Music Generator Suno Raises Capital at $500 Million Valuation; Altman Went Behind ‘Her’ Back
New York isn’t known for its roomy apartments, which is why my electric piano is currently gathering dust on top of my kitchen cabinets. So I was especially excited to use Suno, an artificial intelligence music generator. It turns simple text descriptions like “a dark disco song about when you're not around” and “a romantic garage song about being trapped in an AI music factory” into songs that often sound professionally made.
The app’s usage, if not revenue, has attracted investors. Lightspeed Venture Partners, known for backing generative AI firms including Stability AI, Mistral AI, Glean and Pika, is leading a Series B deal valuing the two-year-old Suno at $500 million, including the new capital, according to a person involved in the round. The precise size of the round couldn’t be learned, but the person said the company has raised $125 million in total, including from Nat Friedman, Daniel Gross and Founder Collective.
Suno, which comes from the Hindi word for “listen,” uses its own audio model to generate the music and OpenAI’s ChatGPT to generate lyrics and titles. In the eight months since Suno launched its first product, its founders say 10 million people have made songs on its site, and some of those songs have gone semi-viral on X. Stone was the first song to cross 1 million listens, and here are some catchy electronic ditties. If AI music is this credible after a year or two of development, just imagine how good it could be five years from now.
But the Cambridge, Mass.,-based startup and other rivals like Udio face potential threats. Some musicians and producers are concerned that AI music generators are using their work without permission to train AI models. CEO Mikey Shulman wouldn’t disclose exactly what data Suno’s audio model is trained on, but said that it’s a combination of both music and speech and that the company “works very closely with lawyers to make sure what we’re doing is legal.” Notably, Suno says it’s also working closely with musicians such as DJs Flosstradamus and 3LAU.
In an era where AI researchers and developers are increasingly facing scrutiny and lawsuits around their training data, Suno’s lawyers will need to work longer hours. Just last week, Sony Music, whose artists include Lil Nas X and Celine Dion, sent letters to more than 700 companies warning them against using their music without explicit permission.
Suno’s founders didn’t comment on whether they’d received one of these letters, but they said they believe their technology could help artists in the long run. For instance, someday Suno could generate music based on users humming melodies or tapping out beats, similar to how artists create music today. That could speed up artists’ production process. And artists might benefit from listeners who want to quickly remix their songs using Suno, including by changing the songs’ genre. Voice artists also could opt to receive a commission every time a Suno user creates a song with their voice, Shulman said.
Suno also is exploring how to compensate artists in exchange for using their data, he said. Already, large language model developers such as OpenAI have struck licensing agreements with content producers such as the Financial Times and the Associated Press. And some image repositories like Shutterstock have reportedly licensed their data to AI developers, too.
Much of the company’s new funding will go toward hiring employees (it has less than 20) and renting servers to run its models. Audio-generating models tend to be just as compute-intensive as text models, Shulman said. Suno users can generate four songs for free, pay $9.90 for 600 songs and $29.90 for 3,000 songs. But the company isn’t wedded to that model, he said.
Here’s what else is going on…
‘Her’ Puts Egg on ‘His’ Face
On Monday, Scarlett Johansson showed OpenAI CEO Sam Altman is still having trouble being “consistently candid.”
Sunday night, OpenAI posted a blog post explaining how it chose various voices for its impressive new voice AI assistant, which is coming soon to ChatGPT and we think is pretty impressive. One of the voices it demonstrated earlier this month, which it calls Sky, bore a resemblance to Johansson’s, as you can see in this demonstration video. But in its post, OpenAI said it paid voice actors top dollar to clone their voice and added that “AI voices should not deliberately mimic a celebrity's distinctive voice.” It then added: “Sky’s voice is not an imitation of Scarlett Johansson but belongs to a different professional actress using her own natural speaking voice.”
Uh, OK. Not only was the post “oddly defensive” about the ScarJo connection, as my colleague Aaron Holmes put it, but there’s no way that a person with ears could believe the connection was a coincidence. Johanssen portrays a voice-based AI in the (excellent) 2013 film “Her,” and Altman has talked a lot about his appreciation of the film. He even tweeted “her” after OpenAI’s demo of the voice assistant.
At the same time it posted the defensive blog, OpenAI tweeted that it had “heard questions about how we chose the voices in ChatGPT, especially Sky. We are working to pause the use of Sky while we address them.” A careful and considerate company was essentially admitting a mistake and preemptively taking stock, it seemed.
Nope. The source of OpenAI’s anxiety—the Black Widow herself—then appeared. In a statement to the press, she presented her side of the story: Altman approached her last year about cloning her voice, saying she could “bridge the gap between tech companies and creatives and help consumers to feel comfortable with the seismic shift concerning humans and Al.”
After she declined the offer and later heard the Sky demo, Johansson said, “I was shocked, angered and in disbelief that Mr. Altman would pursue a voice that sounded so eerily similar to mine that my closest friends and news outlets could not tell the difference.” Her lawyer then contacted Altman to ask for details about how OpenAI created Sky, prompting the company to take down Sky beginning Sunday night. She then suggested that legislation should be passed to protect individual rights from such intrusions.
Sure, it would have been embarrassing for Altman to proactively admit to his failed negotiations with Johansson in the original blog post or in a tweet. But not doing so only made him seem silly when the truth came out. Although Altman issued an apology hours later, saying OpenAI was “sorry to Ms. Johansson that we didn’t communicate better,” the saga over Altman’s unforced error surely isn’t over.—Amir Efrati
Deals and Debuts
See The Information’s Generative AI Database for an exclusive list of private companies and their investors.
Scale AI said it raised $1 billion from existing investors at a $14 billion valuation, up more than 80% from its valuation in a prior funding. The Information first reported on the deal in March.
Recommended Newsletter
Every weekday, The Briefing helps executives get smarter about the latest in tech, media and finance. Subscribe for free now.
New From Our Reporters
What We’re Reading
Thank you for reading the AI Agenda Newsletter! I’d love your feedback, ideas and tips: [email protected].
If you think someone else might enjoy this newsletter, please pass it forward or they can sign up here.