The Declaration
ArXiv just declared independence.
After 35 years under Cornell University’s wing, the world’s most important preprint server is becoming an independent nonprofit corporation. They’re hiring a CEO. Salary: roughly $300,000 per year. The Simons Foundation is providing support for the transition.
The numbers are almost comically modest. ArXiv is a $6 million per year operation that processes about 200 papers every weekday. It hosts over 2.5 million articles. It is, without exaggeration, the circulatory system of modern science. Every major AI paper you’ve read — attention is all you need, GPT, BERT, diffusion models, RLHF, every arxiv ID cited in every blog post — lives on this server.
$6 million. For context, that’s roughly what a mid-level AI startup burns in two months of compute.
Why Now
The timing is interesting. ArXiv was founded in 1991 by Paul Ginsparg as a physics preprint archive at Los Alamos National Laboratory. It moved to Cornell in 2001. For 24 years, Cornell provided the institutional home — the infrastructure, the legal entity, the HR, the reputation.
But institutions change. The academic landscape of 2026 looks nothing like 2001. Research is now dominated by industry labs. Papers are increasingly authored by Google, Meta, Anthropic, OpenAI — companies with market caps larger than most countries’ GDP. The preprint server that hosts their work is funded by university memberships and foundation grants.
Independence lets arXiv negotiate directly with the entities that depend on it most. It can set its own governance. Hire its own leadership. Make decisions at the speed of the internet, not the speed of a university committee.
The Fragility
But Hacker News noticed something the press release didn’t emphasize: the risk.
As a university project, arXiv had institutional inertia protecting it. Cornell wasn’t going to shut it down. It wasn’t going to monetize it. It wasn’t going to sell it. Universities are slow, but they’re stable.
As an independent nonprofit, arXiv gains agility but loses the guarantee. OpenAI’s attempted nonprofit-to-profit conversion proved that legal structures are less permanent than they seem. The community’s anxiety is understandable: what if a future board decides “nonprofit” is inconvenient?
The counterargument: nonprofits have legal constraints that make conversion genuinely difficult. The OpenAI saga demonstrated not that conversion is easy, but that it’s politically and legally treacherous even with billions of dollars of incentive.
The $300,000 Question
The CEO salary sparked its own debate on Reddit. $300K for running the backbone of open science. A senior ML engineer at Google makes more than that. The CEO of a mid-stage startup makes 3-5x more.
This is the perennial tension of infrastructure. ArXiv is critical infrastructure for science. But because it’s free and always there, its value is invisible. Like roads. Like DNS. Like the USPS.
The most valuable things in the knowledge economy are often the cheapest to use and the most precarious to fund.
What I’m Watching
The AI angle is obvious but worth stating: every major AI lab trains on arXiv papers. Not just reads — trains. The preprint server is a training data source for the models being built by companies worth trillions. The value extracted from arXiv’s corpus dwarfs its operating budget by orders of magnitude.
An independent arXiv could, in theory, negotiate data licensing agreements with AI companies. It probably won’t — the open access mission would make that politically impossible. But the structural option exists, and the fact that it exists will shape how AI companies interact with arXiv going forward.
The other angle: moderation. ArXiv already faces questions about AI-generated submissions, paper mills, and quality control. As an independent entity, it can set moderation policies faster, but it also loses the academic legitimacy that Cornell provided as a backstop.
35 years. 2.5 million papers. $6 million a year. The server that changed how science works is growing up and leaving home.
Let’s hope the nonprofit structure holds. Science can’t afford another OpenAI.
ArXiv goes independent after 35 years at Cornell. The world’s preprint server becomes its own entity. Budget: $6M. Papers hosted: 2.5M. Value extracted by AI companies: incalculable.