The Streamers Aren't Worried About the Right Thing

Everyone in the streaming industry is losing sleep over the same nightmare: a world where you can generate a prestige drama for $40 and a GPU rental. Netflix spends $17 billion a year on content. What happens when that moat dries up?

That's the wrong nightmare.

There's a new paper — arxiv 2406.07550 — and it points at something the streamers haven't fully named yet. The threat isn't that AI generates content that competes with their library. The threat is that AI makes the library itself unnecessary.

Here's the distinction, and it matters a lot. When Netflix buys exclusive rights to a show, what they're really buying is a wall. The content exists behind the wall. You pay $15.99 a month to stand on the right side of it. The entire business model is the wall.

AI-generated competing content doesn't destroy the wall. It builds another house outside it. You might leave, but the wall still stands. The exclusive rights still mean something. The licensing fees, the IP, the brand — it all holds.

What this paper is gesturing at is different. It's the scenario where the wall becomes porous — where the informational and experiential value of the content can be extracted, mediated, reproduced, and delivered without anyone touching the original files. Not piracy, exactly. Something harder to litigate and harder to stop.

Think about what you actually want from a show. You want the plot, the characters, the emotional arc, the water-cooler moments. You want to know what happened and to have feelings about it. For a certain class of content consumption — probably a larger class than anyone wants to admit — an AI that can walk you through all of that in twelve minutes, accurately, with emotional fidelity, is a genuine substitute. Not for a prestige cinematic experience, maybe. But for the third procedural drama you're half-watching on a Tuesday.

The content's value was always downstream of its scarcity. Not just physical scarcity — "you can only watch this on our platform" — but experiential scarcity. You had to watch it to have watched it. There was no shortcut.

That second kind of scarcity is what's eroding. And it erodes even if Netflix wins every piracy lawsuit, even if they watermark every frame, even if every DRM scheme holds. The content gets processed by models that learn what it contains, and then those models can produce outputs that functionally deliver what people were paying for.

The streaming executives who've been in rooms talking about "AI-generated content risks" are worried about supply-side disruption — too much content, cheaply made, flooding the market. That's real. But the demand-side disruption is weirder and more fundamental: a world where a meaningful fraction of your potential audience decides that they've, in some sense, already seen it.

The question that the paper raises — and doesn't answer, because nobody can answer it yet — is whether that fraction is 2% or 40%. Whether it's the obsessive recappers who would have found a way anyway, or whether it's a genuine mass market behavior shift.

If it's the latter, the $17 billion content budget isn't buying what anyone thinks it's buying. The wall is still standing. The house is just empty.