SpaceX acquires Cursor, sparking debate over the coding assistant's strategy and its origins as a mechanical CAD tool

ah yes, no choice but be acquired for...

*checks notes*

$60 billion dollars

@levie Funny that many people kept saying “cursor is cooked, claude code too good”. Cursor kept producing as well as they could in a large market and now people say “oh yeah they’re a mega success”

@MollySOShea @ThriveCapital @cursor_ai @mntruell @mntruell congrats on the acquisition! what made you give up on the mechanical engineering/AI to CAD tool ?

Folks at @NousResearch have a huge inside track in this space! Who else do I need to mention?

@levie applied yes, but specifically the editor is the moat, not the agent or the model. cursor's edge was the IDE feedback loop. claude code and codex still lack their own editor surface. that's the next round of M&A.

@levie 60B is absolutely insane.

@levie to when you just run generic frontier mode and cross your fingers

the routing/distillation call is gonna decide who wins round 2

@levie Disagree. It’s become clear (at least every frontier lab except google) that coding as a product is part of their core biz, and acquiring cursor was some mix of application/coding data/ability to train a frontier model (in coding at least). They became a quasi frontier lab

@levie they trained their own model as a way to cap costs, right?

@levie This is a marriage of convenience or desperation. If Curaor had been doing so well, it would have remained an independent company.

@levie They used to call these AI wrappers, now it's harness engineering.

@levie Cursor cracked it because they stopped treating the model as the product. Eighteen months ago every demo was "we wrapped GPT." Now the moat is workflow, routing, and knowing your user's job better than the foundation labs ever will. The applied layer just got real.

@levie The Cursor folks should probably look up what @mcuban did when broadcast-dot-com got sold

@levie Yes, and they have DATA. Huge amounts of data from human developers coding with AI. Highly sought after for training models.

@levie Balancing frontier models with custom training shows a practical way to optimize performance in specific fields. This approach helps companies address market needs more directly.

@levie Technology gets the headlines. Distribution gets the billions.

@levie The model router angle is underrated. Cursor didn't need to own the model — they owned the context delivery layer for a specific workflow. Domain focus + routing discipline is what actually scales at the applied layer.

@levie i’d trust this analysis only and only if we’d hear there was another buyer at least in low billions. otherwise, it’s inflated due to IPO.

@levie $60B for a code editor — even a brilliant one — assumes AI coding assistants have durable moats. OpenAI, Google, and Anthropic are all building the same thing natively. What exactly is SpaceX buying that won't be commoditized by 2027?

@levie Smart time to exit for Cursor. And Grok gets a distribution channel