Tencent Music lets AI songs onto QQ Music, but puts fully AI tracks on a different royalty model

Tencent Music Entertainment (TME), which runs QQ Music, Kugou, Kuwo and WeSing, is betting on AI music from both sides. It sells the tools that make AI songs, and it pays the royalties when people stream them. On its fourth-quarter call in March it said those two roles would be kept apart: songs made wholly by AI do not share royalties the way human recordings do. Its first-quarter results, out today, say AI-generated songs are now a growing share of daily new releases.

What did Tencent Music say?

On the 17 March call, chief executive Ross Liang said more than 10 million users and over 150,000 professional creators use TME's one-stop AI music production platform. Music Business Worldwide reported that the tools include track refinement and AI-generated vocal demos.

Getting a song from those tools onto the streaming service is easy. In April 2025 MBW reported that TME's AI Songwriter, part of its Tencent Music Venus platform, writes lyrics and vocals from a short prompt, and the result can be published to QQ Music in one click.

Asked on the call whether AI would eat into royalties, management drew a line. In the English interpretation of the call, as transcribed, TME said royalties and revenue sharing still come from professionally and officially produced content, and that it saw no material change there. A song made wholly by AI has "a different royalty sharing model". AI-assisted music that is still sung by a human voice stays on the traditional model. Liang also said several hits on music charts in the previous three months were made by AI, that most AI content is not original, and that some carries copyright risk.

TME did not say what the different model pays, how it decides that a track is wholly AI, or how many tracks fall into each group. The Q1 release doesn't say either.

What does the Q1 release add?

Not much on royalties, but the tone is clear. Executive chairman Cussion Pang says in the release that AI is widening who makes content but "does not replace human creativity", and that it makes premium IP more valuable. The company says it used AI to speed up production and to revive classic catalogue through "high-quality, authorized AI covers".

The business backdrop is softer. Total revenue was RMB7.90 billion (US$1.15 billion), up 7.3% year on year. Membership revenue rose 6.6% to RMB4.57 billion, and the release gives no paying-user or ARPPU figures. MBW noted in March that TME had stopped reporting key subscriber metrics every quarter, and that the shares fell nearly 25% on the day of the Q4 results.

Does Chinese regulation require this?

Not as far as we can find. We looked for a reported Chinese "incentive plus royalty" rule for AI tracks and found no official source for it, so we've left it out. The rule that does apply is about labels, not pay. China's Measures for Labelling AI-Generated and Synthetic Content, issued in March 2025 by the Cyberspace Administration and three other agencies, took effect on 1 September 2025 and cover audio. Generation services must embed labels in their output's metadata, and where explicit labels are required, audio gets spoken or rhythmic cues. Nobody may maliciously delete, alter or hide those labels. Platforms that distribute content must check files for those metadata labels and show a visible notice on labelled content. DLA Piper's summary gives the same scope. The royalty split is TME's own commercial choice.

How does the West compare?

Deezer is the most open Western service about AI. In January it said it was receiving about 60,000 AI-generated tracks a day, roughly 39% of daily deliveries, while fully AI music made up at most 3% of streams. Up to 85% of streams on those tracks were fraudulent in 2025. Deezer tags fully AI songs and keeps them out of algorithmic recommendations, but it says it demonetises streams it detects as fraud. Its announcement doesn't say that genuine streams of a fully AI song are paid any differently.

What we think

TME has put the line in an unexpected place. It isn't at "used AI" versus "didn't". It is at the voice: on TME's own description, a song with an AI-written melody that a human sings stays on the traditional model, and a song with a synthetic singer doesn't. For working musicians that is the more useful line. Most of the 150,000 professionals on TME's platform are presumably using AI somewhere in the chain. A rule that punished any AI use would hit them, while a rule built around the vocal mainly hits mass-produced, fully synthetic uploads. It also gives singers a commercial reason to exist inside AI-assisted production, which no Western licence has done yet.

There are two problems. The first is opacity. "A different royalty sharing model" could mean a smaller share, a fixed bonus, or nothing at all, and TME hasn't said which. Human creators can't judge whether their pool is protected, and AI users can't judge what they're signing up for. A company that reports AI as a growth story to investors should publish the rate.

The second is conflict of interest. TME both supplies the AI tools and decides which pool their output lands in. China's labelling rules help here, since TME's own generators have to tag what they make and QQ Music has to check for those tags. That makes the vocal test enforceable for anything made inside TME's ecosystem. It's much weaker for songs made elsewhere with the labels stripped, which the rules ban but which is exactly what a bad actor would try.

Deezer's numbers show why a separate pool matters more than it seems. If fully AI tracks are close to 40% of uploads but only a few per cent of streams, then pro-rata payouts barely notice them today. But per-stream rates are set by dividing one pot among everything, so the risk grows with every percentage point of listening that AI takes. Western services have so far tackled AI through fraud detection and recommendations. TME is the largest platform we've seen say in public that the royalty pool itself is the place to draw the line. Spotify, Apple and Amazon should have to explain why they haven't. The fair version would be this: label the track, keep fully synthetic vocals in their own pool, publish the rate, and let human performers opt in to the traditional pool by proving a human sang.

Sources

• Tencent Music Entertainment Group Announces First Quarter 2026 Unaudited Financial Results, TME, 12 May 2026
• Tencent Music Q4 2025 earnings call transcript, StockAnalysis, 17 Mar 2026
• Earnings call transcript: Tencent Music Q4 2025, Investing.com, 17 Mar 2026
• 3 things we learned from Tencent Music's Q4 earnings call, Music Business Worldwide, 18 Mar 2026
• Tencent Music lets musicians generate AI tracks and send them directly to QQ Music, Music Business Worldwide, 23 Apr 2025
• Measures for Labelling AI-Generated and Synthetic Content, Cyberspace Administration of China and others, March 2025
• China released new measures for labelling AI-generated and synthetic content, DLA Piper, 24 Mar 2025
• Deezer confirms demonetization of up to 85% of AI-music streams due to fraud, Deezer, 29 Jan 2026