Micro-Drama Got Cheap to Make. Not to Show.

I’ve been asking the same question of everyone I meet who runs a streaming service, and I’ve lost count of how many times I’ve asked it:

What does it cost you to deliver a thousand hours of video?

Not the total CDN bill. That one they have, or finance does. I mean the unit, and I mean that unit specifically. Hernán López and his firm Owl & Co built a whole framework around streaming unit economics, and one of the metrics in it is RPMH, revenue per thousand hours. In their Vertical Economy Report 1H26 ↗ (opens in a new tab), they use it to compare how well TikTok, Reels and the rest turn attention into money. It’s a good metric. Hours are what your audience gives you and revenue is what you get back for them.

It only measures one side though. So you know what an hour of watching earns you, and you don’t know what it cost you to hand over, which means you don’t really know what it’s worth.

Hardly anyone carries that second number around, and when you hear why, it’s fair enough.

For most of the history of subscription streaming it was too small to matter. Someone pays you thirty-five dollars a month, you ship them video all month, and the shipping is a sliver of what they’re worth. It’s noise against the subscription. Nobody builds reporting around noise. So it sits inside a cloud bill somebody opens once a quarter, bundled in with forty other line items, and everyone gets on with their day.

That held up for about fifteen years. I don’t think it holds anymore, and last week I finally saw the reason written down by someone other than me.

Owl & Co put out the first serious attempt to size vertical video as a whole. If that’s not your corner of the world, vertical drama is the category of phone-shaped shows. Two-minute episodes, sixty or eighty to a season, apps like ReelShort and DramaBox. They came back with $150 billion outside China this year.

The number that surprised me was on page nine.

Three free, ad-supported apps, Melolo, FreeReels and PineDrama, went from 15% of the category’s total time spent to 83%, in six quarters. Advertising across the category is up more than fourfold since the start of last year, and the report is careful to note that the actual revenue is still mostly sitting with the paid incumbents.

So the watching moved to free, and the money hasn’t followed it yet.

Now go back to the reason nobody tracks the cost side. It was noise against a thirty-five dollar subscription. Against a viewer who pays you nothing it isn’t noise, it’s most of what that viewer costs you. The file is the same size either way. Your CDN has no idea whether the person watching has a card on file.

And there’s a worse version of that viewer. The one who watches the free episodes at the front of a series, hits the paywall, and closes the app. You bought that click with an ad. Then you paid to stream them the free part. Then they left.

I want to be careful here, because this is where people usually overclaim. Free tiers are correct. Every media business I’ve worked in built one, and vertical is right to be doing it. The point isn’t that free is a mistake. It’s that free changes which costs matter, and this one has been sitting in the “doesn’t matter” pile for years on the strength of an assumption that just stopped being true.

There’s a second thing in the report pushing the same way. New releases went from 1,659 in the first quarter of last year to 5,145 last quarter. In that quarter, releases were up 25% and total time spent went down 4%. AI-assisted productions start around thirty thousand dollars now, so making more of them is a rational response to them getting cheap. Owl & Co say volume can be cannibalistic, which is polite. ByteDance ended up pulling thousands of AI series off Hongguo because they were dragging completion rates down.

Making an episode got dramatically cheaper this year. Moving one onto a phone costs what it cost in January.

If you’re reading this and you don’t run a vertical app, I’d still park it somewhere. Vertical is just where it’s easiest to see, because the whole category flipped in six quarters instead of six years. If you run an ad tier, a free window, a FAST channel, you’re on the same curve with a longer runway.

You know your RPMH, or you can have it by lunch. What does a thousand viewing hours cost you to deliver? And what does that look like on the night everybody shows up at once?

Hit me up ↗ (opens in a new tab) if you’ve got both halves, I’d like to see them.

(I work on the delivery side of this, so weigh me accordingly. The report is Owl & Co’s, it’s public, and it’s one of the best maps of this category anyone has published.)