The Universal – has music’s AI future finally been sold? UMG CEO sets out his AI vision

Since the birth of popular music, artists have inspired a sliding scale of fandom: casual listeners, people who love a particular song but can take or leave the rest, plus the hard core of fans and 'superfans', those devotees of an act or genre, who will see the shows, grab albums in every format, and buy the t-shirt from the global merch stand.

Since the 1990s, the Web has certainly put bands in touch with their fans in a way that stadium shows can’t rival, but local gigs can. Meanwhile, the original Napster arguably proved the business model of digital downloads, despite labels’ (and musicians’) opposition at the time.

But that was thirty years ago. Today, 'superfan' opportunities are rapidly evolving and are “super-charged by AI”. That’s the bold claim of Sir Lucian Grainge, Chairman and CEO of one of music’s remaining majors: the Universal Music Group (UMG); the first music CEO invited to speak at NVIDIA’s GTC conference. 

Note that word “responsible”. It’s the critical element, because 2026 finds much of the industry in defence mode – and that’s before you factor in AI. With paid-for downloads ebbing away, and vinyl the preserve of a wealthy, patient few (though on-demand vinyl is a promising new model), artists only receive micro-payments from Spotify, Apple, Amazon, Deezer, and other streams.

With 90% of songs receiving fewer than a thousand listens, most artists receive nothing at all for their work. And that’s because streaming’s big beast Spotify changed its rules a few months back: it now only pays royalties on tracks that have amassed over a thousand streams. But even a thousand streams create revenue of a dollar at most – three months after those streams took place.

So, you can do the math yourself: 10,000 streams mean ten dollars, perhaps, and 100,000 $100 – divide by four if you are a four-piece band, and then defer that income by a quarter. Such figures fail to cover even a fraction of most acts’ costs, let alone provide a sustainable living. And that assumes the artist owns the rights to their music.

Thus, Spotify maximises its income, while sitting on fees for a quarter. Its Executive Chair and former CEO Daniel Ek has grown rich – with an estimated net worth of $7.8 billion – on the work of hundreds of thousands of unpaid artists. 

And into an environment that, many artists would say, is toxic to original music makers comes AI, the very technology that Grainge is now celebrating. In a strategy briefing recently, he confirmed: 

Our embrace of responsible AI technologies continues to be very aggressive. We're forging partnerships across a spectrum of artist creation and fan engagement initiatives.

Friend or foe?

Since the launch of ChatGPT, most artists – struggling to make a dollar – see AI as the enemy of the creative sectors, not as an ally. Music platforms such as Suno have been trained, by CEO Mikey Schulman’s own admission, on high-res audio scraped from the Web. Thus, subscribers who can’t write, sing, or play an instrument can produce automated, professional-sounding compositions – based on scraped musical performances – with no more effort than a prompt.

So, it is not for nothing that over a thousand artists, such as Sir Paul McCartney, Hans Zimmer, Kate Bush, and Damon Albarn, supported a protest album last year called ‘Is This What We Want?’, featuring just the ambient sounds of empty recording studios.

But UMG’s Grainge only sees upside in AI:

I'm very aware that a large swathe of the investment community looks at the intersection of AI and media and sees only some of the risks. I want to be very clear: we fundamentally disagree with that view. We believe AI represents unprecedented commercial opportunities for UMG and our artists, in both the near and the long term. We're working tirelessly to shape the business models and the legal and legislative frameworks that will form the foundation of a responsible AI ecosystem.

There’s that word again: responsible, with the bonus of “unprecedented commercial opportunities”, no less. Even so, some might find Grainge’s comments may seem surprising to some, given that UMG scored a significant victory against the AI sector last summer.

In June 2024, UMG sued generative AI platform Udio for “mass infringement of copyrighted sound recordings, copied and exploited without permission”. Ten months later in August 2025, the US District Court for the Southern District of New York found for the plaintiffs. As a result, Udio, in its then form, was replaced by a new service that just allows the use of content licensed from UMG.

Ironically, Udio’s users complained about “their” work being infringed, with all ownership rights in the generated works reverting to the platform. Similar actions against Suno and its ilk continue – one reason for AI companies lobbying the US government to overturn copyright actions against them.

AI companies have variously claimed that anything on the Web is free and fair game for AI training – regardless of how it got there; that the use of copyrighted content is essential to see off the threat from China; and that generative AI is ‘transformative’ and does not retain copies or produce derivative works.

However, the latter argument – which persuaded some courts to find against the plaintiffs in AI copyright cases – were put in doubt by research published this month by Xinyue Liu, et al.

A research paper, Alignment Whack-a-Mole: Finetuning Activates Verbatim Recall of Copyrighted Books in Large Language Models [https://arxiv.org/abs/2603.20957v2] found – as the title suggests – that fine-tuning a prompt has the effect of uncovering verbatim source data, proving that, in whatever form it exists, the AI has effectively retained a copy.

Obviously, that research referred to books, and was the second paper this year to reach that conclusion. However, the findings, which applied to all the LLMs tested – including GPT-4o, Gemini 2.5 Pro, and DeepSeek v3.1 – have legal implications for every AI copyright case.

The authors note: 

Our findings offer compelling evidence that model weights store copies of copyrighted works and that the security failures that manifest after finetuning on individual authors' works undermine a key premise of recent fair-use rulings, where courts have conditioned favourable outcomes on the adequacy of measures preventing reproduction of protected expression.

Universal appeal 

But back to Grainge. Just why is he so enthusiastic about AI, given its recent history of threatening his industry’s existence, including the value of both original work and copyrighted content? His rationale:

“I encourage people to spend time to really understand the work that's being done and the opportunities that lie ahead. Personally, I've never been more energized about the possibilities we are pursuing. And once again, we face another exciting transformation. 

Listingt some things that excite him, agreements with Udio and Stability AI [with whom UMG has signed a strategic development agreement] are cited, as well as the signing of a licensing agreement with Klay Vision [to create an AI music foundation model]:

Klay's Large Music Model is trained entirely on licensed music. It will evolve AI experiences for 'superfans', while respecting the rights of artists and songwriters. We're excited about this company's vision and applaud their commitment to ethically – to ethicality – in generative AI music.

Implicitly, therefore, 'superfan' opportunities will lie not in meet-and-greets, privileged seating, or limited-edition merch, but in content generated from licensed originals. In this way, UMG’s artists will receive an income from works that, in some way, further their original compositions, perhaps, or produce new ones using their distinctive voice, musicality, or style.

All of which begs the question of who would own the rights in such an AI-generated composition? Would it be Universal, or the artist on whose portfolio it was based? An AI-only work cannot be copyrighted, but one originated by an artist can be, if the AI played a guided, secondary role. But would the artist own intellectual rights in it? That remains to be seen, but in either case it becomes a sellable good.

That aside, this may be good news for the few artists who are on a major label, perhaps – assuming they support the generation of works they did not actively create themselves. Doubtless some may see real creative potential for discovery, play, and experimentation – just as many artists have encouraged listeners to remix tracks from stem files in the past. But others who maintain strict control over their catalogue, or scarcity in their outputs, may not appreciate AI being thrust upon them in this way.

But there is more to come, said Grainge – hence his platform at NVIDIA GTC last week:

In December, we revealed we're also collaborating with Splice, the world's most popular music creation platform. Together, we are building a roadmap for the development of commercial AI tools rooted in creative control and sonic excellence.

And last month, we unveiled the first of its kind alliance with NVIDIA, the world leader in AI computing. Our shared ambition is to transform and enrich the music experience for billions of music fans around the world. This collaboration will cover everything from artist tools to music discovery to fan engagement. NVIDIA articulated the relationship perfectly when they said, ‘We're entering into an era where a music catalogue can be exploited like an intelligent universe, conversational, contextual and genuinely interactive. And we'll do it the right way, responsibly with safeguards that protect artists’ work, ensure attribution and respect copyright.

Great news, in itself, though NVIDIA appears to be playing both sides, as a partner of several AI companies that scraped the Web for training data. And there's clearly more to come where that came from: 

How phenomenal! Our work with NVIDIA will be a multiyear partnership and, like our other AI initiatives, create significant win-win potential in market-led solutions. 

And all this is giving the customers what they want, is a bottom line conclusion from the CEO:

Our strategy for these AI deals is informed by a significant amount of consumer research, both our own and third party – we're just not sticking our finger in the wind. Our insights team recently conducted a global study on consumer attitudes towards AI and music. The key takeaway is that consumers want AI driven by human intent or AI as an enhancement of, and not as a replacement for, human creativity. Plus, consumers are asking for transparency with respect to how AI is used in the creation of music. This research underscores our belief that AI isn't just an incremental revenue opportunity. It's going to introduce entirely new formats.

My take

Doubtless many artists will take the win from licensed platforms that respect their original work, voice, musicianship, artistry, and catalogues. But it’s notable that Grainge just talked about listeners’ wishes and beliefs, and not those of artists themselves.

But what about the thousands of acts who aren’t on a major, but instead release through independent labels that maintain closer and more nurturing relationships with artists – many of which, like Bella Union, were founded and are run by musicians? Can they afford to take on the might of the highly capitalized US software sector?

And what of the millions of acts who believed in the promise of the Web, but now find themselves denied any income at all from their work, while facing the threat of automated competitors and fake AI acts?

Yet again, music may end up being the kind of two-tier market that has dogged the industry since day one: majors who eat everything that moves, and a grassroots sector that struggles even to survive – let alone to be heard.

(How many of the 140,000 tracks uploaded to Spotify today did you listen to? Most will only get heard if the artist can persuade an influencer or a playlist to feature them – which means more money changing hands than the act will ever see in return.)

Remember, the COVID pandemic saw the mass closure of local music venues, many of which have never reopened. Many musical instrument shops have since shut too, because musicians have less and less disposable income. The music tech sector could have stepped in and supported them, but arguably its most prominent leader, Ek, chose autonomous warfare instead as his preferred investment vehicle.

The good news is that one major label, at least, is investing time, energy, and money in an AI strategy that appears ethical in intent – and Universal won a notable victory against an AI start-up that scraped the Web rather than licensed its training data. All of that is a win for creatives, in general terms.

But will UMG’s artists really be center stage of these deals, or will UMG cream off most of the profits? The main point of Grainge's briefing was a pitch to investors and Wall Street analysts, remember, so I guess we’ll see.

Meanwhile, let’s remind ourselves of a classic tune, courtesy of Damon Albarn, who wrote these memorable words, and Blur. Altogether now:

This is the next century
Where the universal's free
You can find it anywhere
Yes, the future has been sold.

Every night we're gone
And to karaoke songs
How we like to sing along
Though the words are wrong.

It really, really, really could happen.