What K-Drama Clips Taught Me About Arbitrage

I found a YouTube channel that posts scenes from Korean dramas as Shorts. It was created only a few months ago and already had around 60 million views. The videos were not reviews or explanations. There was no person appearing on camera to talk about the shows. It seemed to be mostly scenes taken from different dramas, cropped vertically, subtitled and uploaded several times a day.

My first question was whether it made money. Sixty million views sounds like money, although YouTube Shorts views are one of those numbers that can look much larger than the business underneath them. The channel might be monetised. It might be rejected for reused content. The owners of the dramas might be claiming the advertising revenue. It might have made a few thousand euros, or almost nothing. There was no obvious way to tell from the outside.

Still, someone had managed to generate 60 million views without making the thing people were watching.

The television productions had already absorbed nearly all the cost. Someone had written the scripts, hired the actors, built the sets, lit the scenes, recorded the sound, edited the episodes, marketed the shows and convinced a streaming service or broadcaster to distribute them. The YouTube channel entered at the very end of that chain. It found one scene where a character was humiliated, rejected, exposed, defended or suddenly revealed to be rich, moved it into a vertical frame and presented it to an audience that might never watch the full drama.

I initially saw this as a slightly dubious content farm. It probably is one. But the more I looked at it, the less exceptional it seemed.

A podcast clipping company does something similar. It does not create the two-hour conversation. It identifies the 30 seconds most likely to circulate without the rest of it. An affiliate site does not make the hotel, laptop or insurance policy. It positions itself between the person searching and the company selling, then takes a percentage for sending the customer across. A dropshipper does not make the product and may never touch it. They find something cheap and badly presented in one place, rename it, photograph or render it more attractively, and put it in front of people who would never search for the original supplier.

Agencies often work this way too, although we use more respectable language for it. A client in New York pays an agency €15,000 for a website. The agency hires a designer and developer in another country for a fraction of that amount. The agency may genuinely add value through sales, direction, accountability and communication, but part of the business is still based on the fact that the same labour is priced differently depending on who packages it and where it is sold.

Public-data companies do not necessarily create the information they sell either. Company registrations, procurement records, property transactions, court decisions and government statistics may already be available for free. The business is often in collecting them, cleaning them, connecting them and turning them into something a person can use without understanding how five government databases work.

Even many AI products are versions of this. A general model produces an answer for a few cents, but a company places a narrow interface around it and charges users €9.99 a week. Sometimes the wrapper is almost embarrassing. Sometimes it removes enough uncertainty and complexity that the customer is happy to pay. The underlying capability is available elsewhere, but not in the exact form, language or context where the user wants it.

All of these businesses are moving something from a place where it is cheap, ignored or difficult to use into a place where it becomes more valuable. The thing being moved is not always a physical product. It might be attention, labour, language, timing, trust, format or information.

The K-drama channel is doing format arbitrage. A scene that originally sits somewhere inside a sixteen-hour series becomes a self-contained 40-second emotional event. The channel removes nearly everything that makes watching a drama difficult. There is no need to choose a series, subscribe to a service, learn the characters or wait four episodes for something to happen. A woman walks into a room, everyone treats her badly, and then they discover she owns the company. The entire emotional mechanism fits inside one scroll.

It is probably worse than the original work, but it is better adapted to the container it has been moved into.

That seems to be one of the internet’s main economic functions. It reveals that the same thing can have very different values in different contexts, then makes it easy for someone to stand between those contexts.

A developer in Estonia has one price when selling directly to a local company and another when their work is packaged through an American consultancy. A government document has almost no commercial value as a PDF buried on a ministry website, but may become extremely valuable when converted into an alert that tells a salesperson which company to call today. A two-hour interview can receive 20,000 views while one sentence cut from it reaches ten million people. A service that took a human three days last year may take an AI agent 20 minutes now, while the customer still evaluates it according to last year’s cost.

I had thought about arbitrage mostly as a financial activity. Something costs €5 in one market and sells for €7 in another, so the trader captures the difference. Online, the markets are less obvious and the prices are often hidden. The spread might exist between languages, between countries, between formats, between what a customer understands and what a supplier can technically provide.

Once I started looking at internet businesses this way, almost everything began to resemble arbitrage.

This also made me think about my idea of building many small AI apps. I had been thinking about Jensen Huang having 60 direct reports, and whether I could have 60 apps instead. Each one could collect what I have learned about a particular subject, become slightly more useful over time and perhaps eventually turn into something valuable. The cost would not be 60 human salaries. It might be a few AI subscriptions, hosting bills and whatever attention I could distribute among them.

There is a more cynical version of the same idea. Build 100 apps quickly, give each one a slightly different name and keyword, and let the App Store decide whether any of them catches demand. Most fail. One ranks for an obscure search term, converts a few users into subscriptions and pays for the rest. The portfolio is not really a collection of deeply considered products. It is a large number of cheap attempts placed inside someone else’s distribution system.

That is extremely close to what the K-drama channel is doing. It uploads many clips and lets YouTube discover which face, scene, title or emotional reversal people cannot stop watching. The uploader does not need to understand the audience completely in advance. The recommendation system runs the experiment.

This logic is becoming difficult to resist because AI has made another attempt so cheap. Another app, another landing page, another video, another directory, another translated version, another service variation. When the cost approaches zero, making only one thing begins to feel almost irrational. Why make one bet when I can make 60?

I am attracted to this partly because it gives one person the surface area of a company. I could have an app about public Wi-Fi, another about design history, another about events, another about health data, another about whatever I happened to spend a week learning. Each could have agents maintaining it, collecting information and making small improvements. Perhaps one would find a real audience. Perhaps the others would still preserve knowledge I would otherwise lose.

But I am not sure whether that is a company, a portfolio, a personal knowledge system or just a more technologically elaborate way of starting things and not finishing them.

The K-drama channel made the weakness of this model easier to see. It has accumulated 60 million views, but what has actually accumulated for the person running it?

They probably do not own the dramas. They do not own the actors’ performances. They do not own the audience relationship. YouTube decides when the channel appears, who sees it and whether it earns anything. If the copyright owners complain, the videos may disappear. If YouTube changes its recommendation system, the views may stop. If another channel copies the same format, there may be very little preventing it from producing the same result.

The channel has produced an impressive number, but the number may not have made the next video more defensible than the first.

This is the question I keep returning to with all these small AI products: what becomes easier or more valuable after the first one exists?

Does the next app reuse a distribution channel I control, or does it start from zero inside the App Store? Does it improve a shared model or database? Do users recognise a common brand? Does every project contribute knowledge that makes the others better? Is there a technical or operational system underneath them that compounds, or am I simply repeating the same cheap experiment 60 times?

It is possible to produce a lot without accumulating much.

An agency can complete 100 websites and still depend on finding the next client through the same cold outreach as the first. A channel can reach millions of viewers who never remember its name. A directory can contain thousands of records copied from public sources but have no correction process, community or historical depth. An AI service can look magical until the customer realises they can get nearly the same result by opening the model directly.

This does not mean the arbitrage is fake. The gap is real. The K-drama uploader has noticed that streaming companies possess thousands of emotionally effective scenes but are not distributing them in the form YouTube currently rewards. The uploader is doing that work instead. The agency really does remove risk for the client. The directory really does make fragmented information usable. The narrow AI interface may be far easier for a normal person than a blank chatbot.

Moving value can itself be valuable. The person who creates the original thing is not automatically the only person who contributes anything. Distribution, selection, trust, translation and presentation all matter. Sometimes they matter more commercially than production.

But I think there is a difference between using arbitrage to enter a market and mistaking the arbitrage for the whole business.

A temporary pricing difference can give a small operator room to begin. AI-assisted work can be sold at prices established when the work required a team. Public information can be packaged before institutions make it accessible themselves. A successful idea from one country can be adapted before local competitors notice it. An overlooked format can grow inside a platform before the original rights holders understand the demand.

The interesting question is what gets built during that period.

An agency can use the margin to develop specialised knowledge, tools and relationships that remain useful when everyone has access to the same AI. A public-data product can accumulate history, corrections and integrations until the cleaned version becomes more valuable than the source. A clip channel could develop an audience that trusts its taste, then move into criticism, discovery, interviews or licensed distribution. A portfolio of small apps could reveal one recurring problem worth building more deeply.

The arbitrage creates the opening, but something else has to grow inside it.

Maybe this is too demanding. Not every online operation needs to become a permanent company. Someone may run a clip channel for a year, make some money and stop when the account is removed. A temporary opportunity can still be a rational use of time. There is no moral requirement to build infrastructure, a brand or a 20-year institution around everything.

But I think I need to know which game I am playing.

If I publish 60 apps, am I hoping to capture temporary App Store search traffic, or am I building a system where each app makes the whole portfolio better? If I use AI to produce client work faster, am I just keeping the difference while it lasts, or am I learning something that becomes difficult to copy? If I build a database, does each additional entry deepen an asset, or does it merely create another page Google may or may not send traffic to?

The distinction is not visible in the beginning. Both versions can look exactly the same from the outside. They may use the same tools, publish the same products and even make the same amount of money. The difference appears later, when the gap begins to close.

The model becomes cheaper for everyone. The platform changes its rules. The supplier starts selling directly. The copyright owner learns how to produce Shorts. The customer discovers the tool. The cost advantage disappears. The copied information becomes easily accessible at the source.

Then you find out whether the business was inside the gap or whether it had used the gap to build something else.

The K-drama channel still fascinates me because the scale is so disproportionate to the apparent effort. Sixty million views from selecting and repackaging scenes someone else spent millions making. It is difficult not to look at that and think there must be thousands of similar opportunities scattered around the internet, expensive things sitting in the wrong format, useful data trapped in the wrong interface, expertise stuck in the wrong language, human-priced work that has quietly become machine-cheap.

There probably are.

But the channel also shows how little ownership can sit underneath an enormous number. Someone else made the drama. YouTube supplied the audience. The copyright owner may collect the money. The operator noticed the gap and moved faster than the institutions around it.

That might be enough for an arbitrage operation.

I am less sure it is enough for the kind of company I want to build.

Connected reading

• Sixty Bets Against Forgetting
• 100 Apps, €0?
• The Prototype Has Become Too Cheap to Trust
• Every Interface Contains a Business Model
• The Business Model Inside the Regulatory Lag

AI authorship note

This essay began with Kris Haamer investigating a fast-growing YouTube channel that repackages K-drama scenes as Shorts, then asking what other online arbitrage operations exist. AI supported the channel and policy research, the comparison across content, labour, data, distribution and AI arbitrage, drafting, editing and the publishing workflow. Kris supplied the original example, the connection to his app-portfolio ideas, the title, the writing style and the editorial direction.