On Friday, August 14, SpaceX closed its acquisition of Anysphere — the roughly 300-person company behind the AI coding tool Cursor — at an implied equity value of $60 billion. It is the largest acquisition of a venture-backed startup ever recorded.
Not a dollar of it was cash. The deal was all-stock: Anysphere’s shares converted into roughly 389.3 million SpaceX Class A shares, priced off a volume-weighted average of the seven trading days before closing, with assumed options and restricted stock taking the total issuance to about 391 million. SpaceX’s X67 subsidiary merged into Anysphere, and Anysphere became a wholly owned SpaceX subsidiary reporting into SpaceXAI, the division formed when SpaceX and xAI combined.
That last sentence is the whole story, and almost nobody led with it.
Cursor’s core feature is that it does not care which lab made your model. Its router sends a given request to Anthropic’s Claude, OpenAI’s GPT models, or xAI’s Grok, depending on the task. Developers have paid for that neutrality for three years. As of Friday, the company operating the router is also the company that makes one of the three options — and loses money every time the router picks one of the other two.
What $60 Billion Bought
Anysphere was founded in 2022 and shipped Cursor in early 2023. It reached $100 million in annualized recurring revenue faster than any software company on record, crossed $1 billion by late 2025, and by June 2026 was running at roughly $4 billion annualized — about $2.6 billion of it enterprise, according to Reuters. It did that with a headcount of around 300.
| The Deal | Detail |
|---|---|
| Announced | June 16, 2026 (SEC Form 8-K) |
| Closed | August 14, 2026 |
| Implied equity value | $60 billion, all-stock |
| Shares issued | ~389.3M converted; ~391M total Class A |
| Cash paid | $0 |
| Break fee if it had failed | $1.5B cash + $8.5B in compute |
| Target headcount | ~300 employees |
| Target revenue | ~$4B annualized (~$2.6B enterprise) |
| Price per employee | ~$200 million |
| New parent | SpaceXAI (SpaceX + xAI) |
Read the break fee again. Had the deal collapsed, SpaceX had agreed to hand Anysphere $1.5 billion in cash and $8.5 billion in compute for the trouble. A $10 billion consolation prize is not what you offer for a code editor. It is what you offer when you have decided you cannot afford to lose the distribution.
That reading is supported by how the deal was originally framed. In April, SpaceX put two options on the table: a $10 billion partnership, or a $60 billion acquisition. It exercised the acquisition option on June 16, four days after SpaceX itself went public.
The Router Is the Product
It is worth being precise about what Cursor sells, because the conflict lives in the details.
Cursor is not a model. It is the layer between you and the models — the thing that decides which one sees your codebase, how much context it gets, and which requests are worth an expensive frontier call versus a cheap fast one. Cursor Router, live since early August, arbitrates across Claude, GPT and Grok. That arbitration only has value if the arbitrator is indifferent to the outcome.
It is no longer indifferent. Every request the router sends to Anthropic or OpenAI is revenue leaving SpaceXAI for a direct competitor, and margin the parent could have kept in-house. xAI reported a loss of $6.35 billion in 2025. The commercial logic of moving even a modest share of Cursor’s traffic to Grok is not subtle, and it does not require anyone to act in bad faith — only to keep optimizing.
Nothing Has Been Announced. That Is Also Information.
To be fair to SpaceXAI: at the time of writing, no pricing change, no model-access change and no rebrand has been announced. There is no roadmap post from SpaceX or xAI leadership on what happens to third-party models inside Cursor. Elon Musk’s public comment on the closing was one word: “Welcome!”
What exists instead is a sequence.
- August 9 — staff were told at an internal meeting that the Cursor brand would most likely be phased out over the following months as the business merges into SpaceXAI, according to one report. No such change has been confirmed publicly.
- August 12 — Grok 4.6 became available inside Cursor. That is two days before the paperwork closed, which means the engineering integration was running ahead of the legal one.
- August 14 — the deal closes. Cursor is listed alongside Grok Build, Grok Bot and Grok API in the SpaceXAI product family. Grok is listed first.
- August 18 — Cursor Origin, a code hosting platform, ships. Four days after close, the tool that reads your code now also offers to store it.
Grok 4.6, for the record, scored between 69.9% and 70.8% on CursorBench v3.2 and 61 on the Artificial Analysis Intelligence Index — competitive, not dominant. Which is precisely the situation in which default placement is worth more than benchmark wins.
That comment is the entire risk expressed in two sentences, and note what it is not. It is not opposition to the deal, or to Grok, or to Musk. It is a person describing the one change that would make a tool they use six hours a day worse, and having no way to find out whether it is coming.
What SpaceXAI Says It Brings
The pitch from the buyer’s side is compute. SpaceXAI claims access to the largest GPU fleet in the world, including the Colossus supercomputer, and argues that Cursor’s own in-house models — the Composer line — can now be trained and iterated far faster and served more cheaply than an independent company could manage. Cursor’s COO, Jordan Topoleski, described the arrangement as “an order of magnitude more compute.”
That claim is currently a company assertion. No independent benchmark or customer result has confirmed the cost advantage.
It is, though, a coherent thesis. Andreessen Horowitz — an investor, so read accordingly — argued in a post on the deal that iteration speed rather than raw model quality decides the AI coding market, citing enterprise customers now generating around 65% of production code with AI, and one Fortune 500 firm going from 150,000 to 800,000 lines of code shipped per week. If that is the game, then the company with the most GPUs to burn on experiments has a real structural edge, and Cursor spending $8.5 billion worth of someone else’s compute rather than raising for its own is a rational trade.
What Wall Street Thinks It Bought
SpaceX went public on June 12, 2026, four days before announcing the acquisition. Its stock (NASDAQ: SPCX) rose 1.5% to $143.45 on the news — a mild reaction for a $60 billion purchase, which is what happens when the consideration is your own paper rather than your cash.
Morgan Stanley maintained a $300 base price target and projected the acquisition could add up to $13 billion in SpaceX revenue by 2027. Against roughly $4 billion of annualized revenue today, that number is not a forecast of Cursor’s current business. It is a forecast of Cursor’s current business plus whatever cross-selling the SpaceXAI stack makes possible — which is another way of saying the Street has priced in exactly the integration developers are worried about.
What This Means If You Just Write Code
Nothing today. Possibly quite a lot by winter.
The specific things to check are unglamorous and concrete: whether the model picker still lists Anthropic and OpenAI models at the tiers you pay for; whether the default selection for a new project changes; whether router behaviour under “auto” shifts without a changelog entry; whether your enterprise agreement still names the labs your code may be sent to; and whether Cursor Origin’s hosting terms differ from the editor’s.
The broader point applies well beyond one editor. A large share of the AI tools people rely on are routers — assistants, IDEs, agent frameworks, search products — that sit above several labs and claim to pick the best one for you. Almost all of them are, at some level, owned by someone. This is the first time one of the big ones has been bought outright by a lab whose model it routes to, and how it is handled will set the template for the rest.
What To Watch
- The model picker. Removal of a competing model, or a downgrade in which tiers get access to it, is the concrete event. Everything else is speculation until then.
- Whether “auto” routing gets documented. A published routing policy would settle this quickly. Continued silence is itself an answer.
- The Cursor name. If the brand folds into Grok Build, the reported internal guidance was accurate and the rest of the integration timeline is probably accurate with it.
- Enterprise renewals. $2.6 billion of annualized enterprise revenue was sold on neutrality. Watch whether large customers re-sign, and what they negotiate about data handling.
- Anthropic and OpenAI’s response. Both now sell coding tools competing with a platform their models are billed inside. The cleanest counter is to make their first-party tools better than the router.
The Takeaway
SpaceX closed a $60 billion all-stock acquisition of Anysphere on August 14, the largest venture-backed startup exit on record, paid entirely in about 391 million SpaceX Class A shares. Cursor — roughly 300 people and about $4 billion in annualized revenue — is now a wholly owned subsidiary inside SpaceXAI, alongside Grok.
The product it sold was neutrality: a router that picks Claude, GPT or Grok on the merits. Its owner now makes one of those three and lost $6.35 billion at xAI in 2025. No change to pricing, model access or branding has been announced. Grok 4.6 was integrated two days before the deal legally closed.
The number worth keeping is not $60 billion. It is $10 billion — the cash-and-compute break fee SpaceX agreed to pay if the deal fell through. Somebody decided that owning the layer between developers and their models was worth more than the models themselves. Whether they were right is now a question about a dropdown menu.