SpaceX had a choice.
Pay for Cursor's data as pure R&D and hope a model ships. Or rent Colossus to Cursor, get the data engine running, and buy a one-year right to acquire for $60B if the model wins — or walk away for $10B if it doesn't.
They chose option two.
The signal came a month before the announcement. Elon tweeted three words about Cursor's supposedly proprietary model: "Yeah, it's Kimi 2.5."
The base model was free. Available to anyone.
What Cursor built on top was two years of professional engineers correcting real code inside private systems. The asset no lab can buy, replicate, or brute-force regardless of how many engineers they hire.
Not the editor. The last mile.
Coding is where AI makes the most money today. SpaceX just bet $10B to find out if Cursor's data plus xAI's compute beats what OpenAI and Anthropic have approximated without it.
With SpaceX IPO math, $10B is a rounding error. A $60B acquisition might turn out to be the best capital allocation in tech this year.
I wrote about what this means for every vertical AI company - why no lead is safe, why open source is actually vertical AI's best friend not its threat, and what the defensible moat actually looks like in 2026.