Mikey Shulman on Suno and the Future of Making Music

Suno CEO Mikey Shulman tells Harry Stebbings that his goal is to make more people active participants in music: creating, editing and sharing as well as listening. He describes the ambition as making musicians, not just making music. Stebbings says he uses and pays for Suno, while Shulman leads the company whose product they discuss.

The team initially expected to build an enterprise audio-understanding tool, drawing on its natural-language processing background. Shulman says audio generation improved sooner than anticipated, so the team quickly shifted its attention to making things rather than only analyzing them.

He cautions against assuming that scaling music models will solve every problem in the same way as improving objective benchmarks in text. Music is subjective, he says, and developing models with good taste requires techniques beyond simply making them larger.

Shulman wants music to be more like a video game: interactive, rich, enjoyable alone and with friends. He believes people will pay for a more engaging experience, though the size and shape of that future market remain uncertain.

Both expect customers eventually to care more about the product experience than which model powers it. Shulman predicts that model versions may cease to be the focus of releases, but does not know when that might happen.

Suno uses a transformer model, Shulman says. He describes its competitive focus as audio representation rather than inventing a new model architecture, building on techniques from the open-source text-model community.

Stebbings asks whether greatly increasing supply will make music cheaper. Shulman accepts that the average value of an individual piece of content may fall, but argues that broader participation and engagement could increase the value of music overall. This is the Suno CEO's economic view, not a demonstrated market outcome.

On quantum computing, Shulman sees extraordinary potential but believes fundamental physics and research challenges remain. He questions whether conventional venture timelines match the field and suggests public investment may be needed. His career and funding advice reflects his assessment of the risks, not a forecast he can guarantee.

Shulman looks for underappreciated talent because Suno cannot match the pay offered by the largest AI labs. He values economists' ability to reason about natural experiments and experimental physicists' ability to run high-quality experiments quickly.

Suno is headquartered in Cambridge, Massachusetts. Shulman thinks working on the distinctive problem of aligning models with subjective human taste can attract researchers; he notes that existing preference-alignment techniques may not be the right long-term answer for music.

Usage data and A/B tests can inform the product, and Shulman expects more personalization in future. He acknowledges that it is difficult to decide how a music model should represent different people's tastes.

Asked whether large model providers will enter applications, Shulman expects them to try, though he does not know whether music will be a priority. He argues that Suno must win through pleasurable product experiences, not merely a larger underlying model.

Stebbings praises the speed of creating a first song. Shulman says Suno closely watches the first experience and has observed that even small artificial delays make users like it less. In the company's tests, ten seconds was worse than eight; he does not expect one-second generation soon.

One signal the company watches is whether users reach the free-tier paywall on their first day. Shulman interprets that as a sign they enjoyed spending time creating, even if they do not subscribe. A single discarded song with no return visit suggests the experience did not work.

Suno began charging from its first Discord product while retaining a free tier. Shulman says early subscriptions signaled that people valued the service, offset GPU spending and helped the team identify whom to interview: those who subscribed and those who reached the limit but did not.

GPU spending is the company's biggest expense and, Shulman says, several times payroll; Suno also operates a substantial research cluster. He hopes compute prices fall but expects the company might use more compute. He argues that simply hiring more researchers will not produce proportionate gains.

Stebbings raises the June 2024 record-industry lawsuit over training data. Shulman acknowledges that Suno's training data includes some copyrighted works and asserts that this is lawful. That is Suno's position in a disputed legal matter, not a settled legal conclusion; the interview does not establish the lawsuit's outcome.

Shulman says litigation consumes time and money and would prefer cooperation with the existing music industry to grow the market. He says some industry participants already work with Suno and that some parties had spoken with the company before the suit. His description reflects the interests and perspective of the defendant company.

He does not know how disputes between AI companies and incumbents will ultimately be resolved. He rejects both an approach that ignores incumbent concerns and one that seeks to eliminate new entrants through litigation. Losing a trial would be bad for Suno, he says, though he does not believe it would necessarily end the company.

Shulman wants a much larger population to enjoy making music rather than merely giving existing creators a modest productivity improvement. He thinks conventional music creation requires practice and tools that many people find difficult; Stebbings counters that challenging activities can themselves be rewarding.

Whether AI-generated music can safely train future AI models is, Shulman says, an open technical question: feedback into training could introduce serious problems. Suno works on it somewhat, but he hopes larger model developers solve the most difficult parts.

He would prefer listeners and creators not be separated into two disconnected worlds of AI and other music. Switching apps creates friction, and he imagines making and sharing music connected to listening at parties, during workouts or on a commute. He does not prescribe which companies must combine their products.

Asked how artists feel about Suno, Shulman says most artists he personally speaks to use and like it. He explicitly recognizes that this is a biased sample; it does not establish the views of artists generally.

Stebbings proposes an artist-controlled model trained on that artist's own material. Shulman says he would welcome such arrangements, subject to rights and contracts. He says Suno currently blocks prompts using an artist's name and positions the service for original music rather than impersonation.

With an artist's consent, Shulman imagines tools for the artist, songwriters or even fans to make authorized variations. He also considers whether an artist could receive a share when users create music inspired by them. These are proposed models, not announced agreements or existing entitlements.

He says Suno users pay for the experience of creating, distinct from listening. He hopes new models could allow fans to pay artists directly or pay for an artist-approved version of a model. In his view, such interactions need not fit the existing division of streaming revenue.

A remix contest with Timbaland illustrates the kind of engagement Shulman wants: fans could work with the music of a favorite creator. He says the producer works with Suno on product development, receives early access and provides public recognition for the company.

When asked about compensation, Shulman says Suno does not pay Timbaland cash and identifies him as an adviser. The conversation discusses equity in that context; the partnership is a commercial relationship and should be considered alongside the endorsement.

Shulman thinks pop has become more homogeneous in melody, harmony and song structure, shaped by digital production, streaming and social algorithms. He also notes that those same algorithms can bring music to far more people; he does not frame every change as inherently good or bad.

Music discovery is already heavily algorithmic, he argues. What becomes popular can reflect platform recommendations and promotion as well as the music itself. He cautions that people at Spotify or TikTok would know more about their systems than he does.

He sees TikTok as important for short-form discovery and Spotify for its algorithmic playlists. He hopes music services become more engaging, noting that much listening today occurs in the background while people work or do other activities.

Shulman says generative-video companies have asked Suno for an API to add background music, but Suno has declined. His stated reason is strategic: the company wants to make music a more valuable, active experience rather than simply supply inexpensive background audio.

One product decision he regrets is not leaving Discord sooner. Suno launched a thin web app in November after starting on Discord in August; he says 90% of traffic moved to the web in five days. Discord remained useful as a community and feedback channel.

For Shulman, a good interface makes a previously difficult workflow intuitive. He gives the example of covering an entire song in another genre, rather than merely reproducing the controls of an old digital audio workstation.

He argues that an empty text box is not the ideal interface for every AI product. He hopes music tools will become intuitive enough that people no longer need to think in terms of prompts; in his view, the quality of the music matters more than the sophistication of the instruction.

Shulman says he sometimes wishes Suno had not made exceptions for remote work. The company was built in person, and he says some exceptions remain worthwhile, but judging and managing them becomes harder as the team grows. This is his management preference, not a universal finding about remote work.

He says Suno had raised a little over $125 million at the time of the interview. He views capital as a tool that can unlock qualitatively different opportunities. He also cautions founders to understand that investors' reserved capital and pro rata rights can favor existing investors in successful rounds.

Charging from day one gave the company a reason to pay attention to revenue. Shulman says building a large platform matters more to him at this stage, but revenue growth signals that users value the experience. He says Suno has prioritized growth while trying not to sacrifice revenue.

Music is inherently social, he says, but social experiences need not follow a single model. A platform might support private exchanges between friends alongside professional creators with large audiences, much as a social network can serve both intimate and public interactions.

Stebbings says the speed of change in work, content and information makes him uneasy. Shulman is optimistic, but rejects treating AI's effects as inevitable: he believes people must actively build the future they want instead of waiting for it to happen.

He identifies two futures he would dislike: unauthorized, uncompensated impersonation of artists, and a hyper-personalized stream so absorbing that music becomes less social. Both are hypothetical scenarios he uses to explain why design choices and consent matter.

A song made for a partner, by contrast, is social in his view. He does not want all music to become personalized creation; he also values art centered on an artist's own expression. His goal is a broader set of experiences rather than one dominant format.

Shulman predicts that taste and selection will matter more as tools make production easier, while traditional technical skills matter less. He compares selecting strong AI-generated music to curating playlists. This is his forecast for creative work, not a settled measure of artistic value.

Asked about an AI winter, he says he does not know whether disappointment among finance or technology insiders will extend to the wider world. He says his own view is shaped by his Cambridge environment, just as Stebbings sees a London finance perspective.

His instinct is to resist bans on social platforms, though he stresses that he is not an expert on the question. As Suno CEO, he says his hardest task is focus: the company may see dozens of important opportunities but can pursue only a few at a time.

Shulman says having children made him more ruthless about prioritization and more patient with fast learners. He also argues that people can overlook a technology's main effects by focusing on secondary consequences.

In his MIT Sloan teaching, he sees AI as potentially giving people access to a competent tutor or assistant, while recognizing that it makes existing homework easier to complete. His response is to change what he teaches to prepare students for a world where such tools are used. This is his view of education, not proof of equal access or effectiveness.

Unexpected Suno uses include making music with children and personalized sales or marketing outreach. Shulman says the company noticed commercial use through business-card spending data, but does not claim a measured effect on sales outcomes.

For the future of music, Shulman hopes many more people will spend more time making and sharing it, with AI enabling participation and personalization. He predicts more people could earn a living from music, comparing the change to digital photography; that is an aspiration and forecast from the CEO of an AI music company.

Stebbings thanks Shulman for the conversation and says he enjoys the product. Shulman thanks him in turn.