リアルタイム通信へのAV1大規模導入：Metaの取り組みAdopting AV1 for Real-Time Communication (RTC) at Scale

Meta has detailed a multi-year program to bring the AV1 video codec into its real-time communication (RTC) stack, the technology that powers video calls across products such as Messenger, Instagram, and WhatsApp. The move matters because real-time video has stricter constraints than streaming on demand, and a more efficient codec can translate directly into better call quality on constrained networks, lower bandwidth costs, and improved reach in regions where connectivity is limited.

AV1 is an open, royalty-free codec developed by the Alliance for Open Media, an industry group whose members include Meta, Google, Amazon, Netflix, Microsoft, and others. It is positioned as a successor to VP9 and a competitor to HEVC, and it is widely cited as offering meaningful compression gains over older standards such as H.264 and VP8 at the same visual quality. Those gains are attractive for RTC, but they come at the cost of higher computational complexity, which is the central tension Meta describes navigating throughout the effort.

According to the company, the work began with codec selection, weighing AV1 against the codecs already deployed in its calling infrastructure. Streaming services have adopted AV1 primarily for video on demand, where encoding can be performed offline and decoders have time to buffer. Real-time communication is different: the encoder and decoder both run live, latency budgets are measured in milliseconds, and there is no opportunity to re-encode. This means the theoretical compression advantage of AV1 must be balanced against whether a device can sustain real-time encoding and decoding without overheating, draining battery, or dropping frames.

Device eligibility is therefore a recurring theme. Meta explains that not every phone or computer can run AV1 in real time, particularly on the encode path, which is more demanding than decode. The company appears to rely on a tiering or gating approach, enabling AV1 only where a device has sufficient hardware or software capability and falling back to other codecs otherwise. Hardware AV1 decode support has been expanding in newer mobile chipsets and GPUs, while software decoders such as dav1d are used to broaden compatibility where hardware support is absent. On the encoding side, real-time-oriented implementations are needed, since reference encoders like libaom were designed with quality and offline use in mind rather than the speed RTC requires.

Rate control is another pillar of the work. In real-time calls, the available bandwidth fluctuates constantly as users move between networks, and the encoder must adjust its bitrate moment to moment to avoid congestion and latency spikes. Meta notes that it had to adapt rate control specifically for AV1's characteristics so that the codec responds quickly to bandwidth estimates from the transport layer. This is closely tied to techniques common in modern RTC systems, including scalable video coding and simulcast, which let a sender produce multiple quality layers so that participants on different connections each receive an appropriate stream.

Error resilience rounds out the technical agenda. Real-time media is typically carried over unreliable transport, so packet loss is expected rather than exceptional. Unlike a streaming player that can pause and rebuffer, an RTC system must keep the picture moving through lost or late packets. Meta describes investing in resilience so that AV1 streams degrade gracefully under loss, using mechanisms to limit error propagation and recover quickly rather than freezing the video.

The broader context is an industry-wide shift toward AV1 in interactive media. The codec has been integrated into the WebRTC ecosystem, browsers have added support, and several large platforms have signaled interest in using it for conferencing as hardware acceleration becomes more common. Meta's account is notable mainly for the scale and the operational detail, covering not just compression results but the practical questions of how to deploy a demanding codec across a heterogeneous fleet of consumer devices.

Meta frames the project as ongoing rather than finished, and the published findings are likely intended both to document its engineering choices and to inform others working on similar systems. As more devices ship with native AV1 acceleration, the trade-offs the company outlines around eligibility and complexity should ease, potentially making efficient, royalty-free real-time video more widely available across the industry.