TechFlowPost published a translated commentary on June 18 by Tara Tan from The Strange Review, titled “Full Stack or Out: The Calculation Behind xAI’s $60 Billion Acquisition of Cursor.” The article centers on SpaceX-affiliated xAI’s $60 billion stock acquisition of Anysphere, the company behind Cursor. Tan’s central argument is direct: the deal is not primarily about purchasing market share in AI coding tools, but about gaining access to the high-quality training data produced every day by 7 million developers as they write code through Cursor.
The source also makes the author’s perspective explicit. Tara Tan is a partner at Strange Ventures, and “full-stack” is described as one of her firm’s own investment themes. TechFlowPost’s editor’s note therefore reminds readers that the piece carries a venture-capital viewpoint. The article opens with a compact thesis: building products has become 10 times easier than before, so companies need to be 10 times more ambitious than before. Within this framing, code generation is described as the strongest killer application for large language models so far.
Anthropic and Claude Code as the growth benchmark
Tan begins by using Anthropic as the comparison point. According to the original article, Anthropic’s revenue rose from an annualized revenue run rate of $87 million in January 2024 to $47 billion in May 2026, a roughly 540-fold increase over 28 months. The commentary attributes that pace to two engines working at the same time: top-down enterprise partnerships and bottom-up developer adoption.
On the enterprise side, the article says Claude is the only frontier model available across all three major cloud platforms. On the developer side, the key product is Claude Code. Tan writes that Claude Code is the fastest-growing product in Anthropic’s history, scaling from zero to $2.5 billion in annualized revenue within nine months. The piece also cites a market-share figure: Anthropic now holds 54% of the enterprise AI programming market.
Cursor’s acquisition and the value of developer data
In Tan’s reading, Cursor represents the same type of bet for SpaceX. The source states that SpaceX announced the acquisition of Anysphere, Cursor’s parent company, in a $60 billion stock deal. Cursor was incubated at MIT four years ago. As an AI coding tool, it is used by 7 million developers every day, has reached $2 billion in annualized revenue, and is described as the highest-revenue AI coding tool in its category.
The commentary does not ignore Cursor’s recent loss of share. Over the past year, its market share fell from 41% to 26% as Claude Code gained ground. But Tan’s point is that xAI is not buying Cursor mainly for that share. The value lies in the signal generated when developers write code inside Cursor. The original article describes this coding activity as one of the strongest training-data signals in the AI field and as the missing piece that Grok needs to strengthen its capabilities.
How Colossus, Grok, X and Cursor fit together
The article breaks xAI’s existing stack into three parts: Colossus is compute, Grok is the model, and X is the application. The limitation, as Tan phrases it, is that X is a place where users scroll, while Cursor is a place where developers write code. In this framework, an application is not only a distribution channel. It is also a source of data and workflow feedback that can continuously feed model improvement. By bringing Cursor into the system, xAI gains a developer work environment that can connect directly with the model-training loop.
This leads to the broader idea Tan says she has been considering since the OpenAI and Nvidia transaction last September: to become a major AI company, a firm has to be full-stack. The article describes the logic as a loop. Better products produce better infrastructure by generating more data; better infrastructure then feeds back into better user experience. The original post also includes a chart from the author’s team explaining the “full-stack loop,” and Tan says this has long been Strange’s core investment logic.
Why model companies move upward into applications
According to the commentary, going full-stack produces two outcomes. First, the unit economics of building and training models become sustainable. Second, a company can obtain proprietary training data from the application layer, differentiating itself from other model providers. Once user data and workflows are locked in, they become a moat. Based on that logic, Tan writes that the next few years will show model companies either growing applications internally or pursuing aggressive upward acquisitions to absorb the application layer directly.
The article closes by returning to a phrase that has become popular among founders: because building products is now 10 times easier than before, a company has to be 10 times more ambitious to win. Tan writes that, at this point, the statement has played out across different sectors. The original TechFlowPost page lists the author as tara tan, with the handle @taratan. It also includes TechFlow’s official community information, including the Telegram subscription group TechFlowDaily, the Twitter official account TechFlowPost, and the English Twitter account BlockFlow_News.