a16z Infrastructure Fund Completes Three Acquisitions Exceeding $75 Billion This Year
SpaceX has acquired AI programming tool Cursor, owned by Anysphere, in an all-stock deal valued at approximately $60 billion. The transaction has closed, and Cursor is now a wholly-owned subsidiary of SpaceX AI. Previously, SpaceX secured the $60 billion acquisition option in April or a breakup arrangement of about $10 billion, exercising the option after going public in June. a16z has held shares since Series A and, upon exit, held about 26% in total with Thrive.
Cursor was founded in 2022 by four MIT students and raised $2.3 billion in Series D led by Accel and Coatue in November 2025, with a valuation of $29.3 billion; it was also pursuing a new round with a valuation of about $5 billion before the acquisition. At the time of the deal, annualized revenue was in the range of $2 billion to $4 billion, with about 50,000 enterprise customers and total funding of approximately $2.7 billion to $3.4 billion. Founders hold about 20%, corresponding to about $12 billion; the employee option pool is about 11%. Shares were exchanged for approximately 389 million shares of SpaceX Class A stock.
AMD is acquiring World Labs, founded by Fei-Fei Li, for approximately $8.2 billion in an all-stock deal, expected to close by the end of the year. Li will serve as AMD's Executive Vice President and Chief Scientist, reporting to Lisa Su. World Labs focuses on spatial intelligence and world modeling, generating and simulating interactive 3D environments from text, images, and video, with a projected valuation exceeding $1 billion in 2024. This year, it has completed about $1 billion in financing, with investors including a16z, NEA, Nvidia, and Autodesk. The company has about 70 employees and previously acquired the robotics simulation team SceniX.
Stripe has acquired the AI model routing platform OpenRouter, with reports indicating a price of about $7.5 billion. The founders are set to receive about $1.5 billion, while investors will receive around $6 billion. In May, the Series B valuation was only about $1.3 billion, reflecting a premium of about five times within three months. OpenRouter provides a unified API to route between hundreds of models from OpenAI, Anthropic, Google, Meta, etc., based on price and performance, with annualized revenue rising from about $5 million to approximately $140 million. a16z participated in early rounds. Databricks had previously participated in the bidding.
Unexited groups are simultaneously raising valuations: Databricks completed $5 billion in financing in August, with a valuation of $190 billion and annualized revenue exceeding $7 billion, up over 80% year-on-year; Mistral raised about €3 billion in September led by Samsung, with a valuation of about $24 billion, targeting $1 billion in annual recurring revenue by year-end; ElevenLabs is negotiating a secondary transfer with a valuation of about $22 billion, with annual recurring revenue of about $600 million; Fal is discussing a new round valued between $15 billion and $20 billion, with annualized revenue of about $800 million; Periodic Labs, founded by former OpenAI research VP Liam Fedus, has a Series A valuation of about $7.5 billion.
Buyers are from industries needing coding data, world models, and model routing entry points: SpaceX is using publicly traded stock to acquire Cursor's engineering distribution and code data, AMD is using stock to acquire physical world models and Li's team, and Stripe is using its payment network to create a routable token distribution entry. Sellers are early shareholders and founders who have already achieved high multiples. Funding is shifting from primary market follow-on investments to industry acquisitions, with strong event-driven characteristics. Beneficiaries are early funds and founders who entered Cursor, World Labs, and OpenRouter; pressured parties are late buyers who still need to purchase inference and model layers at public market multiples, as well as independent inference clouds losing their largest coding clients to SpaceX.
Source: Public Information
ABAB AI Insight
a16z's infrastructure team has bet on "who controls the developer's daily workflow" rather than creating another large model: Cursor has evolved from a $400 million valuation editor in 2024 to a programming agent layer; World Labs is betting on Fei-Fei Li's vision and spatial intelligence path that has continued since the ImageNet era; OpenRouter is turning multi-model switching into a payment-level entry point. The same group has previously done the same thing repeatedly during the cloud and developer tools cycle: first occupy the interface, then wait for industry buyers to use stock to acquire.
The capital path is not about burning training budgets for labs, but selling already established distribution products to giants lacking entry points. SpaceX needs coding data and engineer habits, AMD needs to sell chips into workloads that "understand the physical world," and Stripe needs to make token spending routable. Early funds use small checks to acquire interface companies, while industry players pay premiums with their own highly valued stocks, forming a closed loop of "primary market pricing, secondary market stock delivery."
Similar paths can be seen in Nvidia's investment in and proximity to Hugging Face, Google's acquisition of Wiz, and Meta's purchase of WhatsApp: whoever becomes the default entry point for developers or enterprises first can be sold at a premium in the next round. The current stage has shifted from "expanding model parameters" to "controlling workflow and routing rights," with Cursor, OpenRouter, and Fal all at this level, rather than competing with Anthropic and OpenAI on parameter scale.
This is a transfer of pricing power. The prices of code generation, voice, image video inference, and model routing are no longer unilaterally determined by labs, but by who controls the engineer's desktop, who controls the token bill, and who controls the physical simulation entry. The mechanism is: industrial capital lacks data and distribution and is willing to pay far above revenue multiples with its own stock; venture capital nurtures these entry companies into assets that can be bought in one go.
ABAB News · Cognitive Laws
Entry points are more valuable than models because they can be bought in one go
Stock acquisitions are cheap printing for high-valuation companies
Those who first occupy workflows will eventually sell to those who lack workflows