On a frosty February evening in Cambridge, Massachusetts, Mikey Shulman is spinning up a new song. His electric bass guitar hangs idly on a nearby wall. A 61-key synthesizer and drum kit remain untouched a few doors away. Instead, he types a few sparse phrases – pedal steel guitar, country Americana folk, acoustic guitar — into his startup Suno’s AI music generation software.

A few seconds later, a song comes to life: fluid guitar strums and human-sounding vocals with a smooth Southern accent soar over an upbeat tempo. It’s instantly catchy, like if Ella Langley met Lana Del Rey.

The tune isn’t a chart-topper or a summer hit, but it’s evidence enough for why more than 100 million people have now used Suno to make music. Suno-created songs have gone viral on TikTok, debuted on Billboard charts and racked up millions of streams. Over 7 million songs are made on the app every day, catapulting it to the top of the Apple App Store’s most downloaded music apps in April — surpassing Spotify.

“The technology finally allows for billions of people to be creative, to have the fruits of their labor, to feel fulfillment in a different way,” says CEO Shulman, 39. He calls it a “new form of consumer entertainment.”

But that’s come at the cost of professional musicians. In its early days, Suno said that it trained its AI model on tens of millions of copyrighted songs scraped from the internet, triggering fierce backlash. In 2024, some 200 artists including Katy Perry, Billie Eilish and Nicki Minaj called out AI companies for training their models on artists’ work without their permission. In July 2024, Universal Music Group, Sony Music, Warner Music Group and the Recording Industry Association of America hit Suno with a massive lawsuit, alleging that it illegally downloaded millions of copyrighted recordings from YouTube to train its model without getting permission from rights holders or providing compensation. “It’s a copyright chop shop,” an industry executive told Forbes.

Suno has denied the claims, and the lawsuit is ongoing. “What we do isn't illegal,” Shulman says. “It’s like listening to a lot of music and learning from it.” Instead, he argues, Suno is simply leveling the playing field. There’s an unfair asymmetry in the music world. Music production has historically been limited to a small pool of savants. Most people just consume, listening to music or singing along at concerts. But Shulman, who admits he’s average at playing the drums and guitar, says the lack of technical skill shouldn't stop anyone from becoming a musician. Now with AI, that’s possible.

“We've become the Ozempic of the music industry. It's like everybody's on it and nobody wants to talk about it.”

The music industry’s griping hasn’t stopped the four-year-old startup from becoming a smash-hit success. The startup’s annualized revenue tripled from $100 million in October (about $8 million that month) to $300 million in February (about $25 million that month). In 2025, the startup’s revenue was about $150 million, Forbes estimates.

VCs are sold too. Suno has picked up $375 million in funding from top-tier investment firms like Menlo Ventures, Lightspeed Venture Partners and Matrix, nabbing a $2.45 billion valuation in November. This year, the startup returned to the Forbes’ AI 50 list, which spotlights the most promising AI startups in the world, after debuting in 2025.

Today, more than 2 million users hand over anywhere from $8 to $24 per month so they can generate and download hundreds of songs (they also own the commercial licenses to their creations). Most people feed their own lyrics or prerecorded voice memos into the system, then add a short description of the style and genre they want it to produce, like “melancholic indie pop” or “soft pop piano ballad.” Hobbyists might use it to add a drum beat to vocals or adjust the pitch of their voice.

“Consumers don't experiment with tools,” says Menlo partner Amy Wu who led Suno’s $250 million Series C funding round. “They will only use a product if it's bringing them joy, really adding value to their lives.”

Despite the outcry, some professional music producers and songwriters are on board, using the program as a demo machine of sorts where they paste pre-written lyrics into the software to generate different ideas for a track before refining it further in an audio editor. But they’re doing it under the radar.

“We've become the Ozempic of the music industry. It's like everybody's on it and nobody wants to talk about it,” Shulman says. In September, Suno launched Studio, an audio workstation that allows users to create, edit and layer tracks, speeding up parts of producers’ workflows.

But while Suno might be fun for aspiring songwriters and people who want to make a customized song for their mom’s birthday, does anyone actually want to listen to AI music? Is it even good? And what does AI slopification mean for the human artists struggling to make it in an increasingly saturated music industry?

These are existential questions both for Suno and the industry itself. The record labels, after at first fighting Suno in court, have started to come around. In November 2025, Suno settled with Warner and struck a deal to use licensed recordings for its music generation model and limit downloads to paid subscribers. It’s a win-win for both companies. For the $6.7 billion (2025 revenue) record label conglomerate, the partnership is a “newfound revenue” source, CEO Robert Kyncl tells Forbes, allowing the company to tap into Suno’s revenue that it can then share with artists and songwriters who opt in to license their music for training. “Tools like Suno make it super easy [for anyone] to create,” he says.

Others, like $14.4 billion (2025 revenue) Universal Music Group, are in a deadlock. The record label believes AI-generated songs should be restricted to dedicated applications. They shouldn’t be downloaded and shared across social media and streaming platforms, where they compete with human artists and make it harder for them to get paid from an already-shrinking royalty pool.

“We want to make sure artists don’t have to compete with a machine that can spit out millions of tracks.” 

Michael Nash, chief digital officer and executive vice president at UMG, said on a Billboard podcast that this issue is a “hat hanger” in the settlement talks between the music giant and Suno after it sued Suno in 2024. And until that’s resolved, negotiations are at a standstill.

“There’s a limited number of ears in the world and a limited number of minutes or hours that people can listen to music,” says Ron Gubitz, executive director at the Music Artists Coalition. “We want to make sure artists don’t have to compete with a machine that can spit out millions of tracks.”

But Shulman knows that to some degree, he’s already won. Trying to discourage people from sharing AI music on social media is futile, especially since he sees Suno becoming a ubiquitous part of how music is made in the future. Plus, he scoffs at the music industry’s fixed-pie mentality: If AI encourages more people to engage with music, that could add more dollars into the ecosystem for all artists.

“I don't want to get to a world where there's a distinction between AI-generated music and non-AI generated music. It's all going to have AI in it somewhere,” he says, speaking to a roomful of Harvard Business School students in February. “A tremendous number of professionals use Suno, and it will end up in little bits and pieces in the music that you listen to.”

Luckily for human artists, fully Suno-generated tracks sound like something you might hear on the radio, and forget as soon as they’re finished. The ingredients are there, but anything fully AI generated, from lyrics to bass lines, lacks human quirks and emotional resonance. As one industry executive puts it: “Music is the language of emotion… Are you going to want a robot to tell you what it's like to get over a broken heart?”

Suno’s solution is a “weirdness” slider, which you can turn up to add abstraction and randomness to a composition. Around the 60% mark, it starts to produce decent ideas. But as good as AI is at aping the sounds of the past, it still can’t quite make something truly fresh without a human there to guide it — yet.

Music has always been a crucial part of Shulman’s life. Growing up in Peter Cooper Village in New York City, he began playing the piano at age four and switched to bass guitar at 12. In college, he started a band, performing at gigs in the city. “I pretend to play guitar. I pretend to play drums. I'm not very good,” he says.

Then he went off to Columbia University for a degree in applied physics. In 2015, Shulman got his PhD in physics from Harvard and joined data analytics firm Kensho as a machine learning engineer. There, he led a team of 40 data scientists and met his cofounders Georg Kucsko, Martin Camacho and Keenan Freyberg (all of whom have musical backgrounds). In 2018, after financial intelligence firm S&P Global acquired Kensho, the four of them worked on training audio models to transcribe earnings calls. In the evenings, they would go back to Camacho’s basement for jam sessions.

They soon realized that while audio data was messy and difficult to work with, it held a lot of value to train speech and music models. In February 2022, they left S&P Global to start Suno. In 2023, they released Bark, a popular open source text-to-speech model that generated sound snippets, realistic speech and non-verbal sounds like laughter and sighs.

“I don't want to get to a world where there's a distinction between AI-generated music and non-AI generated music. It's all going to have AI in it somewhere,”

Training AI to create music is tricky. The Suno team initially thought it would take 100 times more compute and money to train a model that could produce audio that sounded like music. Unlike text, which can be neatly parsed out into distinguishable units in the form of words and phrases, the fast and continuous signals that make up sound are far more challenging to encode. Soon after ChatGPT launched in November 2022, Suno had its first technical breakthrough: it figured out a way to represent elements of music that helped teach the model the structure and style of a song. Sitting at a kitchen table at Kucsko’s Cambridge house in late 2022, the four of them listened with excitement to the model’s first AI-generated melody that actually sounded like a song.

“Without really telling it anything, it learns to build up from short little audio pieces to longer and longer coherence, all the way to full songs,” CTO and cofounder Kucsko says.

They first launched as a Discord bot in September 2023. Within the first month, more than a thousand paying subscribers from all corners of the world like Brazil and Portugal flocked to the tool, even though early versions of the model sounded terrible, Kucsko says. That opened it up to a “fire hose of feedback,” cofounder and president Camacho says, prompting the team to add the ability for people to upload their own recordings.

The cofounders intentionally never taught the model the traditional rules of music. Limiting its understanding to 12 notes would mean the AI wouldn’t be able to conjure up a sound that’s completely new. “The crazy thing is the model doesn't even know that there are instruments or voices,” Shulman says. “It's all just sound.“

Suno’s technical edge might not last long as competitors pick up momentum. Rival Udio, founded by former Google DeepMind researchers in 2023, has $10 million in seed funding and 3.3 million monthly users. (It reached settlements with UMG and Warner after it was sued by the record labels in 2024 alongside Suno.) In February, Google acquired Suno challenger ProducerAI and trained its own music generation model, Lyria 3, which is embedded in Gemini and can compose entire songs including vocals, lyrics and instrumentals. Suno’s key differentiator isn’t the model itself, Shulman says: It’s the product that keeps users coming back to make more music.

Forbes 2026 AI 50 List

The Forbes Artificial Intelligence 50 List of 2026 spotlights promising AI-driven businesses. See the leaders driving the future of the industry.

VIEW THE FULL LIST

Musical AI slop is already starting to dominate streaming platforms. In April, French music app Deezer said 75,000 AI tracks are uploaded every day, making up about 44% of total daily uploads. Last fall, Spotify said it purged 75 million “spammy” tracks from the platform. Apple Music has started requiring record labels and distributors to include “transparency tags” for music uploads where a material portion of the content (album artwork, song, lyrics and music videos) is created with AI. Now, Spotify is rolling out a similar feature to credit AI in parts of a song where it was used.

But the actual consumption of AI music is low, only accounting for 1% to 3% of total streams, according to Deezer. A leading executive at a major record label says AI music is “more prevalent in the narrative than in practice yet.” “There are some hits here and there,” he says. It’s an open question whether AI music will be a commercial hit in the future.

Worse, 85% of streams on AI-generated songs were deemed fraudulent in 2025, Deezer said. “It's a royalty diversion scheme tool that has been characterized as a fraud fodder factory,” an industry executive tells Forbes.

“You’re not going to win. There’s no fighting AI.”

Some artists have already seen the impact. Tony Justice, a Knoxville, Tennessee-based independent artist best known for country hits like “Last of the Cowboys” and “Life on 18 Wheels,” says his income has taken a hit as AI-generated songs infiltrate streaming apps. He says he’s had to look for sponsors and other avenues to make up for that lost revenue.

“You just feel robbed,” Justice says. “You feel hijacked and stripped of all your hard work.” In June 2025, he filed a class action suit against Suno along with thousands of other artists, alleging copyright infringement. The startup has moved to dismiss some of the claims and the lawsuit is ongoing.

While some musicians have outright rejected Suno, others are starting to embrace it.

“You’re not going to win. There’s no fighting AI,” Grammy Award-winning DJ and music producer Diplo said in a recent interview, adding he doesn’t need anyone to sing his songs anymore because AI voices are so good. Alex Pall and Drew Taggart of the Chainsmokers have used AI tools like Suno and Udio to spin up new ideas for songs, like using it to see what a song would sound like in a female voice. Pharrell Williams has said AI can help with the “minutiae” of song making. Rapper, singer and songwriter Will.i.am is a vocal proponent, even teaching a course on AI agents at Arizona State University. He’s said 2026 is the last year that only humans walk the red carpet for Grammys. Next year, it’ll be “robotic artists” who’ve used AI to create their hits.

About a year ago, Los Angeles-based rapper Thurz started using Suno to create his most recent album. He typically samples music from big-name artists and records, then pays their estate a 65% cut of the copyright and royalties, leaving barely anything for him and his collaborators. This time however, he turned to Suno to generate compositions that sounded like they’re from the ‘60s or ‘70s. Now he just pays a $24 subscription fee. The AI software has also helped him save time on going into the studio to record and produce tracks.

“If I don't have a producer that's accessible for me at the time, I can literally beatbox an idea,” Thurz tells Forbes. “I could do a little drum percussion sound for eight bars and [Suno will] turn that into a real idea.”

Shulman thinks the bigger business opportunity is with consumers. In the future, he sees a world where AI music generation becomes a part of how musicians connect with fans. For example, Taylor Swift could release an interactive version of her album with lyrics or samples that fans could play with for an extra fee, or an unfinished song that they could use AI to complete, he says.

Even with the industry’s sentiment starting to turn in Suno’s favor, AI music is still fraught and messy. In October, a song called “I Run” by the producer Haven went viral on TikTok, started trending on the Billboard charts and hit 13 million streams on Spotify. But the voice, which Haven had generated using Suno, sounded suspiciously like British artist Jorja Smith. The producers said it was unintentional, but when Smith’s label complained, Spotify and other streamers pulled the song down for artist impersonation.

So Haven re-recorded the song — with a real human voice this time. It’s now been streamed more than 160 million times.

More from Forbes

ForbesWhy This AI Law Firm Is Ditching The Billable HourForbesThis 28-Year-Old AI Founder Thinks His Customer Service Startup Can Beat Out Companies 10x His SizeBy Rashi ShrivastavaForbesCoders Worry The AI From This $2 Billion Startup Could Replace Their JobsBy Rashi ShrivastavaForbesHow France’s Mistral Built A $14 Billion AI Empire By Not Being AmericanBy Iain Martin