Industries / Media, arts and recreation / Media and entertainment / Video games / Esports
Broadcasting
Broadcasting is showing matches to an audience: filming and directing the game, adding commentary, carrying the feed on television or streaming sites, selling or sharing the right to show it, and counting who watched.
7
Milestones achieved, from 2000 to 2024.
3
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2000
A cable channel for games
OnGameNet, a South Korean cable channel owned by the On-Media group, was founded in 2000 around a StarCraft league that had begun as a television programme. One of its league finals in Busan drew as many as 100,000 spectators. KED Global, the English service of the Korea Economic Daily, calls it "the world's first e-sports and gaming-dedicated broadcasting station".
LessonKED Global reported in 2021 that the channel lost viewers once game makers built their own broadcast studios, and that CJ had put it up for sale at a value under 10 billion won (about US$8.9 million). ).
Credit
On-Media; CJ ENM from 2009
2011
Twitch launches, Amazon buys it
Twitch began in June 2011 as a spin-off of Justin.tv, a general live-video site, to carry only live video of people playing games. By Amazon's figures, in July 2014 more than 55 million unique visitors watched more than 15 billion minutes made by more than 1 million broadcasters. Amazon named "stadium-filling esports organizations" among those broadcasters. On 25 August 2014 Amazon agreed to buy Twitch for about US$970 million in cash.
Credit
Twitch Interactive; Amazon
2019
Two counts of one event
Riot Games reported that its 2019 League of Legends World Championship (Worlds) had an average minute audience of 21.8 million, the average number watching in any given minute. It said the figures came from Nielsen. Inven Global counted 3.9 million peak concurrent viewers, the most watching at one moment, for the year's most-watched match. That count left out Chinese viewing, which it called "notoriously hard to properly source and confirm". The gap has lasted. Esports Charts counted 6,941,610 peak viewers outside China at the 2024 final and 3,981,986 at the 2026 final. Bilibili, a Chinese video site, said more than 20 million people watched the 2026 final on its own platform, by a method Strafe notes is not comparable.
Lessona viewer figure means little until you know who counted, which platforms were included, and whether it is a peak, an average or a total.
Credit
Riot Games and Nielsen; Inven Global; Esports Charts
2020
ESPN closes its esports desk
ESPN started a section on competitive gaming in 2016 and televised or streamed events such as the Overwatch League playoffs. In November 2020, during layoffs of about 300 staff, it said it would "cease operations for our dedicated daily esports editorial and content". Sources told Front Office Sports that the business "didn't scale enough to make a meaningful impact".
Lessona major sports broadcaster's backing did not by itself build an audience big enough to keep a dedicated unit.
Credit
ESPN
2021
China blocks a streaming merger
On 10 July 2021 China's State Administration for Market Regulation, the competition regulator, prohibited the merger of the streaming sites Huya and DouYu, which Tencent would have controlled. AVCJ reported that, by the regulator's figures, the two held 40% and 30% of China's game live streaming. Tencent already held 40% of the market for running online games. The companies ended the deal on 12 July 2021.
Lessona competition regulator stopped one company from owning both the games and the main places where people watch them.
Credit
State Administration for Market Regulation
2024
Twitch leaves South Korea
Twitch ended its Korean business on 27 February 2024. Its chief executive, Dan Clancy, wrote that its network fees there were "10 times more expensive than in most other countries", even after it tried peer-to-peer delivery and capped video at 720p. In November 2023 Twitch had held 52% of Korea's streaming market and AfreecaTV (now SOOP) 45%, by Mobile Index figures cited by Korea JoongAng Daily. StreamElements and Streams Charts reported that Naver's new site, Chzzk, more than doubled its monthly hours watched between January and March 2024.
Lessonone country's network fees pushed out a global platform, and local sites took over its streamers.
Credit
Twitch (Dan Clancy)
2024
Co-streamers carry the broadcast
In January 2020 Activision Blizzard gave YouTube exclusive streaming rights to its esports, including the Overwatch League and the Call of Duty League. Riot moved the other way, towards co-streaming, in which creators rebroadcast the official feed on their own channels with their own commentary. It said its 2022 Worlds final would be co-streamed "for the first time". In November 2024 it said English coverage of China's LPL league would come from co-streams only, with no "official broadcast in the traditional sense". By Esports Charts' count, 70.6% of LPL viewing outside China in 2024 came through co-streams, 18.4% on Marc "Caedrel" Lamont's channel. Stream Hatchet, a measurement firm, says co-streams made up 52.9% of the 2.7 billion esports hours watched in 2025.
Lessononce creators were allowed to show the match, much of the audience watched through them rather than through the official channel.
Credit
Riot Games; creators such as Caedrel
Watch for
AI subtitles on live casts
Level Infinite, Tencent's games publisher, says its IgniteX team built a system for live broadcasts. It separates the casters' voices from game sound, turns their speech into text and translates it by machine. It then puts subtitles on the stream "under 15 seconds" after the words are spoken, in "up to 12 languages at once". It says the system has run on Honor of Kings events, including the KPL league's grand finals, with linguists watching for errors, and gives no accuracy figure.
Achieved whenan outside party measures the subtitles' accuracy and delay on recorded broadcasts and publishes the results.
Claimed by
Level Infinite, June 2026
AI play-by-play commentary
MOBA-VL is a vision-language model, an AI model that watches video and writes text, with 9 billion parameters. It was built to commentate live on three MOBA games (multiplayer online battle arenas, the genre of League of Legends and Dota 2). It was trained on 860 professional matches, about 460 hours. On the authors' own benchmark it scores 63.25 against 55.12 for an earlier model, StreamingVLM. They also report that their training method raises event recall, the share of match events the commentary mentions, from 34.5 to 42.1 over ordinary fine-tuning.
Achieved whenother researchers repeat the scores, or a broadcaster uses such a model on air and publishes viewers' ratings.
Claimed by
Shengyun Zhong, Xinkang Zhao, Ziyuan Chu and Linchao Zhu (arXiv preprint), September 2026
A 750-million audience claim
The Esports World Cup Foundation said its 2025 event in Riyadh recorded 750 million viewers and 350 million hours watched. It said viewing peaked at 7.98 million during Gen.G's League of Legends win, across 35 languages and 97 broadcast partners, and named no measurement firm or method. Esports Charts counted a peak of 1.2 million for that final.
Achieved whenthe Foundation publishes how it counted, or an independent firm publishes figures on the same basis.
Claimed by
Esports World Cup Foundation, August 2025
Open
01
Score captions on esports casts
A public test of speech-to-text software on English esports commentary. People type out reference transcripts of about three hours of casts by hand. Each model is then scored on word error rate, the share of words it gets wrong, overall and separately for player handles, team names and game terms.
The briefCredit · unclaimed02
An open count of viewers
Open-source code that asks each platform's official API every minute how many people are watching an event's official channels and listed co-streams. It saves every reading and adds the channels together minute by minute. It then reports the event's peak beside Esports Charts' and the organiser's figures.
The briefCredit · unclaimed03
Grade AI-translated casts
A test of machine-translated subtitles on Chinese-language commentary. Two bilingual people translate an hour of casting into English as references. Machine subtitles are then scored for meaning, for errors on player and character names, and for delay.
The briefCredit · unclaimed
Open 01 · the brief
Score captions on esports casts
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of speech-to-text software on English esports commentary. People type out reference transcripts of about three hours of casts by hand. Each model is then scored on word error rate, the share of words it gets wrong, overall and separately for player handles, team names and game terms.
Why it is still open
Football and basketball commentary have measured tests: SoccerNet-Echoes scores Whisper on soccer commentary, and a 2026 study found Whisper got 21.7% of words wrong on 421 NBA segments. The captioning firm 3Play Media reports sports error rates "3x higher than the best performing industries", partly because of "player and coach names". The MOBA-VL team transcribed 460 hours of esports commentary with Whisper large-v3, found that "raw ASR commentary contains name errors", and reported no error rate against hand-checked text. This research found no published error rate for esports casts.
Built with what exists
OpenAI's open Whisper large-v3; ElevenLabs Scribe v3, which accepts up to 1,000 keyterms such as player handles (see the Note "Fewer word errors"); the jiwer Python library to count errors; matches from official league channels' public archives.
First step
Type out 10 minutes of one LEC cast by hand and run it through Whisper and Scribe v3.
Done looks like
Three hours of casts from at least four games, each minute checked by two people, with word error rates for each model with and without a list of names, and a separate error rate on names. References and scores are published with links and timestamps. Audio is shared only with the organiser's permission.
Open 02 · the brief
An open count of viewers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Open-source code that asks each platform's official API every minute how many people are watching an event's official channels and listed co-streams. It saves every reading and adds the channels together minute by minute. It then reports the event's peak beside Esports Charts' and the organiser's figures.
Why it is still open
Published counts come from commercial firms and organisers who release neither code nor minute-by-minute data. Stream Hatchet's report page gives no method and no list of platforms. Strafe notes that co-streamers' separate peaks were "not necessarily reached at the same time", so they cannot simply be added. YouTube warns that the counts its API returns can differ from its own cleaned "despammed" counts. This research found no open tracker that publishes its code and raw data for a whole event.
Built with what exists
Twitch's Helix API "Get Streams" endpoint, which returns a viewer count; the YouTube Data API's concurrentViewers field; Python on a small always-on server; each platform's API terms.
First step
Log the LEC's official Twitch and YouTube channels and ten co-streamers for one match day.
Done looks like
Minute-by-minute logs for every tracked channel at five events. Each event's peak is the largest total at a single minute, with the difference from Esports Charts' published peak reported and explained. Code and logs are public.
Open 03 · the brief
Grade AI-translated casts
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test of machine-translated subtitles on Chinese-language commentary. Two bilingual people translate an hour of casting into English as references. Machine subtitles are then scored for meaning, for errors on player and character names, and for delay.
Why it is still open
Level Infinite reports a delay of under 15 seconds for its live subtitles and says linguists watch for mistakes, but it publishes no accuracy figure. This research found no published test of machine translation on esports commentary.
Built with what exists
Whisper large-v3 or Scribe v3 for the Chinese transcript (see the Note "Fewer word errors"); a general AI model such as Claude for translation; Unbabel's open COMET translation metric; paid bilingual raters.
First step
Have one bilingual volunteer translate 10 minutes of a public KPL archive and compare it with one machine pipeline.
Done looks like
60 minutes of KPL or LPL commentary with two human reference translations. Each machine version is scored by COMET and by three raters on a five-point scale, with name errors and delay reported. Scores and timestamps are published.