Cognition CEO denies report that SpaceX tried to acquire the startup | TechCrunch

On August 19, 2026, Bloomberg reported that SpaceX had approached Cognition — the startup behind AI coding agent Devin, last valued at $25 billion — with acquisition interest, only for CEO Scott Wu to publicly rebuke the claim within hours of publication. Writing on X, Wu said Cognition "is not for sale" and stated flatly that the two companies had never held talks.

The Claim

Bloomberg's report, citing people familiar with the matter, framed SpaceX's interest in Cognition as part of a broader push to close the gap with OpenAI, Anthropic, and Google in the AI race. The timing made the story immediately plausible: SpaceX had closed a $60 billion acquisition of AI coding tool Cursor just the week before — a deal that has already yielded a concrete product in Grok 4.6, a new model released jointly by Cursor and SpaceX that claims improved performance on coding tasks and multi-step agentic workflows. Acquiring Cognition would have given SpaceX a second, larger foothold in the same market.

Yahoo Finance's coverage of the story confirmed the basic framing that SpaceX is positioning itself as a genuine competitor in the AI applications layer, not merely an infrastructure provider. Cognition's product Devin counts Mercedes-Benz, Citi, and Goldman Sachs among its enterprise customers, and the company is reportedly in new funding conversations that would push its valuation to $40 billion — up sharply from the $25 billion attached to a May 2026 round. That trajectory makes Cognition one of the more valuable independent AI coding companies still available to acquire.

Wu's denial was unambiguous in its public form but notably narrow. He disputed that any acquisition talks had occurred. He did not, however, address a separate thread in Bloomberg's reporting: that Cognition could become a customer of SpaceX's surplus GPU compute capacity — a business SpaceX has already begun building through a confirmed supply arrangement with Anthropic. That gap between what was denied and what was left unaddressed is the most telling detail in Wu's response.

What We See

The pattern here is familiar: a Bloomberg scoop citing unnamed sources, a swift public denial from the target company's CEO, and a story that splits cleanly between what was said and what was not. Our read is that Wu's response does almost nothing to resolve the underlying strategic question — whether SpaceX and Cognition are, or will be, commercially entangled — because a compute supply relationship and an acquisition are not the same thing, and he only addressed one of them.

The GPU capacity angle deserves more attention than it has received. SpaceX has been selling excess compute capacity from its ground infrastructure to third parties, with Anthropic as a confirmed customer. If Cognition were to route training or inference workloads through SpaceX infrastructure, the two companies would be operationally linked without any equity transaction required. Yahoo Finance's coverage, which closely tracked TechCrunch's reporting, did not push further on this distinction — both outlets treated the denial as the closing beat of the story rather than the opening of a follow-on question about the compute deal.

The Windsurf backstory is worth understanding to place Cognition's current posture in context. Earlier this year, Google paid $2.4 billion to acquire Windsurf's CEO and core research staff — not the company itself. Cognition then stepped in to purchase Windsurf's remaining assets, subsequently laying off 30 people and offering buyouts to roughly 200 of the inherited workforce. That sequence reads less like a startup in distress and more like one executing a deliberate consolidation play: absorbing a competitor's technology while shedding its headcount overhead. A company on that trajectory, reportedly targeting a $40 billion valuation, is not obviously looking for a buyer.

Where It Falls Short

The most significant limitation of this story is one that applies whenever a CEO denies a Bloomberg report citing anonymous sources: both can be simultaneously true in different senses. Wu may be accurately reporting that no formal acquisition conversations occurred from Cognition's perspective, while Bloomberg's sources observed something real — early-stage interest on SpaceX's side, an informal inquiry that never reached the stage of what Wu would call a "talk." Denials of this kind are rarely false; they tend to be precise in ways that leave the original claim standing at arm's length.

There is also a sourcing gap that limits how far this story can be triangulated. A piece from BeeInCrypto that reportedly covered the SpaceX-Cognition angle from a market and crypto perspective was inaccessible at publication time, blocked behind a security verification wall. That removes one potentially independent reporting chain from consideration. What remains is Bloomberg's sourced claim on one side and Wu's public statement on the other, with no third party who has spoken to people at both companies.

Grok 4.6 also deserves a caveat. Cursor and SpaceX's claims that it delivers stronger coding performance and improved multi-step agentic task handling remain unverified by independent benchmarks at this writing. The AI coding model space has a long history of vendors releasing self-reported benchmark figures that look compelling until stress-tested against real engineering workloads. Whether Grok 4.6 genuinely competes with or complements Devin at the enterprise level is an open question, and the resolution of that question has direct bearing on how much strategic value a Cognition acquisition would actually have delivered for SpaceX.

The larger arc — SpaceX making a serious play for the AI application layer through the Cursor acquisition, GPU resale to Anthropic, and now alleged interest in Cognition — appears credible regardless of whether this specific deal was pursued. Three of those four data points are confirmed. The fourth is the one Wu disputed. That asymmetry matters when reading the denial.

Sources