Vercel Founder Guillermo Rauch: People Want to Create
Vercel founder Guillermo Rauch stated that the coolest aspect of the current wave of artificial intelligence is that everyone wants to create things; aside from the short-term noise, this is immeasurably good for the world, as people want to create and launch, not just consume.
This statement targets the traffic his company is currently experiencing. Vercel started in 2015 as Zeit and rebranded in 2020, with Next.js becoming the mainstream React framework. He also mentioned that the company took ten years to achieve 1 billion deployments, and then added another 1.4 billion in just ten months. In July 2026, he stated in an interview that the platform has approximately 6 million deployments per day, half of which are triggered by programming agents, with the AI gateway averaging around 1 trillion tokens daily. The proportion of agent deployments increased from about 5% in June 2025 to over 21% in February 2026, with about 70% coming from tools like Claude Code. The v0 generates interface code, with reported user numbers exceeding 4 million and over 100 million applications generated within a year.
Business numbers follow deployments. In September 2025, Series F raised about $300 million, with a valuation of approximately $9.3 billion; by February 2026, the annual recurring revenue was about $340 million, a jump from about $100 million at the beginning of 2024. The weekly download volume of the AI SDK is about 3 million times. He publicly stated that the role of engineers has shifted from delivering individual outputs to building factories that can continuously ship, and he has recruited the creator of Cloud Run to oversee Functions, containers, and sandboxes. Vercel Labs is written as a public experimental window. His narrative endpoint is a one-person unicorn: agents triaging, modifying code, and monitoring traffic.
Who is buying and who is selling: the buyers are the sentiment of "anyone can launch" and deployment minutes, while the sellers are self-built data centers and writing infrastructure from scratch. The events are driven by the desire to create, with funds flowing from model invocation fees to hosting, gateways, and preview environments. The beneficiaries are cloud vendors that connect agent outputs to production domain names, while the pressured parties are consumer-grade products that only provide chat windows without publishing pipelines. The short-term drama is model conflicts, while the mid-term bill is every git push.
No new audited revenue was attached to the public statement this week; deployments and tokens are the company's metrics.
Source: Public Information
ABAB AI Insight
Rauch has shifted the narrative from "will models replace humans" to "will humans start shipping products." Vercel has spent ten years building preview links and edge networks; now that models can write React, these pipelines have become the default dormitory for agents. The desire to create is on the demand side, while the number of deployments is on his revenue side.
The capital path is to connect generation to operation. v0 turns prompts into front-end code, Next.js turns front-end code into framework conventions, and the cloud turns conventions into billable functions and bandwidth. Each additional push by an agent adds to the bill. The Series F valuation follows this slope, and the IPO preparation questions whether GAAP can accommodate this level of usage. His recruitment of Cloud Run personnel indicates a concern that computing in the agent era will no longer be a single function call but rather a long-term sandbox.
Comparisons can be drawn to GitHub turning social graphs into code hosting and AWS turning credit cards into servers. Vercel aims to be the Heroku for agents: you handle the ideas, and it handles the domain names. The industry phase is shifting from demonstration applications to production trust—last year it was about who could prototype faster, this year it's about who dares to let agents modify live code. The one-person unicorn is a marketing phrase; factory-style engineering is the real moat.
Structural judgments belong to technological substitution. The mechanism is: code generation shifts the supply curve outward; if consumption habits remain stuck in scrolling through information feeds, the incremental gains will be missed; once habits change to submitting and publishing, the infrastructure layer will consume the time saved by the model layer. The desire to create is framed as moral progress, reflected in higher deployment density on the ledger. Whether the world gets better is another question, but domain names and token billing have already improved.
ABAB News · Cognitive Laws
The place where the desire to create lands is where the bills rise.
Models write code, and platforms accept deployments.
The consumption era scrolls through information, while the shipping era scrolls through preview links.