SpaceX Tried to Buy Cognition. Cognition Said No

Bloomberg reported this week that SpaceX approached Cognition, the company behind the Devin coding agent, about an acquisition. On 19 August, Cognition chief executive Scott Wu said the report was inaccurate, that the company "is not for sale", and that the two firms have not been in talks. Both things can be read at once: someone went looking, and the answer was no.

The timing is what makes this worth five minutes of your attention. SpaceX closed its $60 billion all-stock purchase of Cursor in mid-August. Within days, according to TechCrunch's write-up of the Bloomberg report, it was circling the other independent name in agentic coding. If you build with these tools, the interesting question is not who wanted to buy whom. It is how many of your daily tools now sit inside somebody else's strategy.

What was reported, and what was denied

The reported claim: SpaceX made an approach as part of a broader push to compete with OpenAI, Anthropic and Google. Bloomberg also reported that while acquisition talks ended, the companies continue to discuss a collaboration in which Cognition would use SpaceX computing capacity.

Wu's denial covers the acquisition, not the compute conversation. He did not address that second part publicly. That distinction matters, because a compute supply relationship creates a different kind of dependency than ownership does, and it is the kind that rarely makes headlines.

For context on Cognition's own position: the company raised $1 billion in May 2026 at a $25 billion post-money valuation, and is in early talks for new funding at a valuation of at least $40 billion. Devin is used by enterprise customers including Mercedes-Benz, Citi and Goldman Sachs. That is not the profile of a company that needs to sell.

Who owns the coding agent you opened this morning

Here is the practical map as of today. Ownership type matters more than the logo, because it predicts how the tool behaves under pressure.

Two patterns fall out of that. First-party agents are strategically safe but model-locked: the roadmap follows the model lab's priorities, and switching model families means switching tools. Acquired agents keep their model flexibility on paper, but their roadmap now answers to an owner with unrelated priorities. Independent agents keep both, and are the ones that get approached.

None of these is the "right" answer. The point is that you should be able to say which category each tool in your stack falls into, and you probably cannot right now.

What a denial actually changes

Not much, honestly, and it is worth being clear about that rather than dressing it up.

A refused approach tells you a company had the option to exit and chose not to. That is mildly reassuring about continuity and mildly concerning about price, since a company raising at $40 billion has to grow into that number somehow, usually through enterprise contracts and eventually through pricing. It tells you nothing about what happens in six months.

What it does not change: your exposure. If a tool disappeared from your workflow tomorrow, the cost is the same whether it was acquired, shut down, or repriced. What happens when an AI startup gets acquired covers the mechanics of that in more detail, and the Cursor deal itself is the closest live example.

Three things worth checking this week

1. Know your exit path from each agent. Can you take your prompts, rules files and configuration somewhere else? An AGENTS.md file and a plain-text prompt library port anywhere. A proprietary workspace format does not. This is the same test as app builder lock-in, applied to your editor.

2. Check whether your agent is model-locked. If your tool only runs one lab's models, an ownership change is not your biggest risk. A model deprecation is.

3. Write down what you would do with two weeks' notice. Not a plan, a paragraph. Which tool you would move to, roughly how long the migration takes, and what breaks. Most teams cannot answer this, and the answer takes twenty minutes to produce.

If you want the shorter version of all of it: the tools consolidating around you is now a normal feature of this market, not an event. Reading how to keep up with AI news without letting it eat your week is a more useful skill than tracking any individual deal. And if you are choosing an agent from scratch right now, the solo developer comparison is a better starting point than a valuation table.

FAQ

Did SpaceX try to buy Cognition? Bloomberg reported that it made an approach. Cognition CEO Scott Wu said on 19 August that the report was inaccurate, that the company is not for sale, and that the companies have not been in talks. Neither Bloomberg nor Cognition has published further detail since.

Is Devin owned by SpaceX now? No. Devin is Cognition's product and Cognition remains independent. SpaceX owns Cursor, which is a different company and a different tool.

Should I stop using Cursor because SpaceX owns it? Not on this news alone. An ownership change is a reason to know your migration path, not a reason to migrate. Judge the tool on whether it still does the work.

What was Cognition valued at? It raised $1 billion in May 2026 at a $25 billion post-money valuation, and per the same reporting is in early talks for new funding at a valuation of at least $40 billion.