Skip to main content
Published 2026-11-09Updated 2026-11-09CC BY 4.0

24/7 streaming in practice: 44,000 hours of pre-recorded live streams

What people actually run when they stream pre-recorded video around the clock: platforms, playlists, formats, bitrates and how long streams stay up. Aggregate data from our own service, September–October 2026.

We run a service that streams pre-recorded video to YouTube and other platforms 24/7. Public numbers about this kind of streaming are scarce, so we looked at our own: about 44,000 hours on air over two months, around 30 streams running at any moment on average. Everything below is aggregated; no figure describes a group of fewer than five accounts.

Key findings

  • 68.1% of all hours on air loop a single video. On paid plans it is 49.5%; on the free trial, 89.7%.
  • YouTube receives 89.7% of the streamed hours. Every other named platform together gets under 3%.
  • 1080p at up to 30 fps is the norm: 72.2% of hours are 1080p or higher, 77.1% run at up to 30 fps.
  • A quarter of the airtime is vertical video (24.7%).
  • 11.9% of hours come from playlists that mix frame sizes or frame rates, which some platforms handle badly.
  • Paid streams stay up longer: a median continuous session of 11.3 hours, against 2.2 hours on the free trial. The longest single session ran 44.8 days.
  • Measured uptime: 99.95% counting every gap against us, 99.98% counting only interruptions that were ours.

Download all figures (CSV)How to cite

Where the streams go

Shares of destination-hours: how long a signal was actually delivered to each platform. A stream sent to two platforms for an hour counts two destination-hours.

Share of destination-hours by platform2026-09-01 to 2026-10-31
YouTube89.7%
Custom RTMP server7.3%
Kick1.3%
Twitch1.2%
Rumble0.4%

"Custom RTMP server" is any address a user typed in themselves: other platforms, restreaming services, or their own servers. We do not inspect what is behind it.

Most streams loop one video

Two thirds of all streamed hours come from a single file played on repeat. Playlists matter more to paying users: half of their hours loop one video, half run a playlist.

Share of hours on air by playlist size7 Sep to 31 Oct 2026
1 video, looped68.1%
2–5 videos16.7%
6 or more videos15.2%

We wrote up what this means for anyone setting up a 24/7 stream in Most 24/7 streams loop a single video.

Formats: resolution, frame rate, orientation

Share of hours on air by resolutionResolution tier the stream was sent at
1080p or higher72.2%
720p27.8%
Share of hours on air by frame rate7 Sep to 31 Oct 2026
Up to 30 fps77.1%
60 fps22.9%

60 fps is more common on paid plans (31.3% of their hours) than on the free trial (13.2%).

Share of hours on air by orientationFrame shape of the videos in the stream
Landscape (horizontal)75.2%
Portrait (vertical)24.7%
Square0.1%

Bitrate of the files being looped

The peak bitrate of the video files being streamed, after our upload processing. The video is sent to the platform as it is stored, without re-encoding, so this is what the platform receives.

Share of hours on air by peak file bitrate7 Sep to 31 Oct 2026
Under 1.5 Mbps9.1%
1.5–4 Mbps18.9%
4–8 Mbps59.6%
8 Mbps or more12.4%

Mixed playlists

11.9% of hours on air came from playlists whose videos differ in frame size or frame rate. Some ingests cope with a change mid-stream; in our own tests Kick dropped a stream within seconds of a frame rate change. If a playlist will run for days, it pays to make every file the same resolution and frame rate.

How long streams stay up

A session is one stream running continuously; restarts with a gap under five minutes are joined into one session. The result barely moves with a 15, 30 or 60 minute threshold.

Paid plansFree trial
Median session11.3 h2.2 h
90th percentile24.7 h24 h (trial limit)
Sessions of a day or more11.6%n/a
Sessions of a week or more2%n/a

The longest single session in the period ran 44.8 days. Many paid sessions end after about half a day; our data does not tell us why, and we have not guessed.

Uptime

99.95% when every gap is counted against us: a failed run that restarted within an hour, any quick stop-and-start (including owners editing their own streams), and a flat allowance for each reconnect or server move. Counting only the interruptions that were plainly ours, 99.98%. Streams that a destination refused for good, for example after a stream key was reset, are not in either number.

Method and limits

  • Data. Production records of TheLoops, 2026-09-01 to 2026-10-31. Internal and test accounts are excluded. Figures are aggregated; no figure describes fewer than five accounts.
  • Hours on air were counted three independent ways, which agree within 2–5%: daily on-air totals 43,500 h; per-minute samples 44,664 h; run history 45,103 h.
  • Stream format figures (playlist size, resolution, frame rate, orientation, bitrate, mixed playlists) come from a log line written each time a stream is placed on a server. Those logs start on 7 September and cover 89% of hours on air. Each stream is weighted by its hours on air.
  • Paid vs free trial. "Paid plans" means the account had a paid plan or a funded balance. Comparisons where either side has fewer than five accounts are left out.
  • What we did not publish, and why. Content categories: most stream and file names are generic, so any split by topic would be guesswork. Geography: country data is incomplete. The share of streams sent to several platforms at once: until 9 November a bug in our service could drop a secondary destination for good, so the delivered numbers understate what users set up.

Using this data

The figures and the CSV are free to use under CC BY 4.0, with a link to this page.

TheLoops, "24/7 streaming in practice: September–October 2026", 2026-11-09, https://theloops.live/research/24-7-streaming-2026