How AI Is Breaking Open Source Revenue While Using Open Source as Training Data

Adam Wathan described Tailwind's situation plainly: "Tailwind is growing faster than it ever has and is bigger than it ever has been, and our revenue is down close to 80%."

The reason: developers no longer read documentation. They ask AI.

This is not a Tailwind problem. It's a structural problem with how open source projects monetize, and AI has collapsed the economic foundation that made documentation-adjacent monetization possible.

The Revenue Model That Broke

Most OSS projects with revenue followed recognizable patterns: paid hosting or cloud versions, courses and tutorials, Pro tiers with advanced features, consulting attached to deep expertise.

All of these depend on developer intent — the developer wanting to learn or use the tool deeply enough to invest time or money. That intent used to manifest as traffic: documentation visits, tutorial views, forum participation.

AI replaced the documentation visit. The developer types their question into a chat interface. The AI answers using knowledge derived from the documentation, the Stack Overflow posts, the blog tutorials — all written by humans, mostly unpaid, and all used to train models that now absorb the traffic that documentation sites used to receive.

The knowledge is still being used. The economic signal is gone.

The Compounding Problem

OSS maintainers are volunteers or small teams on thin commercial margins. The economics of maintaining a popular project are already precarious — the GitHub Blog noted that the gap between project participants and actual maintainers is growing at record rates as new developers flood the ecosystem.

The maintenance burden scales with usage. The funding doesn't.

AI adoption accelerated both ends of this problem simultaneously: usage of popular libraries increased as AI tools integrated them by default, while the documentation traffic that supported revenue-generating products around those libraries collapsed.

More users. More maintenance burden. Less revenue to support it.

What Makes This Different From Previous Crises

Open source has always had a sustainability problem. This one is structurally different for a specific reason: the value extraction is invisible.

When a company uses OSS without contributing, the usage is at least attributable. Someone can see the download count, the GitHub stars, the deployment metrics. The moral case for contribution or sponsorship is legible.

When AI models train on OSS code, documentation, and community knowledge, the extraction is structural and one-time. The outputs are model weights, not artifacts that trace back to source. The economic argument for "you should pay the people whose work made this possible" becomes nearly impossible to make because the chain of value is not traceable.

Tailwind's revenue dropped because developers use AI to answer questions about Tailwind. But the AI's ability to answer those questions came from the documentation Wathan and his team spent years writing. The extraction happened once, in the training run. The revenue loss is ongoing.

The Scale of What's at Stake

The infrastructure that most software runs on is maintained by a small number of people who are often underfunded and frequently burning out. This was true before AI. It's more acute now.

When a critical OSS dependency breaks or goes unmaintained, the downstream cost is enormous — and distributed across thousands of companies who benefited from it for free. That cost is invisible until it materializes. The Eclipse Foundation's 2026 outlook flagged this explicitly: critical infrastructure is maintained by individuals with no institutional backing, and the sustainability gap is widening.

The companies relying on that infrastructure have no plan for what happens when the maintainer stops.

What's Actually Changing

Some approaches are starting to work. Government and nonprofit funding of critical OSS infrastructure is growing — recognizing it as digital public infrastructure rather than a hobby project that should self-fund. Long-term funding partnerships are replacing one-time donations. Frontend Masters committed $50,000 to open source support in 2026 as a direct response to these pressures.

These are better than nothing. They're not sufficient at scale.

The projects that will survive are the ones with clear commercial differentiation beyond documentation and community: hosted services with meaningful operational value, enterprise features that require ongoing human expertise, or products where the OSS version is genuinely a demo of a paid product rather than the product itself.

Projects maintained by the goodwill of a single engineer, funded by the attention economy of documentation traffic, are in structural decline.

The Question Nobody Is Answering

AI companies trained their models on OSS code and documentation. Those models now generate billions in revenue. The OSS projects that provided the training data receive none of it by default.

The standard position is that training on publicly available data is fair use. That position is being litigated.

But the simpler version of the question doesn't require legal resolution: if the value chain is OSS documentation → AI training → AI product revenue, and the project that wrote the documentation sees revenue fall 80% while the AI product grows — that's not a legal question. It's a design question about what open source infrastructure looks like in ten years.

The people building on AI today don't have a good answer. Most aren't asking.