Speed of Light: Custom CDN Achieves Sub-200ms Global Latency

By DeHub Team — 2025-01-03

In Q1 2025 we brought a custom CDN into production, reaching sub-200ms latency globally.

Why build rather than buy

Commercial CDNs are excellent and we used them for years. Two things eventually made building worth it.

Cost structure. We give creators 90% of ad revenue on tokenised uploads and pay the full delivery cost. That means delivery cost is effectively our margin, and it is why we cut bandwidth around 60% with compression work back in 2022. At sufficient volume, general-purpose CDN pricing becomes the dominant term in that equation.

Direction of travel. DePIN Phase 1 had already begun distributing delivery across independently operated infrastructure. Routing everything through a single commercial CDN sits awkwardly with that, for the same reason routing all transcoding through one provider does.

The "globally" part

Sub-200ms in North America and Western Europe is not difficult. That is where the infrastructure is, and any competent setup achieves it.

Sub-200ms globally means the markets where it is genuinely hard — and those are exactly the markets our creator partnerships across 15 countries put us in, and the ones the Dubai event opened.

A platform that is fast in London and slow in Manila is not a global platform. It is a Western platform with international users, and they can tell.

Why latency, not bandwidth

For recorded video, bandwidth is what matters and latency is largely hidden by buffering.

Live is the opposite. Latency is the difference between a stream where the chat is part of the conversation and one where viewers are reacting to something the creator said half a minute ago. Once the delay passes a certain point, the interaction stops being live in any sense that matters — and everything we had just built assumed it was.

Animated on-chain tips shipped in the same quarter. That feature is worthless at high latency: the celebration lands after the moment it was responding to, on top of something unrelated. The CDN is what makes it coherent.

What it does not solve

The last mile. A viewer on poor mobile data in a rural area is not helped by a fast edge node, and no amount of CDN work changes their connection.

That is the constraint behind adaptive delivery generally, and behind degrading quality rather than dropping a stream — a principle set in 2022 and unchanged.

Alongside Livepeer and 99.99% uptime, this is the quarter DeHub's streaming stopped having an excuse attached to it.

How the figure is measured

Worth stating, because latency numbers are easy to publish and hard to interpret.

Ours is measured to the edge, across the regions we actually serve, rather than as a global average weighted by traffic. A traffic-weighted average is flattering by construction: most of your traffic is in your strongest regions, so the number reflects where you are already good.

The distinction matters most for exactly the markets this work was for. A platform can report excellent average latency while being slow everywhere its growth is coming from, and nothing in the figure would reveal it.

We report the worst regions rather than hiding them in an average, which produces a less impressive headline and a number that means something.