontier model on rental GPUs at market rates. Colossus solves that. For SpaceX/xAI, the trade is distribution for compute. Colossus is underutilized relative to its headline capacity, and plugging 1M H100-equivalents of training into a product that already has paying developer seats is the fastest way to turn raw compute into an AI revenue line.

There’s a second motive. SpaceX is widely expected to IPO later this year or early next, and the Musk-merged SpaceX-xAI entity is being marketed at around $1.25 trillion post-merger. An IPO story that says “we own one of the world’s fastest-growing AI developer products” reads a lot better than “we’re a rocket company with a separate AI lab that’s still behind.” The Cursor option is cheap insurance on the equity story.

What Changes for Cursor Users

Short answer: nothing, this week or next. The editor still talks to Claude and GPT. Your subscription doesn’t change. Pricing doesn’t change. The “Cursor” you opened this morning is the same product it was yesterday.

Medium answer, six months out: Cursor will start routing more traffic to internal models trained on Colossus. Composer 2 already did this with Kimi K2.5 as the base; the next iteration will likely use xAI-trained weights with Cursor’s own post-training. The reason to care is defaults — Cursor sets the default model in the editor, and most users never change it. The day the default flips from Claude Sonnet to a Cursor-on-Colossus model is the day Anthropic loses a very large API customer.

Long answer, post-acquisition (if exercised): Cursor under SpaceX is a different product governance story. Musk runs his companies with strong central control, and xAI’s culture is visibly different from Anysphere’s more traditional startup operating model. Users should expect faster shipping, more opinionated defaults, and tighter coupling to xAI’s roadmap. Whether that’s a net positive for your workflow depends on whether you agreed with Cursor’s previous direction (model-agnostic, lots of BYOK flexibility) or want a more integrated stack.

The practical risk for developers evaluating tools right now: lock-in. If Cursor becomes a front-end for xAI models specifically, tools like Windsurf and GitHub Copilot stay genuinely model-agnostic and might look more attractive for teams that value model choice.

The Anthropic Problem

The part nobody in the announcement is addressing head-on: Cursor has been one of the largest enterprise customers of the Anthropic API. Claude Sonnet is the default model for a huge fraction of Cursor sessions. If the editor moves its default to an xAI-trained model, Anthropic loses a substantial chunk of API revenue, and that revenue has been one of the numbers Anthropic’s internal forecasts lean on heavily.

Anthropic declined to comment on the deal to Bloomberg. Read that as a holding position. The realistic forecast: a three-to-six-month grace period during which Claude remains the Cursor default, followed by a phased migration to proprietary Colossus-trained models for the editor’s core flows (completion, in-editor chat), with Claude retained as a selectable option for the users who actually pay attention.

Anthropic’s hedge is its own first-party product. Claude Code has been gaining share among heavy users who want first-party Anthropic tooling, and an acquisition like this, with Cursor visibly picking a side, probably accelerates that migration. Expect Anthropic to lean harder into Claude Code marketing and enterprise pricing in Q2 and Q3.

Open Questions

A few things the announcement didn’t answer, and none of them are small:

• When does the option expire? “Later this year” is vague. If it’s December, SpaceX has a whole year of optionality on a volatile asset. If it’s July, it’s a near-term commitment.
• What counts as exercising the option? Usually these deals have conditions (regulatory approval, IPO completion, minimum revenue targets). None disclosed.
• Does existing Cursor equity roll? In a $60B takeout, the cap-table math gets loud. Cursor has raised from a16z, Nvidia, Thrive, and others; whether they prefer cash or SpaceX equity at a $1.25T strike will shape the deal’s actual closing probability.
• What happens to Cursor’s BYOK flexibility? The editor currently lets enterprise customers plug in their own Anthropic/OpenAI keys. That’s a product bet that may not survive a SpaceX-owned product org.
• Regulatory exposure? SpaceX has deep ties to US government contracting. Adding a company that processes a significant share of US developer code into the parent entity will get a second look from DoJ/FTC even at Musk-era enforcement levels.

FAQ

Is this an acquisition or just a partnership?

Right now, a partnership with an acquisition option. SpaceX has paid (or will pay) at least $10 billion for the current collaboration (compute-sharing via Colossus), and separately holds the right to buy Anysphere for $60 billion later in 2026. The $60B transaction may or may not happen; the $10B is already committed.

How does the $60 billion compare to Cursor’s last valuation?

Cursor is currently closing a $2 billion primary round at around $50 billion, led by a16z and Nvidia. The $60B option strike is roughly a 20% premium over that round, which is normal for a takeout price relative to a recent primary.

Will Cursor still support Claude and GPT after the deal?

For now, yes. Cursor runs on Anthropic and OpenAI models by default, and there’s no announced change to that. Over the next few quarters, expect the default model to shift toward xAI/Colossus-trained weights, with Claude and GPT retained as selectable options. The day the default flips is when your experience changes.

Is Cursor going to become a Musk product?

If the option is exercised, yes. Cursor would sit under SpaceX/xAI and take direction from that org. Product governance, roadmap prioritization, and cultural posture would shift. Users who dislike that direction have alternatives: Windsurf, GitHub Copilot, and Claude Code all stay outside of Musk’s orbit.

What’s the impact on Anthropic and OpenAI?

Anthropic likely loses a major API customer over the next six to twelve months; Cursor is one of its largest enterprise deployments. OpenAI’s exposure is smaller (GPT-4 family is selectable in Cursor but not the default). Expect Anthropic to push Claude Code harder and reprice enterprise contracts.

Does this kill other AI code editors?

No. It redraws the map. Windsurf, Copilot, Replit, and Zed stay independent and become the obvious alternatives for teams that want model choice. The “big two” AI editors (Cursor and Copilot) are now under very different parent companies with very different incentives.

Sources

• TechCrunch — SpaceX is working with Cursor and has an option to buy the startup for $60B — primary deal reporting with valuation history
• Bloomberg — SpaceX Has Agreement to Acquire Cursor for $60 Billion — independent confirmation
• CNBC — SpaceX says it can buy Cursor later this year for $60 billion or pay $10 billion — deal structure details
• Axios — SpaceX nears deal with AI startup Cursor — earlier reporting on the talks
• Cursor’s own note to customers — Cursor’s framing of the partnership

Bottom Line

This is a compute deal dressed as an acquisition option. SpaceX/xAI has chips and no distribution in developer tooling; Cursor has distribution and can’t afford the compute it needs to catch its own model roadmap up to the frontier. The $60B takeout is real, but the interesting number is $10B. That is what SpaceX is willing to pay just to get Cursor’s models trained on Colossus without owning the company. The takeout is optional; the compute coupling is not.

For you, a developer who uses Cursor daily: nothing changes this month. Watch for two signals over the next two quarters. First, the editor’s default model shifts away from Claude. Second, Cursor announces xAI-branded features. When either happens, the deal isn’t theoretical anymore.