Live Encoding at Netflix Switches to VBR
On January 26, 2026, Netflix completed a significant infrastructure change for its live streaming platform: all Live events are now encoded with variable bitrate (VBR) instead of constant bitrate (CBR). The shift, which might appear to be a simple configuration tweak, required the engineering team to re-examine core assumptions about capacity planning, traffic steering, and even which bitrates to advertise for each rendition of a stream.
The change is rooted in how these two encoding modes treat video content. With CBR, every encoded stream is locked to a configured resolution and nominal bitrate—for instance, a 1080p stream targeting 5 Mbps—and the actual output stays near that target over time. The predictability simplifies operations: if a server can safely carry roughly 100 Gbps of live traffic, and each session averages close to its nominal rate, the system can admit about twenty thousand concurrent sessions per server with confidence. During an event, aggregate traffic shifts mainly when viewers join or leave; otherwise, the load profile remains smooth and easy to reason about.
The drawback of CBR is that not all content is equally difficult to compress. A studio interview or flat animation uses far fewer bits than a sequence of rapid camera pans across a packed, high-detail stadium crowd. Under CBR, both scenarios receive the same bitrate, wasting bits on easy scenes and potentially underspending on complex ones.
VBR inverts that objective. Instead of chasing a fixed bitrate, the encoder targets a consistent perceptual quality and allows the bitrate to fluctuate with scene complexity. Simple segments can drop well below the nominal rate, while complex shots can briefly consume more bits to prevent visible artifacts. At Netflix, this is implemented as "capped" VBR using the AWS Elemental MediaLive QVBR (Quality-Defined Variable Bitrate) setting.
Efficiency Gains Measured in Production
The real-world payoff shows up across several metrics simultaneously. Comparing a single episode of WWE RAW encoded at a nominal 8 Mbps with both CBR and VBR reveals the core behavioral difference: the CBR bitrate stays near target with small segment-to-segment wobble, while VBR drops sharply for simple scenes—like a waiting room at the start of the show—and spikes upward for complex sequences, such as a confetti-filled shot. Over a minute, the CBR line is practically flat, whereas the VBR trace swings considerably.
Those swings translate into measurable network benefits at scale. By introducing VBR, Netflix reduced the average number of bytes required to deliver a full event, which lowers the traffic needed to fill servers in Open Connect, its content delivery network, as well as the traffic needed to serve segments to members. Better encoding also improves the viewer experience: A/B tests across multiple Live events recorded about 5% fewer rebuffers per hour while transferring roughly 15% fewer bytes on average, with about a 10% reduction in traffic at the peak minute.
Capacity Planning Must Change
The trickiest issue with VBR is not that bitrate varies—CBR has some variance too—but the magnitude and duration of the dips. Under CBR, a 5 Mbps stream is effectively 5 Mbps on a per-minute basis. Current traffic is a solid proxy for a server's utilization, and steering logic can safely route new sessions to a server seeing low load.
VBR broke that assumption. During an easy, slow sequence, a 5 Mbps VBR stream might generate only 2 Mbps or less for an extended period. The server's aggregate traffic drops, making it look under-utilized, and steering systems begin admitting more sessions to fill the apparent headroom. When the content changes—a fight starts, confetti falls, or the camera cuts to a fast-moving, detailed shot—the encoder can jump to 6, 7, or 8 Mbps on subsequent segments. If the server had accepted many additional sessions during the low-bitrate window, the sudden aggregate spike can exceed what the NIC or network link can sustain, producing latency, packet loss, and viewer-facing stalls or quality downgrades.
Netflix's fix was to change the server-admission logic. Rather than judging available capacity from current traffic alone, the delivery system now reserves capacity based on each stream's nominal bitrate. Even if a VBR stream is currently in a low bitrate phase, the system treats it as able to return quickly to its nominal rate. This keeps traffic-steering behavior consistent between CBR and VBR and prevents a server from accepting too many sessions during a dip, only to be overwhelmed when the bitrate swings upward.
Rebalancing the Bitrate Ladder
Moving to VBR is not as simple as flipping an encoder switch while reusing the existing CBR "bitrate ladder"—the set of resolution/bitrate renditions offered to clients. The encode comparison on WWE RAW demonstrated why: both CBR and VBR streams at "8 Mbps" behave very differently, meaning that equal nominal bitrate no longer implies equal average bit delivery. Reusing CBR settings would leave VBR with fewer bits on average, risking lower quality.
Netflix first applied VBR on the existing ladder and used offline VMAF analysis—a perceptual video quality metric—to measure the gap. Early A/B tests confirmed the concern: overall VMAF came in about one point lower than CBR, with the shortfall concentrated at the bottom of the ladder. The engineering team then compared CBR and VBR encodes rung by rung. Wherever VBR quality came in more than one VMAF point below CBR, Netflix raised that rung's nominal bitrate just enough to close the difference. The high-bitrate streams, which were already close in quality, stayed largely unchanged—including the 8 Mbps rendition in the WWE analysis.
The final result is a ladder with slightly higher nominal bitrates on a few low-end rungs but less overall traffic, because VBR still reduces bitrate during simple scenes. That lets the encoding match the quality delivered by the CBR ladder while preserving the efficiency and stability benefits.
Next Steps for Live VBR
With VBR now behind all Netflix Live events, the team is exploring ways to exploit the new encoding model further. One experiment involves using the actual sizes of upcoming encoded segments in the adaptive bitrate algorithms on devices, rather than relying purely on advertised nominal bitrates. That should allow devices to choose streams matched to how VBR is likely to behave over the next few seconds, rather than how the stream is labeled.
A second area of work is relaxing the conservative capacity reservation. Netflix currently reserves based on nominal bitrates to keep servers safe; by applying a discount informed by observed VBR behavior, the team hopes to recover extra headroom without introducing stability risk.
