How Meta Transcodes Video in the Cloud
When it comes to transcoding video at an operation like what Meta offers, given scale and the complex array of different targets and scenarios, there’s no simple answer to how Meta approaches cloud transcoding. But Meta Technical Program Manager Hassene Tmar guides viewers through the Meta video journey and decisions made around software, hardware, compression efficiency, codecs, and more in this conversation with Streaming Media 2025 Conference Chair Andy Beach at the latest Streaming Media Connect.
Meta’s Scalable Video Processor
Beach says that at Meta’s scale, there are multiple factors to consider for cloud deployments. He asks Tmar, “What are the different sort of levers? What are the different kinds of transcoding schemes you’re using, and then what are the different factors?” He mentions that latency and cost are factors, but there are “others that you have to think about for all the different types of scenarios you’re delivering for.”
Tmar decides to guide Beach through the “journey of a video at Meta”: “So let’s say I take my phone; I upload a video. Now, that video needs to have some kind of transcoding. It needs to be done as soon as possible. It needs to be available for my family, my friends. Even if I send it over Messenger, that’s basically a bit of a server transcoder in there. And so for that we use Meta’s scalable video processor, which is an ASIC video processor that was developed and deployed at Meta and within Meta. And for that we support both VP9 and AVC at this point,” although AVC is mostly for legacy support. This enables “the lowest possible latency to generate those encodes and good-enough quality with VP9.”
What Happens When a Video Goes Viral?
If a video goes viral, it’ll be sent to a lot of people, so Meta needs to reduce the bitrate further, Tmar notes. That necessitates what Meta calls premium encoding, using both hardware and software. For hardware, Meta applies a method called convex hull encoding. “We use hardware in the initial phase to identify the best possible points to encode,” Tmar explains. For software, it’s AV1 and VP9, Tmar says. “And we re-encode in software using the open source SVT-AV1 encoder. We re-encode those points for the best possible compression efficiency” and budget, he says.
Live, Video Calling, and End-to-End Sharing
For live video, Tmar shares that Meta mostly uses AVC “and still in software. For video calling on devices, we use a mix of software/hardware, depending on what’s available and what’s better, but that’s mostly on-device. And then for end-to-end sharing, again, it’s software/hardware based on what’s available and what’s the best compression efficiency.”
Beach ends this journey by saying, “I’d love to see a video of a few seconds in the life of a video at Meta as it walks through that process. That would be a fun roadmap to step through.”
Join conference chair Andy Beach and other streaming media experts in person Oct. 6–8 in Santa Monica, CA, for more thought leadership, actionable insights, and lively debate at Streaming Media 2025.
Related Articles
A number of factors impact codec deployment strategies and decision-making for live streaming, and particularly for a company like Norsk that has been involved with the Greening of Streaming organization from its inception, sustainability and power consumption are key considerations. Norsk's Steve Strong explains how to encode both on-prem and in the cloud with sustainability front of mind while balancing that with cost and other ever-present concerns in this discussion with Streaming Media 2025 conference chair Andy Beach in this clip from the latest Streaming Media Connect.
29 Jul 2025