SpaceX Bought Cursor: Your Bill, Your Code, Cost to Leave

On June 16, 2026, SpaceX agreed to buy Anysphere, the company behind Cursor, in an all-stock deal valued at about $60 billion, as reported by CBS News. It is the largest acquisition of a venture-backed startup on record, and it lands days after SpaceX’s own Nasdaq debut. The deal is expected to close in the third quarter of 2026, pending regulatory approval.

If you write code in Cursor every day, the headline is not the price. It is three quieter questions. Does your bill change? Is your code now training a model owned by Elon Musk’s AI company? And if you want out, what does leaving actually cost, not in subscription dollars but in cost per finished task? This is the answer to all three, with the numbers attached.

What actually happened

SpaceX exercised an option to acquire Anysphere outright, paying in stock issued around its public listing. Cursor becomes a wholly owned SpaceX subsidiary once the deal clears regulators, a transaction InfoWorld flagged as raising direct questions for CIOs about vendor concentration and data control. The same coverage notes the strategic logic: Cursor sits on a firehose of developer interaction data (prompts, iteration cycles, architecture decisions), and SpaceX expects access to that data to improve model training, including for xAI’s Grok. A jointly developed model is slated to ship inside Cursor and Grok Build.

That last part is the whole story for a working developer. The acquisition is not a neutral change of logo. It connects the tool you type into to the training pipeline of a frontier-model lab. Whether that matters to you depends entirely on one setting, which we will get to.

First, the boring reassurance, because it is true and it is load-bearing: in the short term, nothing changes.

Your bill: does Cursor get more expensive?

Not yet. As of June 2026, Cursor’s pricing is unchanged: a free Hobby tier, Individual at $20 per month, Teams at $40 per user per month, and custom Enterprise pricing. Your existing subscription keeps working. Claude, GPT, and Gemini still run inside the editor. The deal does not close until Q3, and acquired products rarely reprice the week the ink dries.

The honest concern is directional. Acquisition pricing pressure tends to move one way, and SpaceX paid $60 billion for an asset that now has its own model roadmap to feed. When the owner also builds the models, the incentive shifts toward steering usage onto house models and pricing tiers toward enterprise revenue rather than the solo-developer budget that built Cursor’s base. None of that has happened. All of it is worth budgeting for.

For what Cursor costs to actually run today, task by task rather than sticker by sticker, the full breakdown of Composer 2.5’s real cost per task is the number that matters, and it is the same number that makes leaving expensive. Hold that thought.

Your code: is it training Grok now?

Here is where you should pay attention, and here is the exact wording, because precision matters more than alarm.

Cursor’s data-use policy draws a hard line at one setting. With Privacy Mode on, the policy states “Customer Data will not be used for training by Cursor,” and Cursor maintains zero-data-retention agreements with its model providers, so those providers do not store or train on your data either. With Privacy Mode off, the same policy says Cursor “may use and store codebase data, prompts, editor actions, code snippets, and other code data and actions to improve our AI features and train our models.”

Read that closing phrase again: “train our models.” That clause was written when “our” meant Anysphere, a coding-tools startup. After close, “our” means SpaceX. The models your code can train are the same models the deal rationale describes feeding into xAI’s Grok pipeline. The policy text does not need to change for the meaning to change. The owner changed.

How Privacy Mode decides whether your code trains a model

Privacy Mode	What happens to your code
On	Not used for training; zero-data-retention agreements apply with model providers.
Off	Cursor states it may store your code and use it to train its models, which after the acquisition belong to SpaceX and xAI.

What is genuinely unknown is what happens to the zero-data-retention agreements and the training language after SpaceX takes control as the new data controller. Privacy policies update on a change of ownership. The current text is favorable if you enable the setting. Whether it stays that way is the open question, and it is the strongest single reason a security-conscious team would move now rather than wait to find out.

So the data verdict is not “Cursor is spying on you.” It is narrower and more actionable: if you are on an individual plan with Privacy Mode off, your code is in scope for training the models of a frontier AI lab, and you can stop that today with one toggle. If that toggle is not enough comfort given who the owner now is, the question becomes what it costs to leave.

The cost of leaving: ranked by cost per task

This is the part every “Cursor alternatives” listicle gets backwards. They rank by subscription price: Windsurf at $15, GitHub Copilot Pro at $10, Cursor Pro at $20, and so on. Sticker price is the opening bid, not the bill. The number that clears is the cost to finish a representative coding task, thinking tokens and tool-calling loops included, which is the only honest way to price an agentic workload.

By that measure, leaving Cursor is expensive, and not in a way the monthly price reveals.

The reason is Composer, Cursor’s in-house model. On Artificial Analysis’s independent Coding Agent Index, Cursor’s Composer 2.5 finished a task for about $0.07 on its standard tier. The frontier agents you would switch to, run in the harness they ship in, cost dramatically more for the same class of work: the Claude Opus agent inside Claude Code came in at $4.10, and GPT-5.5 inside Codex at $4.82. That is a 10x to 60x jump in cost per task, for capability that sits only three to four points higher on the same index.

Cost to finish one task, by tool

Tool	Cost per task	Multiple of baseline
Cursor (Composer 2.5, standard)	$0.07	1.0x
Cursor (Composer 2.5, Fast)	$0.44	6.3x
Claude Code (Opus)	$4.10	58.6x
Codex (GPT-5.5)	$4.82	68.9x

This is a live instance of the pattern we keep finding: the cheaper-looking option is not always the cheaper one to run, and here it runs in the opposite direction from the usual warning. People assume the indie tool is the budget choice and the enterprise-owned one the premium. On cost per finished task, Cursor with Composer is the budget choice by a wide margin, precisely because it built its own model. Leaving it means giving that up.

So the realistic alternatives sort into two buckets, and the bucket you land in decides the bill:

Same-shape IDE tools. Windsurf is the closest editor-for-editor swap, a VS Code fork with multi-file editing that maps onto Cursor’s Composer workflow, though it is itself now owned by Cognition and its pricing lists an individual plan at $20 per month, the same as Cursor Pro. (That the nearest alternative was also just acquired is its own comment on this market.) The catch is that none of these ships a sub-$0.10-per-task in-house model the way Cursor does; once you route work to Claude or GPT, you inherit frontier per-task economics regardless of the monthly sticker.

Terminal and frontier agents. Claude Code and OpenAI’s Codex are the power-user exits, and they are where the $4-5-per-task figures live. They buy you capability and independence from SpaceX’s ownership, at a real and recurring cost-per-task premium. For the agent-by-agent capability and cost picture, the 2026 coding-agent landscape and our Copilot AI-credits breakdown lay out the trade in detail, and there is a fuller list of Gemini CLI alternatives ranked by cost per task for the terminal side.

Model your own numbers before you move. The cost-per-task calculator takes your token volume and loop count and returns the monthly figure for each tool, which is the only version of this comparison that reflects your workload rather than a benchmark’s.

Should you switch?

Decide on the data question, not the deal headline. The economics of staying are still strong, so a switch driven by “SpaceX scares me” alone costs you 10x to 60x more per task for a capability gain of a few points. That is a bad trade on numbers.

The trade flips when the data question is a hard constraint rather than a worry.

The reason to leave Cursor is not the price. It is whether you can accept your editor's owner being a competitor's parent company.

Solo developers and small teams on public or low-sensitivity code: enable Privacy Mode today and stay. The cost-per-task advantage is real money over a year, and Privacy Mode on is a credible answer to the training concern as the policy reads now. Revisit if the post-close policy weakens.

Startups with proprietary code that is itself the product: this is the genuine decision. If your codebase is the moat, “may train our models” under a new owner who builds competing models is a board-level question, not a settings question. Enabling Privacy Mode buys you time; a managed Teams/Enterprise plan with enforced privacy controls buys you more; moving to a tool whose owner is not a model lab buys you out of the question entirely, at the per-task premium above.

Enterprises: you likely already run Teams or Enterprise with Privacy Mode enforced and zero-data-retention in place, so your exposure is lower today. Your action item is contractual, not technical: get clarity, in writing, on what the data-handling terms become after a change of control, and treat the answer as a renewal input.

The bottom line

The SpaceX-Cursor deal is being covered as a record-breaking acquisition. For the person actually using Cursor, it is something smaller and more concrete: a change of who owns the models your code can train, with one setting standing between those two states. Turn on Privacy Mode regardless of what you decide next. Then decide on the data question and not the price, because the price math says staying is cheap and the data math is the only thing the deal actually changed.

Frequently asked questions

Did SpaceX buy Cursor?

Yes. On June 16, 2026, SpaceX agreed to acquire Anysphere, the maker of Cursor, in an all-stock deal valued at about $60 billion, the largest acquisition of a venture-backed startup on record. It is expected to close in Q3 2026, pending regulatory approval, at which point Cursor becomes a wholly owned SpaceX subsidiary.

Does Cursor train AI models on my code?

It depends on one setting. With Privacy Mode on, Cursor states your code is not used for training and it holds zero-data-retention agreements with model providers. With Privacy Mode off, Cursor's policy says it "may use and store codebase data, prompts, editor actions, code snippets, and other code data and actions to improve our AI features and train our models." Privacy Mode is a managed default on Teams and Enterprise plans; on individual plans you have to enable it yourself.

Will my Cursor code train xAI Grok after the acquisition?

The deal rationale, as reported, includes feeding Cursor developer data into xAI's Grok training pipeline, and after close the models Cursor's policy refers to as "our models" belong to SpaceX. With Privacy Mode on, the current policy says your code is not used for training at all. The open risk is whether those data-handling terms change once SpaceX becomes the data controller.

Will Cursor get more expensive after SpaceX buys it?

Not immediately. Pricing is unchanged as of June 2026 (Individual $20/mo, Teams $40/user/mo) and the deal does not close until Q3 2026. The directional risk is that an owner with its own model roadmap may steer pricing and usage toward house models and enterprise revenue over time, but no price change has been announced.

What is the best Cursor alternative, and what does switching cost?

On cost per finished task, leaving Cursor is expensive: its in-house Composer 2.5 finishes a task for about $0.07, while the frontier agents you would switch to (Claude Code with Opus at ~$4.10, Codex with GPT-5.5 at ~$4.82) run 10x to 60x higher. Windsurf (now owned by Cognition, $20/mo) is the closest editor-for-editor swap, but it has no comparable sub-$0.10 in-house model, so per-task cost rises once you route work to Claude or GPT. Switch for independence from SpaceX ownership, not to save money.

Sources

• CBS News. (2026). SpaceX to buy AI coding assistant Cursor for $60 billion [news report]. cbsnews.com/news/spacex-cursor-60-billion-ai-acquisition
• InfoWorld. (2026). SpaceX’s planned $60 billion deal for Cursor raises questions for CIOs [news analysis; deal rationale, data, and Grok pipeline]. infoworld.com/article/4185844
• Cursor (Anysphere). (2026). Data Use & Privacy [vendor documentation; Privacy Mode training and ZDR language]. cursor.com/data-use
• Cursor (Anysphere). (2026). Pricing [vendor documentation; current plan prices]. cursor.com/pricing
• Artificial Analysis. (2026). Cursor’s Composer 2.5: Coding Agent Index [independent benchmark; cost per task]. artificialanalysis.ai/articles/cursor-composer-2-5-coding-agent-index