Why Open Source AI Should Be Not-for-Profit

The story told about Nvidia's success in the chipmaking space in the early 2000s is a fascinating case study in business strategy. On the one hand, Nvidia's release of the proprietary CUDA software layer for their GPU hardware helped bring graphics card programming into the hands of a wider audience. On the other, it created a monopolistic grip on the GPU market, preparing Nvidia for the AI boom that would follow.

Giving out GPUs to university labs for free, all the while holding exclusive rights to the software layer, Nvidia grew explosively thanks to AI in the last few years, much of which was powered by researchers who became dependent on the ecosystem.

When private companies own large segments of the research ecosystem, what is actually being encouraged? When companies give out platforms or infrastructure is it to encourage science for humanity or science dependent on their tech flywheel?

AI and AI derivative research today is so heavily dependent on infrastructure owned by private companies that it seems most research today is purchasing and giving user data to a handful of corporations. An AI lab might purchase GPU machines, cloud credits, hire human experts through startup platforms, and pay monthly subscriptions to coding agents, usually supported by federal or foundation grants creating a funnel of public to private fund transfer. We are already deeply embedded in this ecosystem, even without accounting for their ownership of open-source science.

Is it really "open"? Open source does not mean open platform.

Everyone seems to have followed with zeal OpenAI's conversion from a nonprofit org to a Public Benefit Corporation. But people seem to have missed the politics of open source AI. Under the guise of openness, transparency, and "free" public good collaboration, the open source AI community grew especially as some breakthrough projects such as Deepseek gained wide attention. But who owns it? What are their incentives? Why are these large profit-seeking corporations pushing open source?

• Open source is not the same as open platform: privately owned platforms are closed structures that can arbitrarily take down or selectively promote models and datasets. These systems encourage huge participatory counts in numbers but target spotlight repositories for dramatic effect and media attention. It is built to give the illusion that anyone could be the `next deepseek’ while capitalizing on free AI uploaded and pushed by public (and private) labs. Nothing prevents private companies from making these calls and they do naturally as part of their incentive structure.
• It is difficult for privately owned entities to represent something open source, let alone trust their representation of the ``state of open source.’’ Can we have the CEO of a private company talk to policymakers about open source AI and believe that’s an unbiased representation? It gets harder for the public and policymakers to have healthy discussions surrounding private vs. public, open vs. closed AI when the open source platform has clear end goals of direct or indirect profit and revenue maximizing.

A misalignment between contributors and beneficiaries

The vast majority of open source AI contributions have come from the academic community: the PhD students, the postdocs, academics and their labs. It is the natural and expected practice as being a member of the academic ecosystem to share and build upon one another's work. Just as peer-review and healthy transparent scholarly debates have helped humanity progress in science for hundreds of years. For a long time, this was a system built by and maintained primarily by non-profit organizations and institutions such as libraries, universities, research groups, academic conventions, and conferences.

Open sourcing is a concept that has existed for as long as patenting has existed – for source code even longer. In many cases, it was for the sake of humanity such as with Penicillin and the Polio vaccine, for which the creators had made ethical decisions to not patent for the betterment of the world (1928, 1955). In some, a government mandate to encourage intra-national collaboration or sharing of a public good. AT&T Bell Labs released the license and blueprints of the first-ever transistor as a public utility in response to U.S. antitrust pressures from the U.S. government (1952). Similarly, during World War I, the U.S. government stepped in to force American aviation companies to pool patents into a transparent collective for fee-based use to stop legal chokeholds and encourage aviation innovation and production for war (1917).

Today the spirit of open source AI is vastly different from the purpose and vein of open source science preceding it. Is it suspicious why companies themselves are ``marching for open source’’? Why are for-profit companies encouraging academics, students, engineers to tinker with their stuff for free on a privately owned platform? It’s still a wonder, why do companies openly release their mult-million dollar models open weight and open source? Do Chinese companies want to undercut American competitive edge?

If the discourse of science is owned by a handful of private companies is science really open?

Towards actually nonprofit OpenAI, take 2 or 3?

When OpenAI was first founded as a nonprofit, it, like Deepmind, had the public vision of creating AI for the betterment of humanity. After many years, much has changed – for many companies that projected that vision.

Project Clover is an attempt to revisit that vision. Project Clover is an initiative started by former Hugging Face researchers and academics who realized the limitations of a for-profit open source platform. Nonprofit goods are by no means perfect, but privately owned platforms have concentrated incentives and power structures that are not public serving or prioritize the interests of `science.’ We are working towards:

• A science-first platform should be community-led. This means transparent fundraising and operations; and agenda setting that prioritizes scientific and AI progress (not whether it benefits the interests of stakeholders)
• Search-first platform that gives changes for longtail contributions to be spotlighted, acknowledged, and used
• More sophisticated licensing that allows proprietary and semi-proprietary datasets and model including rare collections to be used with fair remuneration systems
• Operation by consortium. Participating private, nonprofit, public institutions should not own or overpower others.
• Specifically designed to allow and provide unobstructed and open sharing and feedback, encourage free speech
• Public research of AI’s far-reaching political, military, socio-cultural impact

Let's work towards a better future for open source!