NextFin News - South Korea's K-pop industry, worth an estimated $10 billion a year to the national economy, is making a structural bet on artificial intelligence - and it is not about replacing its human idols. The signal that turned AI from a creative experiment into a boardroom imperative arrived with HYBE's 2025 results: record revenue of 2.65 trillion won ($1.86 billion), up 17.5% year on year, alongside an operating profit that collapsed 72.9% to 49.9 billion won. The industry is not automating because the technology is ready. It is automating because the human-idol model is running out of margin.
The Margin Problem That Made AI Unavoidable
The numbers frame the paradox better than any press release. HYBE, the largest of South Korea's entertainment agencies and the company behind BTS, delivered its highest-ever annual revenue in 2025. But the bottom line moved in the opposite direction: operating profit fell 72.9%, and the operating margin shrank from 8.2% in 2024 to 1.9% in 2025. The market read that immediately. In late July 2026, even as BTS's Arirang tour drove second-quarter concert revenue up 243.3% year on year and 630% from the previous quarter, HYBE's shares fell 16.09% in a single session, wiping roughly 2.845 trillion won ($1.96 billion) off its market value. Investors rewarded the human star power and punished the margin.
That punishment is the engine behind the AI turn. K-pop is bigger than ever - more than 300 entertainment companies generate an estimated $10 billion a year in GDP for South Korea, with BTS alone contributing $3.54 billion, research published in a U.S. law journal estimates. Yet the model that built that fortune carries structural cost pressures: training programs that run for years before a group debuts, careers interrupted by mandatory military service, scandals that can erase a brand overnight, and a hard physical ceiling on how many concerts one body can perform in a year. HYBE staged 279 global performances in 2025, up from 172 the year before - a 62% increase in show count that helped lift concert revenue 69%, yet operating profit still fell nearly three-quarters.
The transmission channel is straightforward. A human idol has 24 hours in a day and a career that may last a decade before service or age intervenes. AI-generated or AI-assisted content has neither constraint. HYBE's Midnatt project demonstrated the mechanism in May 2023: using voice technology from Supertone, the AI-audio startup HYBE had invested in during 2021 and acquired a majority stake in January 2023 for 45 billion won (about $32 million), singer Lee Hyun's single "Masquerade" was rendered in six languages - Korean, English, Japanese, Mandarin, Spanish and Vietnamese - with near-native pronunciation, enabling simultaneous global releases from a single recording session. The same system altered his voice timbre to perform a female vocal section. One session, six markets, no additional flights.
That is not a gimmick; it is a unit-economics argument. Every additional language version, every licensed product, every virtual appearance carries a marginal cost close to zero once the model is trained. For an industry where developing a group can cost millions before a single note is released, the operating leverage is the point. The second-order effect is a shift in bargaining power: the scarce asset moves from the idol's time to the agency's models and intellectual property, which can be licensed, localized, and scaled without the artist boarding another plane.
Four Agencies, Three Strategies, One Direction
The Big Four agencies - HYBE, SM Entertainment, JYP Entertainment and YG Entertainment - have stopped debating whether AI belongs in K-pop. They are integrating it into their core business models, each on a different track, and the diversity of approaches reveals how unsettled the end state remains.
SM Entertainment moved first on narrative. Founder Lee Soo-man predicted in the early 2010s that entertainment would evolve into "a world of robots and avatars," and the 2020 debut of aespa operationalized that vision: four human members paired with virtual "ae" counterparts who inhabit a digital universe called KWANGYA. The AI is part of the story, not merely the production pipeline - a conceptual wedge that embeds technology in the group's identity from day one.
HYBE moved on infrastructure. After investing in Supertone in February 2021 and acquiring it in January 2023, the company progressed from tool to product. In June 2024 it launched Syndi8, a four-member virtual pop group whose vocals are entirely generated by AI, operating inside a fictional universe called Nansy Land. Its debut single, "MVP," was released globally on June 27, 2024. HYBE has also pushed the technology toward live performance, where Supertone's real-time voice conversion - first demonstrated when Midnatt transitioned between his own vocal and a female vocal onstage with imperceptible latency - targets the only venue that still requires a body: the stage.
JYP is building the capability organizationally. In December 2024 it rebranded its JYP360 subsidiary as Blue Garage, positioning AI and platform technology as central to expansion, then announced dedicated AI-artist development plans in September 2025 alongside large-scale recruitment - a signal that the company treats this as a hiring and capability problem rather than a one-off project.
YG is the control group. Rather than develop technology in-house, its subsidiary YG Plus handles album and content distribution for Plave, one of the most commercially successful virtual idol groups. That gives YG exposure to the virtual market's data and revenue without taking direct technological risk - a hedge, not a bet.
"I do not know how long human artistes can be the only ones to satisfy human needs and human tastes. That is becoming a key factor in my operation and a strategy for Hybe."
HYBE chair Bang Si-hyuk has framed the shift as a move to "Enter-Tech" - a fusion of entertainment and technology - and has called for broader social consensus on the ethical use of AI in creative fields. The admission that society must be brought along is itself a tell: the technology is ready, but the license to use it commercially is not fully granted.
Virtual Idols Are a Parallel Market, Not a Replacement
The most common misread of this trend is that AI idols are coming for human jobs. The evidence points the other way: virtual performers are building a parallel market with its own economics and its own fans, one that coexists with rather than cannibalizes the human business.
Plave, which debuted in March 2023 under virtual-IP startup Vlast, is the proof case. Unlike fully AI-generated groups, Plave uses real performers in motion-capture suits, with AI refining facial expressions and motion to bridge the so-called uncanny valley. The commercial results are not niche: in March 2024 Plave became the first virtual idol group to take first place on a national music show; its second EP, "Asterum: 134-1," topped major Korean streaming charts; and its concerts drew a total of 40,000 fans across a fan concert in April 2025 and two encore shows at Gocheok Sky Dome in October. In 2026 it entered the KSPO Dome - widely regarded as the dream stage for Korean artists - and launched its first Asia tour less than two years after debut.
"Every creator and artiste will be able to explore and experiment with new ideas using AI technology. This is an opportunity for creators."
Kyogu Lee, founder and chief executive of Supertone, made the point at a technology conference in July 2025. The broader market is pricing the same shift. The global AI music market was approximately $5.2 billion in 2024 and is projected to reach $60.4 billion by 2034, a compound annual growth rate of roughly 27.8%, research firm Market.us estimates. The global virtual human and avatar market is projected to surpass $15 billion in the coming years. These are additive growth curves, not replacement curves.
The most ambitious expression of coexistence comes from Galaxy Corporation, the company of K-pop icon G-Dragon, which is developing humanoid robot idols designed to perform on real stages and interact with fans in person. At the ComeUp 2025 startup conference in Seoul, CEO Choi Yong-ho forecast an era in which real idols, virtual idols and robot idols coexist within a hyper-converged entertainment ecosystem, demonstrating the concept with a humanoid robot performing choreography onstage.
"We're only at 1 or 10% of what's possible."
The Counter-Thesis: Authenticity Is the Product, and AI Cannot Manufacture It
The strongest case against the AI thesis attacks it at its foundation: K-pop does not sell music. It sells perceived human connection - the parasocial bond between fan and idol, built on vulnerability, effort, and the knowledge that the person on stage is real and could fail. If authenticity is the product, then AI-generated performers are selling a counterfeit, and the market will eventually discount them.
This view has evidence on its side. HYBE's own shares were punished in July 2026 despite record results powered by BTS - the most human of assets. The market's verdict was clear: it rewards human star power, not the Enter-Tech narrative. And HYBE's 1.9% operating margin for 2025 shows that AI investment has not, so far, repaired profitability. The company is spending heavily on technology while its core economics deteriorate - which looks less like a transformation and more like an expensive distraction. There is also a regulatory overhang: South Korea is only the second jurisdiction in the world, after the European Union, to adopt comprehensive AI legislation, and the rules around voice cloning, likeness rights, and deepfake content remain in flux. A restrictive regime on commercial AI performances could strand parts of the industry's investment.
The counter-thesis is not a strawman, and it deserves its weight: the agencies may be solving the wrong problem. If fans want humans, then margin compression is simply the cost of doing business in a human industry, not a bug that AI can patch.
But the counter-thesis conflates two different value pools. The human-idol business - concerts, albums, fan meetings - is not being replaced. It is being supplemented by a second pool where the product is not human authenticity but scalable, always-available content. Plave's 40,000-fan concerts prove fans will pay for virtual performers; HYBE's Artist Indirect-Involvement segment, which monetizes an artist's name and likeness without requiring active participation, grew 19.3% in 2025 to 965.8 billion won, and its merchandising and licensing line jumped 35.8% to 570.6 billion won. The question is not whether AI replaces humans. It is whether the parallel pool grows fast enough to change the industry's margin structure.
Verdict: Structural Shift, Cyclical Overlay
This is a structural shift with a cyclical overlay, and confusing the two is the easiest way to misread the moment.
The structural leg is real and durable. Generative AI has permanently lowered the marginal cost of content creation, voice synthesis, and virtual performance. That cost curve does not revert on its own. The agencies that build the capability - HYBE with Supertone, SM with its avatar narrative, JYP with Blue Garage - are repositioning for a world where content scales without proportional labor. The evidence for a regime change is concrete: voice-synthesis quality crossed a commercial threshold in 2023 with Midnatt's six-language release, real-time conversion now works at live-stage latency, and a new legal regime is being written around the technology rather than against it.
The cyclical leg is the post-pandemic concert boom and the BTS return. HYBE's 243.3% second-quarter concert-revenue surge, its 119 concerts in the first half of 2026, and the more than 200 additional shows expected in the second half - which together would be the most since 2021 - are a cyclical high. When that wave recedes, the revenue base will fall, and the AI infrastructure will be tested on its own merits rather than on the glow of a reunion tour.
Three signals will separate the structural story from the narrative. First, margins rather than revenue: HYBE's Artist Indirect-Involvement segment is the cleanest read - if merchandising and licensing does not deliver AI-attributable margin expansion in the full-year 2026 results, with operating margin recovering toward 5% or above, the "AI as margin rescue" thesis is wrong. Second, virtual-idol scale: Plave's 40,000-fan benchmark is the base rate - if virtual-act concert attendance fails to grow for two consecutive years, the parallel-market thesis weakens materially. Third, regulation: South Korea's AI legislation will set the boundaries for voice cloning and likeness rights, and restrictive commercial rules would raise the cost of the entire Enter-Tech model.
The base case is that AI becomes a production and licensing layer beneath the human-idol business: human stars remain the premium product while AI handles multilingual releases, back-catalog monetization, and virtual appearances, with margins stabilizing but not re-rating dramatically. The upside case is that AI-generated acts such as Syndi8 achieve mainstream chart success and touring viability, opening a high-margin second revenue pool that lets the market re-rate agencies as technology-enabled IP owners rather than talent managers. The downside case is that fan backlash and regulatory constraints limit AI's commercial use, leaving the technology as a cost center and the Enter-Tech narrative to unwind.
K-pop's AI turn is not a bet against human idols; it is a bet that the industry can finally monetize the part of stardom that does not require a human body. If the margins follow, the idols were never the product - the IP was.
Data as of August 25, 2026. Company figures from HYBE's regulatory filings and earnings releases; market projections from cited research firms.
Explore more exclusive insights at nextfin.ai.
