November 20, 2026

Sub-Second Video Is a Different Engineering Problem

person using computer keyboard

Ask two teams to build a live video product, and you’ll get two answers with nothing in common. The reason for this is latency, which has been at the top of the list of technical headaches for live video developers for years.

Why Segmented Streaming Can’t Get There

Source: Unsplash

HTTP Live Streaming, or HLS, was introduced by Apple in 2009, and it’s how most people watch video online. HLS cuts video into files of around two to ten seconds. It then publishes them to a manifest and lets the player buffer ahead to ensure continuous video.

That design itself is why it scales. Each segment is a static file, so a CDN caches it at the edge and serves millions of viewers from one copy.

There’s a cost, though. The cost is delay. Standard HLS has a delay of roughly 20 to 30 seconds. However, Apple did launch low-latency HLS in 2019, which managed to lower the delay to around 5 seconds.

The Sub-Second Stack

Source: Unsplash

WebRTC works the other way round. Standardised by W3C and the IETF, it was built for conversation rather than broadcast.

With it, there are no segments. WebRTC sends individual video frames over UDP, and its jitter buffers are measured in milliseconds. Glass-to-glass latency is usually around the 200- to 500-millisecond region.

Unfortunately, though, this comes at a cost. With no cacheable files, WebRTC can’t ride on a standard CDN. Scale then comes from clusters of selective forwarding units, so per-viewer costs are far higher.

Which is Best?

Neither is better on its own. However, with cost control being a concern for 38% of developers, HLS is probably the better choice.

But again, WebRTC, though it costs a lot more per concurrent customer, it has less latency, boosting the live streaming experience for the viewer.

Who Uses What?

Every live video decision comes down to a trade between delay and cost. Segmented streaming is cheap and slow, and real-time streaming is fast and expensive. Both serve two different purposes.

Live auctions, for example, use WebRTC for bidders and presenters but use HLS for viewers. The reason is that bidders need to be able to bid quickly and see updates to prices. For viewers, on the other hand, these immediate updates are less important.

Industrial teleoperation is another example of WebRTC. Remote crane operators, for instance, need instant streams with little to no latency so they can perform their job with accuracy.

Live dealer gaming commonly uses low-latency HLS, though. Online casino games like Gates of Olympus Roulette by Pragmatic Play use glass-to-glass latency as low as 1.5 to 2 seconds. This isn’t WebRTC and is more on the low-latency HLS side of things.

As you can see, one is not better than the other. It comes down to its application, which usually depends on latency or cost.

What Comes Next

Media over QUIC is the next big thing. It’s built using QUIC and WebTransport and aims to hold a sub-second delivery at a much more affordable rate.

Though most CDNs are still delivering using HLS, when QUIC matures and becomes the next go-to option, we imagine a lot of live stream developers will begin adopting it.