Should We Invest Now That AI Music Finally Learned to Stop Rambling?

I noticed an interesting detail: In Mureka V9.5's demo, a piano intro used a full 8 bars of whitespace—just a single note repeating, before slowly entering the verse. In the field of AI music generation, this is almost a technically "unconventional" operation.

Over the past two years, I've seen business plans from at least 30 AI music projects. Almost every one emphasized "melodic richness," "generation speed," and "multi-style coverage." But no one told me why these generated songs sound like cans stuffed with information—melodies so dense there's no room to breathe, vocals like distress signals drowned out by accompaniment, and dynamic differences between sections nearly zero.

This isn't a technical problem, it's an aesthetic problem. And aesthetics are precisely the hardest barrier for AI to break through.

Mureka V9.5's answer sheet, in my view, has its core breakthrough not in being "more accurate," but in being "emptier." It learned to leave whitespace in melodies, let transitions before choruses decay naturally, lower the intensity of verses, and make chorus explosions layered. The addition of this "breathing sense" makes AI music sound less like AI music for the first time.

How high is the technical barrier?

I did a simple comparison, having Suno V4, Udio 1.5, and Mureka V9.5 each generate 10 pop love songs, and asked three music producers to blind test them. Scoring dimensions were as follows:

Dimension	Suno V4	Udio 1.5	Mureka V9.5
Vocal separation (clear distinction from accompaniment)	6.2	5.8	8.7
Section dynamics (strong/weak contrast)	5.0	5.5	8.9
Melodic breathing sense (whitespace and pauses)	4.3	4.7	9.1
Lyric naturalness	7.1	6.9	7.8

This table illustrates one thing: Mureka has opened a clear gap in "dynamic control" and "structural narrative." These two abilities are precisely the weakest links in traditional AI music generation models.

Why? Because most AI music models are trained using "next token prediction," making the model naturally inclined to "fill" every time step, just as language models tend to ramble. To teach the model to "shut up at the right time," you need to annotate massive amounts of professional studio recordings with rests, weak dynamics, and transition sections in the training data, and add penalty terms for dynamic contrast in the loss function.

This requires three conditions: authorization of high-quality copyrighted music libraries, frame-level precise dynamic annotation, and a team that understands music, not just algorithms. As far as I know, the Mureka team has at least two engineers who worked at Grammy-winning recording studios. This background is a scarce resource in the AI music track.

Is the business model viable?

Current main monetization paths for AI music:

• C-end subscription: Monthly fee $9.9-$29.9, targeting short video creators and independent musicians
• B-end licensing: Film scores, game sound effects, ad music, charged per project
• Copyright revenue sharing: Generated songs played on streaming platforms, sharing profits with platforms

Mureka V9.5 currently focuses mainly on C-end, but I think its technology is better suited for cutting into B-end. The reason is simple: C-end users actually have a high tolerance for "AI flavor." Background music generated by AI on TikTok/Douyin is sufficient for dress-up videos. But B-end—like film score directors or game sound designers—are highly sensitive to "breathing sense" and willing to pay a premium for "not sounding like AI."

What is the exit path?

The most likely acquirers in this track are Adobe, Logic Pro, or Apple Music. Adobe needs AI music embedded in Premiere Pro; Apple needs to supplement Apple Music's creation toolchain. If Mureka can secure more than 20 B-end clients and annual contract values exceed $5 million, the acquisition valuation could reach $150-200 million.

But here is a risk: Is the technical moat deep enough? Dynamic annotation and aesthetic data don't sound as hardcore as algorithm patents. Once top companies (like OpenAI's Jukebox 2.0) join this direction, Mureka's lead advantage might be quickly erased.

My judgment: Cautiously optimistic, but need to watch actual paid conversion rates. If Mureka can achieve a paid user ratio of over 15% among monthly active users in the next 3 months, with retention above 60 days, I would consider following up. Otherwise, this niche track might be more suitable as a "technology validation" rather than an "investment target."

Ultimately, the problem with AI music has never been "can it generate," but "after generation, are people willing to listen again." Mureka V9.5 gave me the impulse to "want to finish listening to this song" for the first time. That itself is an investment signal. But between impulse and business model, there is a long monetization path.

Original link: https://www.tmtpost.com/8073797.html