Either full-stack or out: The behind-the-scenes calculations of xAI's $60 billion acquisition of Cursor

null

Author: Tara Tan

Translation: Deep潮 TechFlow

Source: The Strange Review

Deep潮 Guide: SpaceX's subsidiary xAI acquired Cursor's parent company Anysphere for $60 billion in stock, not to gain market share, but to access the high-quality training data generated daily by 7 million developers writing code. Strange Ventures partner Tara Tan uses this deal to make a judgment: to become an AI giant, you must integrate compute power, models, and applications across the full stack. This short commentary breaks down the path that led Anthropic's revenue to increase 540 times in 28 months, and also explains why model companies will crazily acquire at the application layer next. Note that the author is a VC, and full-stack is precisely their investment theme.

Code generation is the most powerful killer application of large language models to date, bar none.

Anthropic's revenue grew from an annualized $87 million in January 2024 to $47 billion in May 2026, a roughly 540-fold increase in 28 months. This rapid growth is driven by two engines working simultaneously: one is top-down enterprise cooperation (Claude is the only cutting-edge model available on three major cloud platforms), and the other is bottom-up developer penetration, relying on Claude Code. This product is the company's fastest-growing, reaching $2.5 billion in annualized revenue in just 9 months. Now, Anthropic holds 54% of the enterprise AI programming market.

Cursor is the same bet that SpaceX has placed.

Yesterday, SpaceX announced the acquisition of Anysphere, the company behind Cursor, for $60 billion in stock. Cursor is an AI programming tool used by 7 million developers daily. It was incubated at MIT four years ago and has reached an annualized revenue of $2 billion, making it the highest-earning AI programming tool in this category. Over the past year, its market share has been declining, from 41% to 26%, because Claude Code has gained ground. But xAI's purchase is not fundamentally about market share.

xAI already has a full stack: Colossus for compute, Grok for models, X for applications. The problem is, X is used for brushing, while Cursor is used for coding. Developer-generated code provides the strongest signals in AI training data, and this is exactly the piece Grok needs to strengthen its combat capability.

This confirms an idea I’ve been pondering since the deal between OpenAI and NVIDIA last September:

To build an AI giant, you must do the full stack.

This logic is becoming clearer. Better products lead to better infrastructure (more data), and better infrastructure, in turn, leads to better experiences. This has always been our core investment logic at Strange.

Caption: The author's team’s investment logic diagram for "full-stack closed-loop"

Building a full stack brings two things:

First, the unit economics of building and training models become sustainable.

Second, you can obtain proprietary training data from the application layer, differentiating yourself from other model vendors. Locking in user data and workflows creates a formidable moat.

In the coming years, we will likely see these moves: model companies developing applications internally, or aggressively acquiring and directly taking over the application layer.

Entrepreneurs now often say: because making products has become 10 times easier than before, companies must be 10 times more ambitious than before to succeed. At present, this seems to be true across all sectors.

— Tara