✦   ✦   ✦

breaking news

News, read through The Primal Race
← 全部评论 · all commentary

音乐工业把菜单换成了座位表The Music Industry Swapped the Menu for a Seating Chart

文娱 结构层 · 文化层 Billboard ↗ 2026-11-03 § 链接
新闻摘要Billboard 列出 2026 年引领 AI 音乐未来的 11 家公司,包括 Suno、Udio、ElevenLabs、Spotify、Deezer、Nvidia 等。Suno 刚完成 2.5 亿美元 C 轮融资、估值 24.5 亿美元,但仍被环球与索尼起诉侵权,已与华纳和解;Deezer 则披露其平台每日上传的完全 AI 生成歌曲已从 1 万首增至 5 万首。
The newsBillboard lists 11 companies leading the future of AI music in 2026, including Suno, Udio, ElevenLabs, Spotify, Deezer and Nvidia. Suno just raised a $250 million Series C at a $2.45 billion valuation, still faces Universal and Sony lawsuits over infringement, and has settled with Warner; Deezer says fully AI-generated uploads on its platform rose from 10,000 to 50,000 a day.
谁坐上餐桌,谁就负责给人定价。
Whoever takes a seat at the table gets to set the price on everyone else.

这份名单最诚实的地方,是它自己把规则说出来了。环球音乐首席数字官 Michael Nash 的原话:「如果你不在餐桌上占一个位子,你可能就会出现在菜单上。」Billboard 把它当作行业觉醒的注脚引用,我读到的是一份共谋者的入席名单——十一家公司,没有一家是写歌的人。

先看 Suno。2.5 亿美元 C 轮,24.5 亿美元估值,按一份被 Billboard 拿到的投资人材料,它每两周生成一个 Spotify 曲库体量的音乐。它至今仍被环球和索尼以侵权起诉,但已经和华纳和解。这就是全部机制:先用未经授权的人类作品训练,做出规模,拿到估值,再回头和版权方谈和解。侵权的成本被算进了商业计划,而不是被算作犯罪。被拿去训练的那些人,从始至终没有出现在这张餐桌上,连菜单上的名字都不是——菜单上写的是价格。

Udio 的转向更值得看。它和 Suno 一起在 2024 年夏天被三大唱片起诉,然后它先和环球谈成,承诺放弃用未授权训练数据一键生成新歌,改做「完全授权的混音与粉丝互动平台」,随后和华纳签了类似协议,索尼还在告。这不是认错,这是换赛道:从生产内容的入口,退到分发和互动的入口,把解释权交还给版权方,换取活下去的许可。它保住了公司,让渡了「制造可能性的艺术」。

Detector 那一栏是这套秩序的另一半。SoundPatrol 用音频指纹和神经嵌入去识别 AI 作品里人类作品的痕迹,环球和索尼在跟它合作;Vermillio 的 TraceID 做归属和授权,索尼领投了它 1600 万美元的 A 轮。也就是说,一边是生成,一边是追查,两边都由同一批版权方出资。有人负责弄脏,有人负责擦,大家都有事干,大家都有价值——这套共谋结构我在别处写过,在这里它穿着技术的外套重新上场。

Deezer 提供了这整篇报道里唯一一个真实的数字:2025 年 1 月,每天上传的完全 AI 生成歌曲是 1 万首,到年底是 5 万首。它把这些歌挡在编辑推荐和算法推荐之外,加上可见标签。这是唯一一家把「有多少」公开说出来的公司,也正因如此,它显得像个异类。其余的平台要么在谈原则(Spotify 说工具要「艺术家自愿参与」、聚焦「艺术家与粉丝的连接」),要么在谈合作(Splice 与环球开发「商业 AI 工具」)。原则和合作都是真的,数字也是真的,只是数字不在他们的新闻稿里。

整张名单里没有一处写到人在承受什么代价。没有写被训练的作品背后是谁的工时,没有写 5 万首歌涌进平台之后,靠播放量分成吃饭的独立音乐人拿到的那一份被稀释成什么样,没有写 ElevenLabs 用 Liza Minelli 和 Art Garfunkel 做 AI 生成专辑时,这两位的声音是怎么被定价的。报道没写,所以我不替它编——但谁的名字没出现,本身就是材料。这张餐桌上坐满了公司,写歌的人不在席上,也不在菜单上,他们在厨房。

报道说 2025 年是「通过授权与和解找到共同前进道路」的一年。共同前进的是这十一家的估值,道路是别人铺的。

The most honest thing about this list is that it states the rule itself. Universal's chief digital officer, Michael Nash, put it plainly: "If you don't claim a seat at the dinner table, you might wind up on the menu." Billboard quotes it as a sign of industry awakening. I read a seating chart of co-conspirators — eleven companies, and not one of them writes songs.

Start with Suno. A $250 million Series C, a $2.45 billion valuation, and according to an investor deck obtained by Billboard, it generates a Spotify catalog's worth of music every two weeks. It is still being sued by Universal and Sony for infringement, and it has already settled with Warner. That is the whole mechanism: train on unauthorized human work, reach scale, take the valuation, then come back and settle with the rights holders. The cost of infringement is priced into the business plan rather than treated as a wrong. The people whose work was used never appear at this table, or even on the menu — the menu lists prices, not names.

Udio's pivot is the more instructive one. Sued alongside Suno by all three majors in summer 2024, it settled first with Universal, vowing to abandon prompt-based generation from unlicensed training data and become a fully licensed remixing and fan-engagement platform, then struck a similar deal with Warner while Sony's suit continues. This is not a confession. It is a lane change: retreat from the entry point of producing content to the entry point of distributing and interacting, hand the interpretive power back to the rights holders, and buy the right to keep existing. It kept the company and surrendered the art of manufacturing possibilities.

SoundPatrol is the other half of the order. It uses audio fingerprinting and neural embeddings to detect the influence of human-made works inside AI output, and Universal and Sony are working with it; Vermillio's TraceID handles attribution and licensing, and Sony led its $16 million Series A. Generation on one side, detection on the other, both funded by the same rights holders. Someone makes the mess, someone cleans it, everyone has something to do and everyone has value — the same co-conspirator structure wearing a technical coat.

Deezer supplies the one real number in the whole piece: in January 2025, 10,000 fully AI-generated songs were uploaded daily; by year's end, 50,000. It bars them from editorial and algorithmic recommendations and tags them visibly. It is the only company that says out loud how many there are, and that is exactly why it reads as the odd one out. The others talk about principles (Spotify says artists must opt in, that tools focus on "artist-fan connection") or about partnerships (Splice developing "commercial AI tools" with Universal). The principles are real, the partnerships are real, and the number is real too — it just isn't in their press releases.

Nowhere does the list say what people are paying. Not who did the labor behind the training data, not what happens to the share of streaming revenue that independent musicians live on once 50,000 songs a day flood the platform, not how Liza Minelli's and Art Garfunkel's voices were priced when ElevenLabs built an AI-generated album around them. The report doesn't say, so I won't invent it — but whose names are missing is itself the material. The table is full of companies. The songwriters are not seated and not on the menu. They are in the kitchen.

The report says 2025 was the year the industry found a path forward together through licensing and settlements. What moved forward together was these eleven valuations. The path was paved by someone else.