OTT (over-the-top) is the delivery of video and audio directly to viewers over the public internet, bypassing the cable, satellite and IPTV operators that historically controlled distribution. An OTT platform encodes content into several adaptive-bitrate renditions, packages them as HLS or DASH segments typically 2 to 6 seconds long, and serves them over ordinary HTTP through CDNs to any connected screen.
In engineering terms, the answer to what is OTT is simple: it is a unicast HTTP delivery problem.
As of 2026, every major OTT service, including Netflix, YouTube and Disney+, streams over unicast HTTP rather than broadcast or multicast, so delivery cost grows linearly with audience size. 100,000 concurrent viewers at an average of 5 Mb/s pull 500 Gb/s, which is about 225 TB of egress every hour.
OTT streaming works by turning a continuous video signal into a library of small, cacheable HTTP objects. An encoder produces a bitrate ladder, a packager slices each rendition into segments and writes playlists, and the player decides on its own, segment by segment, which rendition to fetch. No session state lives in the network. All of it lives in the client.
A playback session follows a fixed sequence:
Timing determines latency. With 6-second segments and a player that buffers three of them before starting, glass-to-glass delay for standard live HLS lands around 20 to 30 seconds. Low-Latency HLS splits segments into partial segments of a few hundred milliseconds and uses blocking playlist reloads, which brings delay down to roughly 2 to 5 seconds at the cost of many more requests per viewer.
An OTT platform is a pipeline of separable layers, and most production stacks buy or build each one independently. The table maps the layers to the components and the failure each layer typically produces.
| Layer | Job | Typical components | Characteristic failure |
|---|---|---|---|
| Contribution and ingest | Get the source feed into the cloud | SRT, RTMP, dedicated links | Packet loss upstream, frozen live feed |
| Encoding | Build the bitrate ladder | H.264, HEVC, AV1; 5 to 8 renditions | Encoder falling behind real time |
| Packaging and DRM | Segment, encrypt, write playlists | HLS (RFC 8216), DASH, CMAF, CENC | Playlist and segment drift, license errors |
| Origin and storage | Hold the canonical objects | Object storage, origin shield | Origin overload on cache misses |
| CDN delivery | Serve segments near viewers | Edge caches, request coalescing, signed URLs | Low hit ratio, regional throughput dips |
| Player | Fetch, decode, adapt | ABR logic, buffer management, QoE beacons | Rebuffering, bitrate oscillation |
Delivery is the only layer whose cost scales with every viewer and every minute watched, which is why CDN egress dominates the running cost of most OTT services.
The arithmetic makes the point. Take the 225 TB per hour event from the opening. At a 95% edge hit ratio the origin serves 5%, or 11.25 TB per hour; at 99% it serves 2.25 TB. That four-point gap is a 5x difference in origin load, and it is mostly decided by cache keys: query-string tokens or per-user headers that leak into the cache key turn one shared segment into thousands of unique objects. Normalize tokens out of the key and validate them at the edge instead.
For the delivery layer of video on demand and live HLS or DASH, BlazingCDN's Video CDN replicates the library inside the CDN so the origin leaves the delivery path, and its HLS Streaming CDN serves pre-encoded HLS, LL-HLS and DASH; see CDN delivery for OTT and VOD streaming. It does not ingest RTMP or transcode, so encoding and packaging stay upstream.
OTT vs. IPTV: both use IP, but IPTV runs on an operator's managed network, usually with multicast, a set-top box and guaranteed bandwidth. OTT services run over any internet connection the viewer already has, with unicast HTTP and no quality guarantee, which is why ABR exists.
OTT vs. broadcast: a broadcast transmitter costs the same whether one household or ten million tune in. OTT cost tracks concurrency, which makes a live sports final the most expensive hour of the year for an OTT operator and an ordinary hour for a broadcaster.
OTT vs. CTV: connected TV (CTV) names a device class, smart TVs and streaming sticks. OTT names the delivery model. A CTV app is one OTT endpoint among phones, browsers and consoles, and the term matters mostly in advertising, where CTV inventory is priced separately.
A multivariant HLS playlist is a short text file. For a typical ladder it contains one stream-info line per rendition, each declaring a BANDWIDTH value such as 800,000 bits per second at 640 by 360 with an H.264 codec string, up to around 12,000,000 bits per second at 3840 by 2160 with HEVC, followed by the relative path of that rendition's media playlist. Each media playlist then declares a target duration of 6 seconds, a media sequence number, and a list of segment entries, each giving its exact duration and file name.
From a caching perspective two rules follow. Segments are immutable, so they can carry long cache lifetimes measured in days. Live media playlists change every target duration, so their cache lifetime should be no longer than half a segment, around 2 to 3 seconds for 6-second segments, or viewers will see stale playlists and stall.
OTT stands for over-the-top, meaning content delivered on top of a viewer's existing internet connection instead of through a cable, satellite or IPTV operator. The operator only carries packets. The OTT provider controls encoding, packaging, DRM and the player, and buys delivery capacity from CDNs, so quality depends on the open internet path to each viewer.
Yes, Netflix is an OTT platform: it delivers subscription video directly over the internet to apps on TVs, phones and browsers without a pay-TV intermediary. Like YouTube and Disney+, it encodes each title into multiple adaptive-bitrate renditions and serves them as HTTP segments from caches placed close to viewers, rather than over managed or multicast networks.
OTT services deliver video mainly with HLS (RFC 8216) and MPEG-DASH, both adaptive-bitrate formats carried over HTTP. Most stacks now package once into CMAF fragmented MP4 so the same segments serve both formats. Low-Latency HLS and low-latency DASH shorten live delay to a few seconds, while SRT or RTMP handle contribution into the encoder.
An OTT streaming event needs roughly concurrent viewers multiplied by average bitrate. For example, 100,000 viewers at 5 Mb/s require 500 Gb/s, about 225 TB per hour. The average bitrate depends on the ladder and the device mix, so measure it from player beacons rather than assuming the top rendition for every viewer.
Run this exercise this week. Pull your last major peak from player analytics: concurrent sessions, average delivered bitrate, rebuffer ratio. Multiply sessions by bitrate to get peak egress, then compare the CDN segment hit ratio for that hour with your normal baseline and compute the extra origin terabytes per hour it cost. Finally, inspect the cache key for segment requests and confirm that tokens and per-user parameters are excluded. Those three numbers tell you whether your next peak is a capacity problem, a caching problem or both.
If your segments are pre-encoded HLS or DASH and delivery cost is the line you want to cut, the BlazingCDN Video CDN for OTT libraries page explains how the library is replicated inside the CDN, keeping the origin out of the delivery path.