Cursor on the quad

In October 2023 four people who had met at MIT announced an $8m seed round, led by the OpenAI Startup Fund, for a text editor.^{4TechCrunchAnysphere raises $8M from OpenAI to build an AI-powered IDE11 Oct 2023} Michael Truell, Sualeh Asif, Arvid Lunnemark and Aman Sanger called their company Anysphere and their product Cursor. “In the next several years, our mission is to make programming an order of magnitude faster, more fun and creative,” Mr Truell said at the time.^{4TechCrunchAnysphere raises $8M from OpenAI to build an AI-powered IDE11 Oct 2023} Programmers, a famously unimpressed audience, hear such promises about their editors every year. This time they were taken at their word.

Three years later the editor belongs to SpaceX. A regulatory filing showed that the $60bn acquisition took effect on 14 August 2026, Bloomberg Law reported, about two months after SpaceX announced it.^{5Bloomberg LawSpaceX Completes $60 Billion Acquisition of AI Startup Cursor14 Aug 2026} That is a long way for a student project to travel. The more interesting question is what was bought. We think the answer is a customer relationship rather than a technology, and that this is a less comfortable thing to have built a $60bn firm on than it sounds.

Faster than a term paper. Cursor valuation at round or sale, $bn (late 2024 is pre-money)

	Valuation, $bn
Late 2024	2.5
Jun 2025	9.9
Nov 2025	29.3
Aug 2026 (sale)	60

Source: TechCrunch (2025), Cursor (June and November 2025), Bloomberg Law (August 2026)³¹²⁵

First, the speed, which is real. In June 2025 Cursor said it had raised $900m at a $9.9bn valuation and grown to more than $500m in annualised revenue.^{1CursorSeries C6 Jun 2025} TechCrunch had reported the previous round at $100m on a $2.5bn pre-money valuation in late 2024, and noted that annualised revenue had been about $300m in mid-April, a rise of 60% in roughly seven weeks.^{3TechCrunchCursor's Anysphere nabs $9.9B valuation, soars past $500M ARR5 Jun 2025} In November 2025 the company announced $2.3bn at a $29.3bn post-money valuation and said it had passed $1bn in annualised revenue; Accel, Thrive and Andreessen Horowitz continued to back it, and Coatue, Nvidia and Google joined.^{2CursorSeries D13 Nov 2025} Annualised revenue is a recent month multiplied by 12, which flatters a business that is still accelerating. Even so, a valuation that rose roughly twelvefold in under two years, from $2.5bn to $29.3bn, is not a rounding error.

The product explains it. Cursor puts the artificial-intelligence help inside the editor where programmers already spend their days. Mr Sanger told TechCrunch at the seed stage that people thinking of “AI plus coding” usually imagined autocomplete, which was the founders’ cue to build something else.^{4TechCrunchAnysphere raises $8M from OpenAI to build an AI-powered IDE11 Oct 2023} A good interface to someone else’s model, sold from $20 a month to individuals and $40 to businesses, is a fine business while the model is good and the rivals are few.^{3TechCrunchCursor's Anysphere nabs $9.9B valuation, soars past $500M ARR5 Jun 2025}

Neither condition is permanent, and here the contrarian case begins. Cursor does not make the thing that makes it valuable. Its users pay for models built largely by other companies, which can and do sell their own coding tools. TechCrunch noted in 2025 that Anysphere had turned down earlier acquisition approaches, including from OpenAI.^{3TechCrunchCursor's Anysphere nabs $9.9B valuation, soars past $500M ARR5 Jun 2025} Depending on your supplier is comfortable until the supplier becomes a competitor.

The founders saw this, and spent 2026 trying to own more of the stack. In March Cursor launched Composer 2, its own coding model. Users on X soon noticed that it appeared to be built on Moonshot AI’s Kimi 2.5. Lee Robinson, a Cursor executive, confirmed that the model “started from an open-source base” but said only about a quarter of the compute spent on the final model came from it, and Mr Sanger conceded that it was a miss not to mention the Kimi base in the company’s blog from the start.^{6TechCrunchCursor admits its new coding model was built on top of Moonshot AI's Kimi22 Mar 2026} We read that as a company doing what its strategy demands and omitting a line from the press release. Building on an open model is thrift. Not saying so is the part that needed a second draft.

The counter-argument is that the moat is the data: millions of accepted and rejected suggestions from working programmers, which is exactly what a coding model wants. That is plausible, and it is the best reason to think the price was not foolish. But a moat that deepens only if the model-makers stay friendly is a drainage ditch. And a buyer that, as Bloomberg Law framed it, is trying to gain ground on Anthropic and OpenAI needs the ditch less than the data.^{5Bloomberg LawSpaceX Completes $60 Billion Acquisition of AI Startup Cursor14 Aug 2026}

For four former students the arithmetic is kind. For the rest of the industry the lesson is plainer. Build on someone else’s engine, and sooner or later the engine’s owner, or a rival with one, comes shopping.