Bitrate is the number of bits used to represent one second of encoded media, expressed in kilobits or megabits per second (kbps, Mbps). In video streaming, bitrate is the single strongest lever on both picture quality and bandwidth consumption: a 1080p30 H.264 stream at 5 Mbps moves about 2.25 GB per viewer-hour, while the same resolution at 2 Mbps moves 0.9 GB and shows visible blocking in high-motion scenes.
Bitrate is a rate, not a size. An encoder is handed a raw frame sequence and a bit budget per second; it spends that budget on residuals, motion vectors, and headers, then discards whatever detail does not fit. When the budget is too small for the scene's complexity, the rate-distortion loop raises the quantization parameter, and you see the result as blocking, banding, and smeared grain.
Three control modes matter in practice. Constant bitrate (CBR) holds output nearly flat, which suits fixed-capacity contribution links and live encoders. Capped variable bitrate (capped VBR) targets an average while allowing peaks up to a ceiling, which is what most video-on-demand ladders use. Constant-quality modes (CRF, or constant rate factor) let bitrate float to hold perceptual quality steady, so a talking-head scene costs a fraction of a concert crowd shot.
| Rung | H.264 bitrate | HEVC / AV1 equivalent | GB per viewer-hour (H.264) |
|---|---|---|---|
| 240p30 | 0.3–0.5 Mbps | 0.2–0.3 Mbps | 0.14–0.22 |
| 480p30 | 1.0–1.5 Mbps | 0.6–0.9 Mbps | 0.45–0.68 |
| 720p30 | 2.5–3.5 Mbps | 1.5–2.2 Mbps | 1.1–1.6 |
| 1080p30 | 4.5–6 Mbps | 2.7–4 Mbps | 2.0–2.7 |
| 1080p60 | 6–8 Mbps | 3.5–5 Mbps | 2.7–3.6 |
| 2160p60 (4K) | 15–25 Mbps | 8–14 Mbps | 6.8–11 |
The one number to carry away: every 1 Mbps of sustained streaming bitrate costs roughly 0.45 GB of egress per viewer-hour, so ladder decisions are billing decisions.
A 1080p60 stream encoded with H.264 at 6 Mbps consumes about 2.7 GB per viewer-hour, while 2026-generation AV1 encoders reach comparable perceptual quality near 3.5–4 Mbps, or roughly 1.6–1.8 GB per viewer-hour. At 10 million viewer-hours per month, that codec change alone removes on the order of 9–11 PB of egress.
Bitrate is decided at the encoder, declared in the manifest, and negotiated at the player. The encoder produces a ladder of renditions; the packager segments each rendition into 2–6 second chunks and writes their advertised bitrates into an HLS master playlist or a DASH MPD; the player's adaptive bitrate logic picks a rung per segment based on measured throughput and buffer level.
An HLS master playlist declares each rung with a bandwidth attribute in bits per second:
#EXTM3U
#EXT-X-STREAM-INF:BANDWIDTH=6200000,AVERAGE-BANDWIDTH=5800000,RESOLUTION=1920x1080,FRAME-RATE=60.000,CODECS="avc1.640028,mp4a.40.2"
1080p60.m3u8
#EXT-X-STREAM-INF:BANDWIDTH=3300000,AVERAGE-BANDWIDTH=3000000,RESOLUTION=1280x720,CODECS="avc1.64001f,mp4a.40.2"
720p30.m3u8BANDWIDTH is the peak segment bitrate including container overhead, not the encoder target. Players use it as the switching threshold, so understating it causes rebuffering and overstating it keeps viewers pinned to a lower rung.
Bitrate vs. bandwidth: bitrate is what the stream demands; bandwidth is what the path can supply. A 6 Mbps rendition on a 6 Mbps link will stall, because segment delivery needs headroom to refill the buffer. Plan for delivered throughput of at least 1.3–1.5 times the rendition bitrate.
Bitrate vs. resolution: resolution is pixel count per frame, bitrate is bits per second. Encoding 4K at 4 Mbps produces a softer, more artifact-prone image than 1080p at the same 4 Mbps, because the bit budget is spread across four times as many pixels.
Bitrate vs. perceptual quality: quality is measured with metrics such as VMAF or SSIM, and the relationship to bitrate saturates. Past roughly VMAF 93–95, added bitrate buys almost nothing a viewer can see, which is why per-title and per-scene encoding beats a fixed ladder.
Adaptive bitrate (ABR) selection runs on three inputs: recent segment download throughput, current buffer occupancy, and the ladder declared in the manifest. Throughput-based heuristics average the last few segments and step up only when the estimate clears the next rung with margin. Buffer-based algorithms such as BOLA ignore short throughput spikes and switch on buffer level instead. Most 2026 production players run a hybrid: buffer-driven in steady state, throughput-driven at startup.
Startup is where ladders are won or lost. Players typically begin on a low or mid rung to reach first frame quickly, then climb, so a ladder whose lowest rung is 1.5 Mbps punishes viewers on constrained mobile paths with a longer time to first frame.
Higher bitrate always means better quality. False beyond the saturation point of the codec and content: a static slide deck at 8 Mbps and at 2 Mbps are visually identical, and both waste bits.
The ladder should be evenly spaced. Rungs spaced closer than roughly 1.5 times apart cause frequent, visible switching without meaningful quality gain. Ratios of 1.5–2 times between adjacent rungs are the practical range.
Audio is negligible. AAC stereo at 128 kbps adds about 0.06 GB per viewer-hour, which matters at the bottom of the ladder where it can be 20–30% of total bitrate.
For deeper reference material on delivery-side behavior and cost per terabyte, the BlazingCDN streaming engineering blog covers segment caching and egress math in more detail.
For 1080p30 with H.264, 4.5–6 Mbps is the practical range; 1080p60 needs 6–8 Mbps. With HEVC or AV1, subtract roughly 35–45% for comparable perceptual quality. Content complexity shifts these numbers: animation and screen capture compress far better than sports, grain, or confetti.
Multiply bitrate in Mbps by 0.45 to get gigabytes per viewer-hour. A 5 Mbps stream uses about 2.25 GB per hour, 8 Mbps uses 3.6 GB, and 20 Mbps 4K uses about 9 GB. Add roughly 2–5% for container, manifest, and HTTP overhead when forecasting egress.
Bitrate is a single encoding rate; adaptive bitrate is the delivery technique of publishing several bitrates and letting the player switch between them per segment. Adaptive bitrate streaming requires aligned segment boundaries and closed GOPs across renditions so switches happen without a visible glitch.
Capped VBR gives better quality per bit for video-on-demand because it reallocates bits toward complex scenes. CBR is preferable for live contribution and for fixed-capacity paths where predictable buffer behavior outweighs efficiency. Live distribution commonly uses capped VBR with a peak ceiling near 1.2–1.5 times the average.
Take one 60-second clip from your hardest content class, encode it at five bitrates spanning half to double your current 1080p rung, and score each with VMAF. Plot the curve and find where it flattens; that inflection point, not a vendor chart, is your correct ladder target. Then multiply the delta in Mbps by 0.45 and by your monthly viewer-hours to get the egress you would save or spend. Most teams find one rung they can drop and one they should add at the bottom.