Product1 distinct publisher3 min readUpdated
A denial of a Bloomberg report on SpaceX buying Cognition leaves the compute relationship unaddressed, and leaves buyers choosing coding tools from a shrinking independent field.
The Product Desk · Product desk

Compiled by The Product DeskSomething wrong?How this is made
Cognition CEO Scott Wu said on X that Bloomberg's Wednesday report that SpaceX tried to acquire his company was inaccurate, that Cognition "is not for sale," and that the two companies have not been in talks [1][2]. The denial arrives a week after SpaceX closed a $60 billion acquisition of Cursor, which is the fact engineering leaders should actually be pricing in: coding-tool decisions are now being made inside a vendor set that is being bought while you evaluate it [3].
Begin with what Wu did not say. Bloomberg reported that the acquisition talks are no longer active but that the two companies continue to discuss working together, possibly with Cognition using SpaceX computing capacity, which SpaceX is selling to other AI companies including Anthropic until it needs the capacity itself [4]. Wu's statement did not address that claim [4]. "Not for sale" and "not commercially entangled" are different assertions, and only one of them was made. Neither company responded to TechCrunch's requests for comment [5].
The buyer's motive is on the record. Musk told SpaceX employees last week that in about four or five years, AI will be 99% of the company's value [6]. The xAI business SpaceX absorbed earlier this year remains relatively early-stage and has fallen behind competitors [7], and SpaceX went public in June with a market capitalisation peaking near $2.3 trillion [8]. Coding assistance is the clearest available path to AI revenue, which is what Anthropic's growth on the back of Claude Code demonstrates [9]. Cursor and SpaceX had already shipped together before the deal closed, jointly releasing Grok 4.6 this month with higher scores on coding and multi-step agentic benchmarks [10].
Cognition is the obvious next target on paper: it sells the Devin coding agent into Mercedes-Benz, Citi, and Goldman Sachs [11], and it remains one of the largest independent coding startups not yet absorbed by a model maker [12]. It raised $1 billion at a $25 billion post-money valuation in late May and, per Bloomberg, is in early talks for a round at $40 billion [13] - a 60% mark-up in roughly three months [14]. SpaceX paid 1.5 times that asking valuation for Cursor [15], which tells you something about how the acquirer values distribution into developer workflows.
For anyone renewing a licence, the relevant precedent is what Cognition itself did after buying Windsurf's remaining assets last year, following Google DeepMind's $2.4 billion acqui-hire of that startup's CEO and top researchers [16]. Cognition laid off 30 people, offered buyouts to the remaining 200 Windsurf employees, and set expectations of 80-plus hour weeks and six days in the office for those who stayed [17]. Acquisition changes the roadmap and the people holding it. And a tool owned by SpaceX inherits Grok's enterprise baggage, including last year's "MechaHitler" incident and this year's nonconsensual sexual imagery scandals [18].
Watch three things. Whether the $40 billion round closes, because outside capital is what funds a refusal to sell [13]. Whether a Cognition-SpaceX compute agreement is announced, since that is the unaddressed part of the denial [4]. And whether Grok models become the default rather than an option inside Cursor, which is the first place a $60 billion purchase should show up in a product you already pay for [3][10].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
Cursor and SpaceX were already working together before the acquisition closed, and this month jointly released Grok 4.6, a model that scores higher on benchmarks for coding and complex multi-step agentic tasks.
Cognition remains one of the largest independent AI software coding startups that hasn't yet been acquired by a major AI model maker.
SpaceX's $60 billion acquisition of Cursor, another AI coding startup, closed last week, a few days before the Bloomberg report.
SpaceX acquired Musk's AI company xAI earlier this year; the xAI business remains relatively early-stage and has fallen behind competitors.
SpaceX went public in a blockbuster IPO in June, with its market capitalization rising to nearly $2.3 trillion at its peak.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One publisher relaying unnamed sources, with the central claim denied
The cluster contains a single source, and its headline claim is second-hand: TechCrunch relays Bloomberg's unnamed sources for both the acquisition attempt and the compute-collaboration talks. The only on-record statement is Wu's denial, which contradicts the acquisition claim and does not address the compute claim. Both principals declined comment. Verifiable hard facts (the closed Cursor deal, the funding history, the Windsurf sequence, Grok 4.6's existence) sit alongside unquantified assertions with no benchmark scores or revenue figures, which caps evidence quality well below the midpoint.
Real transactions and named enterprise users, no usage metrics
Adoption signals are concrete but coarse. A $60 billion acquisition of Cursor has closed, a jointly built model (Grok 4.6) shipped this month, Devin is reported in use at Mercedes-Benz, Citi, and Goldman Sachs, and SpaceX is reported to be selling compute to AI players including Anthropic. What is missing is any quantity: no seats, revenue, token volume, capacity, or deployment scope, and the Cognition-on-SpaceX-compute arrangement is only under discussion. That supports moderate rather than strong adoption.
Headline claim contested and forward framing outruns disclosed substance
The story's most attention-getting element, that SpaceX tried to buy Cognition, is denied by Cognition's CEO and rests on unnamed sources, while the accompanying framing leans on very large forward numbers: a near-$2.3 trillion peak market cap, Musk's claim that AI will be 99% of SpaceX's value in four or five years, and a $40 billion valuation that is only in early talks. Against that sit an early-stage xAI that has fallen behind rivals, benchmark claims with no numbers, and no revenue evidence for the monetization thesis. The overstatement is moderate rather than extreme because the article itself flags the denial, the unnamed sourcing, and the unaddressed compute claim.
Every named party benefits from the version of events it tells
Incentives are unusually legible here. SpaceX, freshly public at a near-$2.3 trillion peak and told by its CEO that AI will be 99% of its value, needs visible AI revenue momentum, which makes coding-startup interest a useful signal. Cognition is reportedly raising at $40 billion, 60% above its May mark, giving Wu reason to deny being for sale to project independence and strength to investors, employees, and enterprise customers, while leaving the compute-collaboration claim conveniently untouched. Bloomberg's sources are unnamed and their interests undisclosed. The article's own record of Cognition's hard-edged post-Windsurf restructuring and Grok's safety scandals shows reputational stakes on both sides.
Low-moderate: structural facts firm, the news itself unresolved
Confidence splits by claim type. The transactional and historical facts (Cursor's closed $60 billion deal, the $25 billion May round, the Windsurf sequence, the IPO, Grok 4.6's release) are firmly stated and internally consistent, and the derived arithmetic follows from figures in the source. But the story's news hook is contested by a named principal, the compute arrangement is unverified and unaddressed, both companies declined comment, and there is only one publisher in the cluster with no corroboration. That supports acting on the consolidation trend while treating the acquisition and compute claims as open.
build
SpaceX went shopping for a second AI coding company five days after closing Cursor2 distinct publishers
build
Grok 4.6 lands in Copilot two days after launch, and the model picker becomes a procurement problem1 distinct publisher
build
Developer habit, priced at $965B: what Anthropic's run actually proves1 distinct publisher
build
SpaceXAI's new Grok terms make "indirect competitor" a deployment risk1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 19, 2026