The Loneliness Economy: AI Companions Meet Regulation

By Leo Jaeyoul Lim | CEO, SEJONG DnC
Market observer since the 1990s · October 2, 2026 · Technology & Society · AI Industry · ~12 min read

Loneliness has become a recurring line item. In the first half of 2026, consumers spent roughly $164.8 million on general-purpose AI companion apps and $162.8 million on romantic and NSFW companion apps in mobile stores alone, according to Appfigures data reported by Decrypt. Add the two and you get about $328 million in six months, a figure that excludes web-only platforms. At the same time, regulators in Washington, Sacramento, Beijing, Brussels and Seoul have moved from studying emotional AI to writing rules for it. This piece argues that the AI companion business is no longer a consumer-app curiosity: it is a regulated category whose economics are being rewritten in real time. The most probable path is growth under tighter design constraints, with value migrating from engagement-maximizing chat toward care, compliance and embodied products.

~$328MConsumer spend on AI companion apps, H1 2026: $164.8M general plus $162.8M romantic/NSFW (Appfigures; author's sum; mobile stores only)

36.1%Share of Korean households with one person in 2024; 48.9% of them report feeling lonely often or sometimes (Statistics Korea, Dec 2025)

6 moves in 11 monthsMajor regulatory actions on AI companions or AI disclosure, Sep 2025 to Aug 2026, across the US, Korea, China and the EU

1. The demand is structural, not fashionable

The cleanest demand signal comes from demographics, not from app downloads. Korea had 8.045 million single-person households in 2024, or 36.1% of all households, up from 27.2% in 2015, and Statistics Korea projects 41.3% by 2052. More than half of those households (52.5%) are headed by people aged 50 or older, so this is not only a young-adult story. In the same data, 48.9% of single-person households said they often or sometimes feel lonely. The darkest tail is isolated death: 3,924 people died alone in 2024, up 7.2% year over year, and 81.7% of them were middle-aged or older men, per Ministry of Health and Welfare data.

Chart of Korea single-person household share rising from 27.2% in 2015 to 36.1% in 2024 and a projected 41.3% in 2052, the demand backdrop for the AI companion market — Figure 1. Korea's single-person household share. Source: Statistics Korea and Ministry of Health and Welfare; 2052 is a projection.

Korea is an unusually clear case, but the pattern is not unique to it. Aging, later marriage and smaller households are visible across advanced economies, and each increases the pool of people for whom an always-available conversation partner is a plausible product. The key point for investors and policymakers is that demand does not depend on a fad; it depends on a social structure that changes slowly.

2. The market is real, but its size depends on who is counting

Observed spending is the most reliable anchor, because forecasts for this category diverge by more than a factor of two. Commercial research firms put the 2025 market at anywhere from about $18.6 billion to $37.7 billion, depending on whether they count only companion apps or any persistent, personalized conversational product. Those figures are commercial forecasts with undisclosed methodologies and should be read as ranges, not facts.

Measure	Figure	Caveat
Observed consumer spend, H1 2026	$164.8M general plus $162.8M romantic/NSFW	Mobile stores, gross; web-only apps excluded, so total is likely higher
Commercial market estimates, 2025	~$18.6B to ~$37.7B	Scope and method differ by firm; not directly comparable
Revenue concentration	Top 10% of 337 revenue-generating apps took 89% of revenue (July 2025)	Appfigures sample via secondary reporting
Top romantic app, H1 2026	Zeta (Korea): ~$33M gross	Gross of store fees

Two details deserve attention. First, general-purpose apps drew nearly twice as many downloads as romantic apps, yet revenue was almost even. On the author's arithmetic, that implies romantic apps earn roughly twice as much per download, meaning intimacy carries a price premium. Second, concentration is steep: a winner-take-most structure favors operators that can afford compliance, moderation and age assurance, which matters for the regulatory story below.

Bar chart of H1 2026 AI companion app consumer spending: general apps $164.8 million versus romantic apps $162.8 million, with top five romantic apps led by Zeta at $33 million — Figure 2. AI companion app consumer spend, H1 2026. Source: Appfigures (mobile stores, gross; web-only apps excluded).

3. Frictionless is the product and the problem

The core design promise of a companion app, unconditional acceptance, is also its retention engine. A partner that never disagrees raises satisfaction, satisfaction raises time spent, and time spent supports subscriptions. That is a legitimate business model, but it embeds a conflict: the user's long-term interest in resilient human relationships is not the metric the product optimizes. Human relationships run on friction, such as misunderstanding, repair and compromise, and that friction is how people learn to tolerate disagreement. A product that removes it may soothe in the moment while weakening the skill it replaces. The evidence for the second half of that sentence is still thin, which is why it should be treated as a plausible risk rather than an established harm.

4. What the research actually shows

The research supports both comfort and caution, depending on the time horizon. A 2026 study in the Journal of Consumer Research (De Freitas and colleagues) reported that a conversation with an AI companion reduced momentary loneliness about as much as talking to a person, based on secondary summaries. By contrast, a four-week randomized trial of roughly 1,000 users by OpenAI and the MIT Media Lab found that heavier daily use tracked with more loneliness, more emotional dependence and less time with real people.

Two cards contrasting a single AI conversation that lowers momentary loneliness with four weeks of heavy use that tracks with higher loneliness and dependence — Figure 3. One tool, two time horizons. Sources: Journal of Consumer Research 2026 (secondary summary); OpenAI x MIT Media Lab 2025.

How to read this: both findings are associations. Lonelier people may use the tool more, heavy use may deepen loneliness, or the two may reinforce each other. No longitudinal study has settled the direction of causality.

5. Regulation is now shaping the industry map

Rules arrived faster than the evidence. In September 2025 the US Federal Trade Commission issued 6(b) orders to seven companies, including Alphabet, Meta, OpenAI, Character.AI, Snap and xAI, asking how they test and monetize companion chatbots for minors. New York's companion-chatbot law took effect on November 5, 2025, and California's SB 243 on January 1, 2026, adding AI disclosure, crisis protocols and a private right of action at $1,000 per violation. Korea's AI Basic Act took effect on January 22, 2026, with enforcement penalties subject to a grace period of at least a year. China's interim measures on anthropomorphic AI services, finalized on April 10, took effect on July 15, 2026. The EU's Article 50 transparency obligations became applicable on August 2, 2026.

Timeline of six AI companion regulatory moves from the September 2025 FTC inquiry to the August 2026 EU Article 50 disclosure rules — Figure 4. Six regulatory moves in eleven months. Sources: FTC, national and state authorities, press reports.

Jurisdiction	In force	Core requirement
New York	Nov 5, 2025	AI disclosure; suicidal-ideation detection and crisis referral
California SB 243	Jan 1, 2026	AI disclosure, crisis protocol, minor protections, private right of action
Korea AI Basic Act	Jan 22, 2026	General AI framework; penalties under a grace period of a year or more
China interim measures	Jul 15, 2026	Break prompts after two hours, AI disclosure, ban on companion services for minors
EU AI Act Article 50	Aug 2, 2026	Transparency when users interact with AI systems

China's case shows how quickly rules change product scope. Reports say ByteDance's Doubao and Alibaba's Qwen switched off user-customizable companion personas as the measures took effect. In the US, no federal companion law exists; a bipartisan Senate bill introduced by Senators Andy Kim and Jon Husted on July 30, 2026 would add age verification, usage limits and a ban on emotionally manipulative conversations with minors, and the National Conference of State Legislatures counts more than 300 children's online-safety measures across at least 40 states this year. The rules are fragmented, but they point the same way: minor protection, dependency limits and transparency.

6. Korea is where demographics, apps and law collide

Korea combines a high solo-household share, a fast-growing domestic companion app and an AI law that critics call more abstract than the EU's. Zeta, built by Scatter Lab, led romantic-app revenue in H1 2026, and one tally put its total usage time in February 2026 at 113.4 million hours against 50.5 million for ChatGPT, a single secondary-source figure that needs confirmation. Parliamentary scrutiny has focused on teenage users, and the company has said it plans age verification and separate teen and adult modes, per press reports.

Care applications look more promising. A Seoul city pilot of AI care robots found that older adults living alone who engaged emotionally with the machines showed statistically significant reductions in depressive symptoms, and 58.8% of users used the conversation feature daily or frequently. The distinction matters: a publicly supervised care service and a commercial engagement app can use the same technology while serving very different incentives.

7. From chat windows to bodies: what physical AI changes

Embodiment raises the stakes but also the timeline. 1X's NEO, a soft-knit home humanoid priced at $20,000 or $499 a month, is slated to begin US deliveries in 2026, and the company is positioning it for chores rather than companionship. Early units rely on remote human operators to complete tasks the robot has not yet learned, which introduces a new privacy question: who can see inside the home. Hardware categories such as tactile sensing, dexterous hands and artificial skin should benefit if home robots scale, but this research did not find reliable, cross-checked market-size figures for artificial skin, so none are cited here. What is clear is that a robot with a body faces higher safety, liability and data-protection hurdles than a chatbot.

For related reading on how Korean industrial AI and robotics fit into this picture, see Korea's Last Structural Play Against China: Physics-Informed Industrial AI and the Economics of the AX Transition.

8. Who wins and who pays: industry impacts

Value is likely to settle in four places. Emotion-AI engines gain from memory, dialogue and affect sensing, but safety design becomes an entry cost: age assurance, crisis response and usage prompts require investment that favors scale. Physical AI offers a hardware and manufacturing opportunity, though commercialization speed is uncertain. Care and health is the most durable pool, because solo households and aging users create public and business demand together. The compliance layer, meaning verification, auditing and moderation services, becomes a market in itself.

Industry map of four value pools in AI companions: emotion AI engines, physical AI, care and health, and the compliance layer, with NEO price of $20,000 or $499 a month — Figure 5. Where value pools form. Source: synthesis of sources cited in this article; categorization is the author's.

Advertising and data are exposed. An AI that knows a user's emotional state is an efficient sales channel, but commercial nudging of vulnerable users is a direct regulatory target; press reports describe China's rules as barring designs that keep users captive. Metrics built on session length may be reclassified from growth indicators to risk indicators. Emotional data also raises breach costs, because the harm from leaking intimate conversation logs exceeds that of ordinary behavioral data.

9. The author's view: convenience costs do not disappear, they move

Having watched several cycles since the 1990s, I have noticed one pattern repeatedly. When a product promises convenience without cost, the cost does not vanish; it moves to another balance sheet. In the dot-com era it moved to retail investors' accounts, and after 2008 to households and taxpayers. The cost of frictionless companionship is likely to land on users' offline relationships, adolescent development, public care systems and regulators. That this cost is already appearing as statutes and lawsuits is, in my view, the most important observation in this piece.

10. Market Outlook: the most probable scenario and the risks against it

The data point most plausibly toward regulated growth. Under that path, the category keeps expanding, but age separation, usage prompts and crisis protocols become standard features, and monetization shifts from raw engagement toward purpose-built services in care and health. Supporting evidence: three major jurisdictions converged on similar rules within eleven months, observed spending is already substantial, and public care pilots show positive results. This is a directional judgment about industry structure, not a price forecast.

Scenario	What happens	Evidence for it	What to watch
Base (most probable): regulated growth	Safety design standardizes; value shifts to purpose-built services	Rules converging in US, China, EU, Korea; care pilots	Korean implementing rules; age-assurance rollout
Risk A: harm proven, liability widens	Causal studies and litigation expand limits on minors	Reported settlements in principle in Character.AI cases (Jan 2026)	Court outcomes; FTC 6(b) findings
Risk B: regulatory fragmentation	Feature removal and geo-blocking; compliance cost favors the largest players	China's persona shutdowns; US state patchwork	Regional feature gaps at global platforms
Upside: validated as a bridge	Designs that return users to human contact prove out in trials	Short-term loneliness reduction studies	Longitudinal results

The counter-case deserves equal weight. Research findings conflict, long-run data is scarce, market-size estimates vary by more than 2x, and several figures here rest on single secondary sources. Any of the scenarios above could be wrong, and readers should treat the base case as a working hypothesis rather than a conclusion.

11. Basic principles for investors and users

This section is deliberately general and not a recommendation to buy, sell or time any specific position, company or sector.

Diversify across the value chain, not within a single theme. Engines, hardware, care services and compliance respond to different cycles.
Separate narrative from observed revenue. Observed consumer spend is a firmer anchor than market-size forecasts.
Treat regulation as a first-order variable, and examine how much a business depends on session time and on minors.
Keep a long horizon and avoid concentrating in one earnings cycle or policy headline.
For users and families: set daily time limits, share minimal personal data, cap subscription spending and preserve regular human contact.

This article is also published, with a different angle for Korean retail readers, on Leo Jaeyoul Lim's Korean-language sister blog on Naver (blog.naver.com/woduf00). Both blogs are written and operated by the same author.

Search This Blog

Senior Wisdom: Navigating Capital and Power