Use Media over QUIC for low latency

Media over QUIC is an attempt to simplify a problem that video and AI systems usually solve with several different protocols at once. The IETF’s current transport draft defines MOQ as a publish-subscribe protocol that runs over QUIC and WebTransport, and says it uses streams, datagrams, priorities and partial reliability to deliver data with low latency. That matters because the protocol is not limited to traditional video playback. The draft says MOQ is “media agnostic,” which means the same transport can carry live video, model outputs, control messages and other time-sensitive objects in one system. ### Why are people talking about MOQ now? The IETF MOQ working group has accelerated work in 2026, with transport draft-18 posted on May 12 and related drafts covering streaming formats, relay behavior and multimodal feedback. (datatracker.ietf.org) The working group page lists active documents on transport, streaming formats, privacy, secure objects and feedback extensions. A recent technical post pointed to MOQ as a better fit for real-time AI pipelines on unstable networks because it can keep latency low without forcing every packet to be treated the same. That claim lines up with the draft’s use of priorities and partial reliability, which let applications favor the newest or most important objects instead of waiting for stale ones. (datatracker.ietf.org) ### What does MOQ do that older media stacks do not? Cloudflare said in a technical post that developers have long had to trade off between WebRTC-style interactivity and HLS or DASH-style scale. It described MOQ as a single foundation for sub-second, interactive streaming at scale, rather than a stack stitched together from separate ingest, relay and playback protocols. (datatracker.ietf.org) The transport draft describes the architecture in similar terms. Publishers send named tracks and objects, subscribers request what they want, and relays can forward or cache those objects in the middle. That relay model is one reason MOQ is being positioned for low-latency distribution beyond a single peer-to-peer session. (blog.cloudflare.com) ### Why does that matter for AI systems, not just video players? March 2026 draft work on multimodal feedback shows the standards effort has already moved beyond plain streaming video. The working group page lists a “MoQ Multimodal Feedback” draft, suggesting contributors are designing for systems that need rapid back-and-forth exchange between media streams and downstream processing. (datatracker.ietf.org) In practice, that maps to AI pipelines where a camera feed, speech transcript, model ranking, alert and operator correction all need to move quickly. MOQ’s object-based publish-subscribe model makes it easier to deliver fresh updates and drop less useful old data when networks wobble, an inference supported by the draft’s emphasis on priorities and partial reliability. (datatracker.ietf.org) ### How could a newsroom actually use this? Remote reporting is one obvious case. A field camera or phone can send live contribution feeds over uneven links, while editors, transcription systems and vision models subscribe to the same stream or derived objects with different priorities. Because relays sit in the middle, the system does not have to open a separate fragile path for every downstream consumer. (datatracker.ietf.org) Deadline workflows are another case. A newsroom that wants near-live clipping, captioning, translation and model feedback during a press conference or breaking event needs low delay more than perfect delivery of every older packet. MOQ is being built around that tradeoff, according to the IETF draft and vendor implementations now being discussed publicly. (datatracker.ietf.org) ### What should teams watch next? May 12 is the latest dated milestone in the core standards track material, with draft-18 now the current transport draft. The MOQ working group page also lists newer companion drafts from March through May 2026, including multimodal feedback and additional streaming formats, which will show whether the protocol matures into production AI and newsroom pipelines. (datatracker.ietf.org 1) (datatracker.ietf.org 2)