Press play on a video and it starts almost instantly — no waiting for a download, no progress bar crawling to 100%. Yet the video you are watching might be a two-hour film, several gigabytes in size. Streaming feels like magic, but it is really a clever chain of tricks: the video is chopped into tiny pieces, compressed ruthlessly, copied to servers near you, and reassembled on your device just ahead of the playhead. Here is each link in that chain.
The core idea: small pieces, just in time
A streaming service never sends you a whole video. Instead, each video is divided into chunks (also called segments), typically 2 to 10 seconds long. Your player downloads a few chunks ahead of where you are watching — a cushion called the buffer — and plays them in order, fetching the next chunk while the current one plays.
This solves three problems at once. You can start watching after downloading just the first few seconds, not the whole file. If your connection drops, you lose only the current chunk, not the entire video. And crucially, the player can change quality mid-stream — because each chunk exists in several quality versions, the player simply starts requesting a different version when conditions change.
Before you can watch anything, the player downloads a small text file called a manifest — an .m3u8 playlist for Apple's HLS protocol, or an .mpd file for MPEG-DASH. The manifest is a map: it lists every available quality level and where to find each chunk. Think of it as the menu and the seating plan in one document.
Compression: why a film fits through your connection at all
Raw video is enormous. Uncompressed high-definition video would need hundreds of megabits per second — far more than most home connections deliver. So before a video is ever streamed, it is compressed by a codec, a mathematical recipe for throwing away information your eyes will not miss.
Modern codecs include H.264 (AVC), the long-standing universal standard; H.265 (HEVC), roughly twice as efficient, which is why it powers most 4K streaming; VP9, Google's open alternative; and AV1, the newest open codec, which squeezes video smaller still at the cost of heavier processing. Codecs exploit the fact that most of a video frame is identical to the previous one: instead of storing every pixel of every frame, they store one full frame and then only the changes between frames.
The same video is then transcoded — encoded again and again at different quality levels, from a small 240p or 480p version for weak connections up to 1080p or 4K for fast ones. This ladder of versions is what makes the next trick possible.
Adaptive bitrate: the player that watches your connection
The smartest part of streaming is adaptive bitrate streaming (ABR). While you watch, your player constantly measures how fast chunks are arriving and how full its buffer is. If chunks arrive quickly and the buffer is healthy, it steps up to higher-quality chunks. If downloads slow down, it quietly switches to lower-quality chunks before the buffer runs dry — so instead of a spinning wheel, you get a slightly softer picture for a few seconds, then a return to sharpness when the connection recovers.
This is why video quality sometimes dips for a moment when someone else in the house starts a download, and why a video that starts blurry often sharpens after a few seconds: the player begins cautiously at a lower quality to guarantee a fast start, then climbs the ladder as it gains confidence in your connection.
The two dominant ABR protocols are HLS (HTTP Live Streaming, developed by Apple and now used almost everywhere) and MPEG-DASH (an open standard favoured by YouTube and Netflix). Both work the same way in principle: chunked video, a manifest, and a player that picks the best chunk for current conditions.
CDNs: the servers next door
Fetching every chunk from one central server would be slow and fragile — imagine millions of viewers all pulling from a single building. Instead, streaming services use a content delivery network (CDN): a global web of edge servers placed in cities close to viewers.
When you press play, your chunks usually come from a nearby edge server, not the origin server where the video lives. If a show is popular in your city, the CDN keeps its chunks cached locally, so the next thousand viewers in the same area get them almost instantly. This cuts delay, spreads the load, and is the main reason a global launch can serve millions of simultaneous viewers without collapsing.
Live streaming works the same way with one extra difficulty: chunks have to be created, encoded and distributed while the event is happening, so live streams typically run seconds behind reality. That delay — the reason your neighbour cheers a goal before you see it — is the price of chunking a live event.
For more on this, see our explainer on undersea cables.
Why videos still buffer
Buffering happens when the buffer empties: chunks arrive slower than the video plays. The usual culprits are:
- A weak or congested connection — the most common cause, and exactly what ABR tries to hide by dropping quality first.
- Wi-Fi trouble — distance from the router, interference from neighbours, or too many devices sharing the airwaves. Our guides to how Wi-Fi actually works and how 5G changes connectivity explain the underlying limits.
- Server or CDN strain — rare for big services, but a huge live event can still overwhelm local capacity.
- An underpowered device — decoding 4K AV1 video takes real processing power; an old TV stick may stutter on a codec it can barely handle.
Practical fixes follow the same logic: move closer to the router or plug in an ethernet cable, lower the quality setting manually, close other downloads, or restart the app so it re-measures your connection from scratch.
Streaming vs downloading: a quick comparison
| Streaming | Downloading | |
|---|---|---|
| Start watching | After a few seconds | Only after the full file arrives |
| Storage needed | Almost none (a small buffer) | The whole file |
| Quality | Adjusts automatically to your connection | Fixed at the downloaded quality |
| Works offline | No (unless pre-downloaded) | Yes |
| Best for | Films, shows, live events | Planes, commutes, patchy connections |
That last row is why most services now offer both: stream at home, download before a flight.
The hidden costs of instant video
Streaming's convenience has a footprint. Every chunk travels through networks and data centres that consume electricity, and video is now the dominant share of internet traffic. Services fight this with better codecs — AV1 can deliver the same quality at noticeably lower bitrates than H.264 — and with smarter CDNs that serve popular chunks from nearby caches instead of hauling them across oceans.
For viewers, the takeaway is reassuringly simple: the spinning wheel is not a mystery and rarely your device's fault. A video reaches you as hundreds of small, compressed, cleverly routed pieces, with a player constantly re-deciding which piece to fetch next. When it works — which is most of the time — you never notice any of it. That invisibility is the achievement.
For the bigger picture, see how the internet actually works.
Takeaway checklist
- Streaming never sends a whole video: it is chopped into 2–10 second chunks and played just ahead of the playhead.
- A manifest file (.m3u8 for HLS, .mpd for DASH) tells your player where every chunk and quality level lives.
- Codecs (H.264, H.265, VP9, AV1) compress video by storing only the changes between frames.
- Adaptive bitrate streaming switches chunk quality up or down as your connection changes, so you get a softer picture instead of a spinning wheel.
- CDNs serve chunks from edge servers near you, which is how millions can watch the same launch at once.
- Buffering means the buffer ran dry — usually a weak or congested connection, not a broken device.
- Live streams run seconds behind reality because chunks must be encoded and distributed as the event happens.
