Personal channels in a box

For years, every streaming platform has been trying to generate more revenue from content they already have.

Archives are packed with great material. Sports clips, concerts, local programming, news, documentaries, interviews, kids’ content, long-tail shows: all of it’s valuable. 

That’s why hundreds or even thousands of virtual channels, each one assembled from existing content, each one targeted to a different audience, use case, region, sponsor, or viewer profile, can be a goldmine. But can you run it?

That’s the question behind our “channels in the box” scaling exercise.‍

Why personalized channels grabbed our attention‍

Personalized channels open up a different path to package and monetize video.

• A broadcaster could create individual music channels based on a viewer’s taste. No need to wait for another pop song to end before seeing your favorite "Freak on the Leash” video.
• A sports platform could turn its archive into team-specific, player-specific, or moment-specific streams. For some, following Max Verstappen’s races 24/7 sounds like a dream.
• A media company could build channels around moods, genres, regions, events, sponsors, or subscriber segments. A breakup playlist can go beyond music.
• A streaming platform could create ad-supported channels in which content and ad breaks are tailored to a specific audience group. Running a channel for millennials about feeding your dog, tweaking your sleep routine, or running your first marathon is simpler than you think.

All of this is a cinch for organizations with deep content libraries. They already own the content. They just need to make it work and bring in more money.

Unified Virtual Channel becomes an ideal solution for running such channels. The solution makes assembling linear streams from existing assets (both VOD and live) possible. That’s without creating a traditional broadcast workflow for every new idea (for example, with AI). Teams can test more channel concepts, serve more audience segments, and create new monetization opportunities without building everything from scratch.‍

From “Can we do this?” to “How far can we take it?”

The question we often hear from customers sounds simple.

‍How many virtual channels can we run on a single server?

Behind that question, though, lies a practical one.‍

How many personalized or ultra-personalized channels could work reliably at scale?

This experiment wasn’t based on just technical curiosity, but on a specific commercial request from one of our customers. 

So we decided to test it properly and run a scaling exercise. We examined the real behavior of virtual channels under load: many channels, real player-like requests, different streaming formats, response times, bottlenecks, and observability.

The project name was simple: 10,000 channels in a box.‍

What we learned about scaling channels

Not all channels act the same way.

A DASH channel and an HLS channel create different request patterns. A fully personalized channel with one viewer per channel behaves differently than a channel where many viewers share the same output. Shorter segments create different load patterns than longer segments. Higher bitrate content can shift the bottleneck from CPU to network capacity.

So the answer to “how many channels can fit in the box?” isn't a clean number.

In tested scenarios, the setup supported thousands of virtual channels with stable response times. Some configurations were stable at lower numbers. Some could go much higher.

Based on the test environment and extrapolation to the target server, the realistic range can swing from several thousand to tens of thousands of virtual channels, depending on the workflow.

That gives a much better starting point for asking specific questions.

• Are these HLS or DASH channels?
• Are they fully personalized or audience-bucketed?
• What segment duration do you use?
• What bitrate profile are we talking about?
• Where will this run? Cloud, on-prem, edge, or a mixed setup?
• What is the expected concurrency?
• How much of the archive will be reused across channels?

These questions turn a big personalization idea into a specific architecture. They lead to real answers.‍

Why observability matters

Another big part of the exercise was observability.

When you scale personalized channels, it’s not enough to know that “something is slow.” 

You need to know exactly what causes the slowdown. Is it manifest generation, the origin under pressure, or storage? Is it TCP connections hitting a limit? An overloaded system, a misconfigured component, what? This is where proper metrics, logs, and traces become important.

If you want to run personalized channels at scale, you need confidence. You need to know what your infrastructure is doing. You need to see where the cost goes. You need to know whether you can add more channels, more viewers, or more content without breaking the experience.

That’s the boring but necessary side of personalization.

We tested the scalability, and yes, it scales

The test setup was built to answer a few practical questions:

• Can we create thousands of channels?
• Can we simulate real player behavior?
• Can the system respond fast enough for actual streaming workflows?
• Can we see where the bottlenecks are?
• Can we understand the difference between HLS and DASH behavior?
• Can this exercise help us give customers a better answer when they ask about scale and cost?

The point was never to create a magic number that applies to every customer. Streaming never works like that. Channel type, segment duration, bitrate, format, personalization level, caching strategy, and deployment model all matter.

In order to achieve reproducible results, we ran everything in AWS EC2, using 32 vCPU c8g.8xlarge instances.

Results
In the tested scenarios, Unified Virtual Channel successfully supported thousands of playlist-based channels under simulated player load.

However, the number of channels depends on factors.

HLS
Stable runs included 1,000 HLS channels with 2-second segments, 2,000 full HLS channels with 4-second segments, and 2,500 personalized HLS channels, each with one viewer. So it really depends on the settings you choose.

DASH
Dash is becoming more scalable. Virtual Channel supports 5,000 DASH channels and serves 99% of requests within the target response window. 

The broader takeaway was not a single fixed benchmark but a practical range. Depending on format, segment duration, bitrate, caching strategy, and level of personalization, a single optimized setup can support anything from several thousand to tens of thousands of virtual channels.

The result varies, and we understand what influences it.‍

What the results mean for media companies

For CTOs, product leaders, and media executives, the interesting part isn’t the benchmark, but the capability.

Instead of thinking in terms of a few big channels, think in terms of targeted experiences. You can create temporary channels around events. You can reuse archive content for niche streams. You can test audience-specific programming. And you can build channels for different markets while reusing the same infrastructure. 

This means you get more options for advertising, sponsorship, and distribution partners. And it means more revenue.

What’s next? PoCs and projects‍

The future of channels may be much more personal, much more flexible, and much more scalable than the old model allowed.

The “10,000 channels in a box” exercise gives us a stronger practical foundation for talking about personalized channels, archive monetization, and virtual channel scaling. It also gives us a better way to work with customers who want to explore these models seriously.

So let’s try it in practice together, shall we?