AI content collapse and what it means for FAST publishers

If you operate or follow the FAST channel ecosystem, the last 18 months have been a structural reset. We have been watching it from the supply side of the exchange. Here is what we are seeing, and what we think it means for the next two years.

The arithmetic just broke

Pluto TV's 2024 cost structure, as far as anyone outside Paramount can reconstruct it, assumed roughly 30% of revenue went to content licensing. That number was already lower than legacy linear cable. It looked like the floor.

It was not the floor. A FAST channel publisher producing AI-generated ambient content today has total content cost under 5% of revenue, including compute, storage, and the salary of the human doing the curation. This is a structural margin advantage, not a cyclical one. It compounds.

What we are seeing on the supply side

Roku reports new FAST channels submitted to their store grew about 4x year-over-year in Q1 2026. Our publisher waitlist tripled in the last 60 days. Most of the new applications come from operators with AI-generated or AI-augmented content libraries.

Quality varies. We have rejected about 60% of waitlist applications in this category, mostly for content quality reasons. The rejections are not because the content is AI-generated. They are because the content is bad in the specific ways that AI-generated content is bad: visual continuity errors, narration that drifts, soundtracks that loop too obviously.

The 40% we accept are hard to distinguish from human-curated content. Audiences do not seem to mind, and in some cases prefer it. The AI sleep stories we are seeing get higher session-completion rates than the comparable human-narrated catalog from a major wellness app. That was not the result we expected.

What this means for advertisers

Inventory is no longer the scarce resource. Discovery is. A campaign budget that bought across 50 FAST channels in 2024 now bids across 200. Most of the new channels are smaller and more thematically specific.

Categorical targeting will not be enough. The IAB taxonomy was not built for a world with 30 channels of variant ambient content under five different brands. Contextual signals, supply path metadata, and per-channel quality scoring all matter more, not less.

Our Resonance engine deprioritizes broad categorical targeting in favor of more specific signals: dwell time per channel, viewer-completion rate per ad pod, lift on attribution events from MMP partners. The plumbing has been there for a while. It is mattering more now.

Two things that will not change

• Audience attention is finite. AI making content cheap does not make people watch more TV. It just flattens the supply curve.
• Ad-supported business models still need impressions that get watched by humans. Falling content costs do not change that math.

What we are doing about it

We are weighting our publisher recruitment toward operators who understand the new economics and run their content with discipline. We are publishing more granular supply-chain metadata to our DSP partners. We are about to announce something that is downstream of all of this. More on that in a few weeks.

The next 24 months will be the most interesting period in CTV programmatic since FAST took off. The supply flood is real. The audience is growing slower than the supply. The exchanges that match well across the noise will earn premium placement. We are working on it.