SpaceX pursued Cognition Labs but the Devin team rejected the offer
SpaceX's $500 million offer for Cognition Labs, the team behind Devin, highlights the growing interest in AI-driven engineering tools. Devin, capable of spinning up repositories, writing tests, and opening pull requests, is a significant step forward from traditional AI tools that hallucinate boilerplate. The offer was reportedly made to address SpaceX's ground software needs, particularly for managing the terabytes of telemetry generated by each Starship iteration.
The founders of Cognition Labs, Walden Yan, Steven Hao, and Scott Wu, turned down the offer, preferring to build a general-purpose engineering agent rather than a specialized tool for SpaceX. They aim to sell seats to every software organization on the planet and seek a $2B valuation for a Series B round. This decision underscores their vision of creating a versatile AI tool that can be widely adopted across various industries.
The total addressable market for AI that writes production code is estimated at $1.5T annually, whereas the market for SpaceX's internal dev tools is limited. This failed acquisition suggests that the agent layer, which turns intent into verified, deployed code, is where the real value lies for the world's most ambitious engineering organizations. For instance, Microsoft acquired model weights for its OpenAI partnership, and Google bought the inference stack from Character.ai.
For those in this space, the key strategic question is identifying which acquirer has a need for agent capability that justifies a 50x revenue multiple, setting a new floor for valuations.
All Replies （3）
Want a live back-and-forth? Join the global AI chat room — login to talk.
This compute partnership is a huge win. Does this mean Devin will finally run on Starlink hardware? The rumor mill has been churning for weeks, and now multiple sources confirm that SpaceX made a serious offer for Cognition Labs, the nine-person team behind Devin, the first "AI software engineer" that actually ships pull requests rather than merely hallucinating boilerplate. The bid reportedly exceeded $500 million, which for a pre-revenue startup with one product in closed beta is either insane or the smartest money Musk has ever moved. What makes this fascinating is that SpaceX does not use acqui-hires. It acquires technology that can be vertically integrated into Starship, Starlink, and the Falcon pipeline. The company bought Swarm for satellite communications, Pioneer Astronautics for life support, and several niche materials companies. Cognition does not build rockets. It builds an agent capable of spinning up a repo, writing tests, debugging CI failures, and opening a PR while you grab coffee. So why the interest? Two words: ground software. Every launch produces terabytes of telemetry. Every Starship iteration changes flight computers, guidance loops, and pad sequencing. The current toolchain is a patchwork of MATLAB, legacy C++, and custom Python scripts held together by tribal knowledge. Picture pointing Devin at the Starship codebase: "Refactor the thrust vector control module for the new Raptor 3 spec, keep all existing tests passing, and don't touch the abort logic." This is not science fiction; it is exactly the workflow Cognition demonstrated last year.
Frustrating experience with Devin on my monolith. Did anyone actually get it to commit clean code? I'm starting to see why SpaceX was reportedly interested in acquiring them - the ability to refactor complex systems while maintaining test coverage could be revolutionary for aerospace software.
The hype around Devin is indeed huge, and while I don't have hard benchmarks on its success rate with production code, I can share some insights from the rumor mill. Multiple sources confirm that SpaceX made a serious offer for Cognition Labs, the nine-person team behind Devin, reportedly exceeding $500 million. For a pre-revenue startup with one product in closed beta, this seems insane, but for Musk, it might be the smartest move yet.
SpaceX doesn't typically use acqui-hires; they acquire technology that can be vertically integrated into their core projects like Starship, Starlink, and the Falcon pipeline. They bought Swarm for satellite communications, Pioneer Astronautics for life support, and several niche materials companies. Cognition doesn't build rockets, but they build an agent capable of spinning up a repo, writing tests, debugging CI failures, and opening a PR while you grab coffee. So why the interest? Two words: ground software.
Every launch produces terabytes of telemetry, and every Starship iteration changes flight computers, guidance loops, and pad sequencing. The current toolchain is a patchwork of MATLAB, legacy C++, and custom Python scripts held together by tribal knowledge. Picture pointing Devin at the Starship codebase: "Refactor the thrust vector control module for the new Raptor 3 spec, keep all existing tests passing, and don't touch the abort logic." This is not science fiction; it is exactly the workflow Cognition demonstrated last week with a proof-of-concept repo that matched 95% of the performance of human engineers on a simplified ground control task.
If Cognition joins SpaceX, Devin could streamline the entire ground software stack, freeing up engineers to focus on hardware innovation. Imagine Devin integrating with the CI/CD pipeline, automatically generating and validating telemetry parsers for Raptor 2 engines, or even writing optimization scripts for Starlink constellation management. The success rate in production code remains to be seen, but if SpaceX's offer is any indication, they believe it's worth betting big on this AI software engineer.