Cloudflare video services. Why now and what’s next

Discover how Cloudflare grew its video services and WebRTC offerings in 2025 and where they are now in relation to the industry.

Cloudflare is considered by many the 4th IaaS vendor (after Amazon, Microsoft and Google). This year, they made quite a few steps in real time communications that shows real interest and investments in the domain of video processing at low latency.

What I want to do in this article, is to go over the announcements made, and try to figure out together what does that mean and where is the industry headed.

The big 3 and WebRTC

Before we dive into Cloudflare, it is important we check what the “other” big IaaS vendors have when it comes to video and real time communications. We will focus on low latencies – below 2 seconds, which leaves out things like YouTube out of this.

Amazon

• Amazon Chime SDK – Amazon’s CPaaS offering for developers
• Amazon Interactive Video Service – a low latency streaming API offering
• Amazon Luna – cloud gaming infrastructure from Amazon

Microsoft

• Edge – web browser built on top of Google’s Chromium. Not much is done here with regards to WebRT as it rely mostly on Google here
• Microsoft Teams – a full communication service for companies. It makes use of WebRTC since it works in web browsers
• Azure Communication Services – the Microsoft Teams CPaaS variant, offering the infrastructure for communication services as an API to integrate with
• Xbox Cloud Gaming – cloud gaming infrastructure for Xbox

Google

The creator and maintainer of WebRTC in a way… and not much to show for it

• libwebrtc – the main WebRTC client-side implementation out there, that is also integrated into ALL web browsers. Means to an end for Google and nothing more
• Chrome – the browser…
• Google Meet – video conferencing / meetings service

–

Amazon and Microsoft have an API/developers play with WebRTC. Google not as much.

Cloudflare seems to be taking this to the extreme. Let’s review how exactly.

TURN servers

In 2021, Cloudflare announced their first service related to WebRTC – managed TURN service.

The main differentiators Cloudflare brought with it?

• Lower price point per Gb consumed
• More regions (300+ today versus less than 50 in all other competitors)
• Use of Anycast

As a first step, this is a simple one to make for IaaS vendors and one that relies on scale.

The fact that none of the other IaaS vendors have launched anything similar while all of them do have TURN servers deployed globally at scale for their other services is quite interesting. They likely don’t see the monetary value in offering that, which I find weird, considering the 100s of other managed services they do offer for developers already.

Calls and Streams

A year later, in 2022, Cloudflare announced Calls and Streams.

Calls was a kind of a managed distributed cloud SFU that developers could integrate with using “pure” WebRTC directly to build their apps. And Streams (its live variant) was about live streaming services.

The problem with this was the lack of a client side SDK. Developers had to figure out the signaling and how WebRTC was actually implemented on the Cloudflare SFU to get it to work. Things got murkier when the time came for optimizing things – you can’t really optimize a client SDK without optimizing the server – they are closely tied. This is why, for example, when you go use an open source SFU like mediasoup, there’s also a client side SDK along with it that you will be using.

I’ve written about this a few months back in my tip & offer emails.

The missing client SDK: acquiring Dyte

Fast forward 3 years and finally Cloudflare decided to do something about it.

Instead of implementing and launching their own client SDK, they decided to acquihire Dyte.

Dyte was another CPaaS vendor from India, which could be considered as a competitor of Cloudflare Calls. They were either using Cloudflare’s SFU to begin with at Dyte, or more likely – post acquisition they have/are shifting their infrastructure to use Cloudflare 👉 keeping their client SDK and UI facade almost as is. Or maybe sunsetting it quietly, now that Cloudflare RealtimeKit has been announced.

Voice AI

There were multiple interesting announcements around Voice AI from Cloudflare this year. It started as part of the Dyte acquisition. In the same announcement, they wrapped in hints around partnerships with ElevenLabs and Hugging Face.

The real deal came only recently when they announced 4 different services/offerings:

1. Cloudflare Realtime Agents – orchestration for Voice AI pipelines. You can think of it as a simplified (less flexible) alternative to Pipecat Cloud or LiveKit Agents in the cloud
2. Pipe raw WebRTC audio as PCM in Workers – an Opus-G.711 gateway that sits on Cloudflare’s edge network dedicated for connecting WebRTC from clients to voice services that require the use of G.711 (=PCM). It also converts from WebRTC to WebSocket for the exact same reason
3. Workers AI WebSocket support – Cloudflare Workers AI is a set of AI algorithms you can run on the Cloudflare network as part of your AI pipeline. They now support WebSocket (and not only HTTPS) because that’s what’s needed for realtime. Interestingly, they made an effort to add Daily’s open source PipeCat SmartTurn v2 for turn detection in conversations, with Daily just releasing SmartTurn v3 (smaller, faster, better). How quickly will Cloudflare integrate with it is yet to be seen
4. Deepgram on Workers AI – Deepgram STT and TTS are now integrated into Cloudflare Workers AI as well. Somehow, this could have been made part of the 3rd announcement, but the decision was to split it

What do we have here?

• An AI framework that is proprietary and cloud hosted by Cloudflare. This is needed by CPaaS vendors to compete in this market. In Cloudflare’s case, the result is interesting since it also covers IaaS and not only CPaaS (=same functionality and implementation, but a different mindset of the developer looking for a solution)
• Utility gateway that is great in reducing headaches when connecting WebRTC clients to online AI services
• A library of AI related algorithms that are preloaded on the platform via the Cloudflare Workers AI. With the introduction of turn detection via PipeCat’s implementation and STT/TTS via Deepgram

In a way, Cloudflare outdid all other IaaS vendors in offering a Voice AI framework targeted directly at developers.

MoQ

MoQ stands for Media Over QUIC. It is a WebTransport based protocol that is being defined for enabling simple live streaming solutions.

The main advantage of MoQ? It is intended to work with current day CDN architectures. So that CDNs can more easily be updated to support MoQ than it is to support something like WebRTC.

Where does MoQ suffer? It doesn’t offer the richness and power of WebRTC when it comes to bidirectional media. Think of echo cancellation as an example.

A week prior to the Voice AI announcements, Cloudflare announced support for MoQ. What they did was launch a MoQ relay service on their global infrastructure for developers to tinker with. This isn’t GA by any means – it is in beta with a protocol that is still being standardized (which is just fine).

In a way, this gives Cloudflare the complete gamut of of supported video streaming protocols in Cloudflare Stream:

• HLS and DASH along with their low latency variant
• WebRTC with WHIP and WHEP interfaces for live streaming (in beta)
• And now MoQ for the next generation of what live streaming is

At the same time, most of the IaaS vendors only do HLS and DASH. For the time being.

No big customers (yet?)

What Cloudflare is lacking is success stories with big customers.

I know of no publicly available name of a vendor that uses Cloudflare’s WebRTC infrastructure for his large-scale deployment.

It doesn’t mean that one doesn’t exist, but it means none has come out publicly. And that’s an issue when looking for trust from new customers who can pick and choose solutions from other vendors in a market that is evolving all the time and is highly competitive.

Cloudflare should remedy this quickly. Either by pushing some of its customers – big and small – to agree to success stories that will be published and promoted. Or by getting a big customer to use their service…

Conferencing versus Streaming

There’s a big difference between conferencing solutions and streaming ones.

This is why MoQ isn’t going to replace WebRTC. At least not in the main use cases where WebRTC is being used. WebRTC is most suitable for conferencing while MoQ is best suited for Streaming.

Does it mean you should switch to using MoQ now?

Maybe. Just note that it is supported only on Chrome, so you might as well wait if what you want is coverage across web browsers.

In the case of Cloudflare, they are making that distinction with these being two separate products:

• Stream, where you can find their streaming products. But not MoQ – that’s handled separately as a playground more than a beta. WebRTC live streaming support is still in beta here
• Realtime, which is where conferencing lives. It also wraps into it the Voice AI capabilities, split between RealtimeKit and Agents (both in beta)

There’s a separation of technologies, mindsets and infrastructure implementations here that make sense.

It is why most vendors who specialize in streaming or even live streaming don’t do conferencing anymore (Dolby did both and shut down their conferencing CPaaS).

Is Cloudflare chewing more than they can swallow

Which brings me to this question:

Is Cloudflare chewing more than they can swallow?

Cloudflare is definitely a large company with resources. It is pouring funding into these spaces. It made an acquihire in this domain.

They are now competing in parallel against multiple vendors each specializing or focusing on one of these two – either Streaming or Conferencing.

Can they be successful in both at the same time?

Conferencing is becoming more competitive again, especially in the domain of Voice AI.

Streams with MoQ requires a paradigm shift in the technologies used and also adds the standardization effort and instabilities that goes along with it.

Doing these all things at the same time is bound to be exhaustive – both in engineering resources as well as managerial ones.

I do hope Cloudflare succeeds in it all – their angle here is innovative and interesting enough to make a difference and enrich our ecosystem.