Licensing is All You Need

OpenAI launched Sora this week, and Sora — and possibly OpenAI itself — are structurally doomed.

Sam Altman has discovered that the hard problems in media are legal. They're about licensing. They're about revenue sharing. And OpenAI is spectacularly unequipped to handle either.

The neural network side is already commoditized. Look at text generation: GPT faces Anthropic, Google, Meta's open-source Llama, Mistral, and a dozen Chinese models. The same bitter lesson will repeat with every neural network. The moat isn't the model. It never was.

Think about what Daniel Ek actually did when he built Spotify. He could have scraped every song off the internet in 2008, hosted them on his servers, and told the record labels to pound sand. Spotify would have lasted about six months. Instead he negotiated every deal, territory by territory, label by label, for years. Those grinding multi-year negotiations with every label and every publisher — that was the work. The streaming technology was straightforward by comparison. Licensing wasn't an afterthought to the business. It was the business.

OpenAI did the opposite. They built their entire operation on uncompensated content and are now discovering that content owners want their cut. Every creator who sees their style replicated without payment becomes an enemy. Every artist whose work went into the training data without permission becomes a potential litigant. You don't build on that and call it a foundation.

And here's where the numbers get ugly. OpenAI has raised capital at valuations that assume software margins — 80%, 90% gross margins, beautiful SaaS economics. Content businesses don't work like that. Spotify runs at 25% gross margins. Netflix hovers around 40%, and that's after decades of optimization. Everyone can see that the promise of AI replacing every job isn't happening. That leaves OpenAI with a normal content business, which is exactly why Sam launched Sora — but that move puts all the IP questions front and center. You can't casually violate copyright if you're not delivering on the AI-machine-god story and are just running a for-profit media company. Once OpenAI starts revenue sharing — and they will, because the alternative is endless litigation — their unit economics collapse. You can't spend trillions on datacenters on Spotify margins.

It gets worse. If they admit they need to license content going forward, they've implicitly admitted they should have licensed everything they trained on. GPT-4, DALL-E, all of it. The liability tail could stretch into the tens of billions.

Daniel Ek picked this exact week to step down as Spotify CEO, which is almost poetic. After two decades of building the definitive content licensing infrastructure, he's leaving just as every AI company is about to discover they need exactly his playbook: patient relationship-building, careful negotiation, and the understanding that content businesses are fundamentally about the people who make the content. Meanwhile TikTok is getting banned and handed to a database boomer. The opportunity here is enormous.

The companies that win will do what Spotify did — the hard, boring, essential work of making peace with content owners. They'll accept lower margins, share revenue, and build without a legal sword hanging over them. That's actually a defensible business. No creator boycotts, no regulatory overhang, just clean growth built on mutual benefit.

The AI industry is about to learn what the music industry learned twenty years ago: you can't disrupt your way around copyright law. You negotiate your way through it. OpenAI is bringing algorithms to a licensing fight.