SM and JYP's investment arm weigh bets on AI idols that talk to every fan differently

Two of K-pop's biggest companies are looking at a different kind of AI idol. Chosun Biz reported on 20 Oct 2026 that JYP Partners, JYP Entertainment's investment arm, and SM Entertainment recently held investment talks with companies that make AI-based virtual artists. Nothing is final. According to the Korean original, some parties have given a "soft commitment", a tentative signal of intent to invest after receiving the companies' materials. The report doesn't name the startups or give any amounts.

What would these AI idols actually do?

They wouldn't copy a real member's voice or face. Chosun describes a model built to talk with each user one-to-one and form a different relationship with each. The AI writes its conversation and reactions in response to what the fan says. It can also send the fan personalised images made to fit what they've been talking about. The stated goal is attachment: the same character should feel like "my own idol" or "an idol I nurture" (내가 키우는 아이돌, in Chosun's own English rendering) to each person.

That is a break with how virtual idols have worked in Korea so far. Chosun's example is Plave, a five-member group that debuted in 2023. Real people perform the voices and movements, and fans call the character on screen the "front singer" and the person behind it the "back singer". Chosun says Plave's agency, Blast, made 85.5 billion won in revenue last year, up about 88%, on album and merchandise sales. The model now being discussed takes the back singer out.

Why not just clone existing stars?

Chosun gives a commercial reason. If an AI chatting with fans says or does something that doesn't match a real artist's personality or values, it could damage that artist's image. So the companies are putting more weight on owning new virtual-artist IP than on recreating the idols they already manage.

Chosun also explains why the money goes through investment arms. Developers reportedly prefer raising investment to running joint ventures with entertainment companies, because joint projects can stall over disagreements such as revenue sharing. An investment-industry source told Chosun that running a service that produces different conversations for every fan needs technology the agencies don't have, and that buying into startups lets them test the business without building everything in-house.

Where does this sit after Supertone?

The talks follow a costly lesson at HYBE. In July 2026 Supertone's shareholders voted to dissolve the AI voice company, which HYBE controlled after putting in nearly 49 billion won across two investments. Music Business Worldwide reports Supertone's 2025 revenue at 2.3 billion won against an operating loss of 15.4 billion won, and notes that SYNDI8, HYBE's virtual girl group that debuted in 2024 with Supertone's voice AI, never became an established commercial act. HYBE called the shutdown part of "a realignment of its AI business strategy".

The other two companies have been circling the same idea. SM's first virtual artist, nævis, debuted in 2024, and The Korea Times reported that LG Uplus's generative AI would make her music videos, shorts and merchandise. JYP's tech unit, Blue Garage, was hiring LLM engineers and a "character interaction engineer" in September 2025 for a virtual AI artist. In a statement cited by Korean media, JYP asked: "Can AI become a being that calls fans' names and interacts with them?"

One caution on JYP Partners. In October 2025 DealSite reported that the wholly owned fund, set up in February 2023, had made no investments in the first half of that year and was short of the staff its licence requires. A deal here would be a real change of pace for it, which is another reason to treat "soft commitment" as a signal, not a done deal.

Chosun also cites a global virtual-artist market of $2.1 billion last year, projected to reach $6.24 billion by 2033. The article doesn't say who produced that forecast, so we'd read it as a sales pitch rather than evidence.

What we think

Supertone seems to have taught the agencies to rent the risk instead of owning it. Taking minority positions in startups is a cheaper way to find out whether fans will pay for a chatbot idol than buying a voice company outright. That's sensible for the agencies. The second-order effects are what deserve attention.

The real shift is who's in the costume. Plave shows that fans will back a virtual group with real performers inside it, and those performers are people with careers and some leverage. An AI idol that improvises every conversation needs no back singer. Korea's standard entertainment contract caps exclusive deals at seven years, so a human group eventually gets to decide whether to stay. A wholly owned character has no such term. It doesn't renegotiate, burn out or leave. Our read: the agencies' reason for avoiding clones is to protect brands, not artists, though it does mean real idols' likenesses aren't being turned into chat partners without them. That's a good outcome, even if it happens for the wrong reason.

The losers are trainees and the people behind the screen. If part of the fan-service job, the late-night messages and the sense of being known, can be done by software, that's one less job a human idol has to be hired for.

For fans, this sells attachment by design. The pitch Chosun describes is to make each fan feel the idol is theirs, and the closer that feels, the better the product works. The report doesn't say how the service would be charged for, what happens to the conversation data, or how younger fans would be protected. Those are the questions to ask before anyone signs.

For songwriters and producers, a character still needs songs. If these acts launch, ask who writes and produces the music, and whether those people are credited and paid like any other session work.