SpaceX’s $60 billion stock-based acquisition of Cursor is not an M&A revival. It’s a seizure of the last defensible layer in the AI stack. Jason Calacanis noted Cursor was paying Anthropic for nearly half its revenue before Anthropic launched Claude Code, a direct competitor. The move proves that startups building on frontier lab APIs are sitting ducks.
The real shift is verticalization. Ali Ansari explained that SpaceX’s compute business already generates more profit than many dedicated AI labs. By owning the developer IDE, SpaceX gains a direct pipeline for high-value workflow data. This data is the fuel for proprietary cognitive loops, where human experts train agents on subjective judgment - the 5% of nuance that separates an AI from a licensed lawyer.
Ryan Daniels of Crosby Legal described his AI-first firm’s model: it employs lawyers specifically to train its internal systems, then offers legal services at flat rates. The billable hour is dying because the product is no longer a model, but a vertically integrated service layer. Value migrates away from the labs dumping capital into J-curve training and toward companies that own the feedback loop.
"The era of simply picking the best model is over. The future of the firm lies in cognitive loops where humans train agents on proprietary, real-world workflows."
- Satya Nadella, This Week in AI
The acquisition priced Cursor at a 60x revenue multiple on a $1 billion annualized run rate. Calacanis calculates it as 20x on a $4 billion run rate, but the premium is for strategic autonomy. With its own compute footprint, Cursor can now train frontier models and bypass the bottlenecks that have slowed xAI’s development.
This vertical stack - compute, model, interface - creates a closed system. Keon from Stacker News Live tracked SpaceX’s pivot toward 'agentic inference' in orbital data centers, solving earthbound energy and cooling constraints. The Cursor buy secures the workstation, the point where proprietary data is generated. Musk is building the AWS of the sky, financed by a trillion-dollar recurring revenue stream.
The warning for startups is blunt. Calacanis advised Y Combinator founders to reject OpenAI’s free credits-for-equity deals, calling them a roadmap for platform cannibalization. The smart move is building 'headless' model routers that can switch between Claude, GPT, and open-source models like DeepSeek. Relying on a single foundational partner is a suicide mission.
"Don't trust the platforms. Anthropic reportedly launched its own coding tool after monitoring Cursor’s token usage, a move Calacanis calls a 'shiv in the middle of the night.'"
- Jason Calacanis, This Week in Startups
The battlefield has narrowed. Ansari predicts nearly 100% of future AI data spending will concentrate on the application layer. There will be millions of specialized agents, but only a handful of companies can afford the capital and compute to build frontier models. The divide is binary: you control the stack, or you are a tenant on someone else's infrastructure.
SpaceX now controls the stack.





