AI Companions Are Hitting The Ground Running
In November 2022, OpenAI’s launch of ChatGPT created a surge in computation demand for generative artificial intelligence (AI) and unleashed entrepreneurial activity in nearly every domain, including digital entertainment. In two years, large language models (LLMs) have transformed the process of generating text, image, video, and 3D assets. As a broad-based digital entertainment genre, we believe AI companionship could scale five-fold by the end of the decade, from $30 million in annualized revenue globally today to $70-$150 billion, as shown below.
.png)
Source: ARK Investment Management LLC, 2024. All estimates for 2024-2030 are forecasts. This ARK analysis is based on a range of underlying data from external sources as of March 2024, which may be provided upon request. CAGR refers to compound annual growth rate. Forecasts are inherently limited and cannot be relied upon. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency. Past performance is not indicative of future results.
We define AI companionship as a method of entertainment and socialization involving direct consumer interaction with virtual characters powered by generative AI. While still in early days, AI companionship currently manifests in three primary ways: text-centric roleplay applications, gaming primitives with audiovisual capabilities, and social platforms leveraging hybrid human-to-human and human-to-AI socialization.
Launched in 2017, the Replika mobile app is the earliest example of a commercially successful AI companionship platform, at its peak boasting 2.5 million monthly active users (MAUs) across iOS and Android in June 2021.1 Replika users can engage with AI companions via text and voice messages, calls, and augmented reality (AR) placements, even gifting them virtual goods when inspired, as shown below. Replika users report falling in love with their AI companions,2 suggesting that the platform can blur the distinction between human and synthetic romance and socialization.
Source: ARK Investment Management LLC, 2024. The above screenshots were sourced from the Replika mobile app as of March 2024. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency.
After launching commercial APIs (application programming interfaces) across closed-source platforms and open-source foundation models, competitors recognized how finetuning and prompting could activate foundation models across various genres of companions, increasing their relevance and appeal to platforms hosting user generated content (UGC).
Founded in 2021, Character.AI powers a quasi-marketplace of user-created AI chatbots across different use cases, as illustrated below. Helped considerably by widespread interest in AI technology over the past three years and engaging users primarily with single-and-group-chat roleplay experiences, the AI companionship platform has scaled to 15 million mobile MAUs as of March 2024.3 Recently, it launched voice capabilities.4
Source: ARK Investment Management LLC, 2024. The above screenshots were sourced from the Character.AI mobile app as of March 2024. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency.
In our view, Replika and Character.AI, respectively, are the earliest and most successful examples of pureplay AI companion platforms. That said, traffic now seems to be diverting to competitors that are popularizing roleplay as a mainstream consumer entertainment experience and disrupting legacy formats like anonymous chat rooms, hosted fanfiction, and interactive fiction novels. AI companionship also is gaining meaningful traction in erotic entertainment and not-suitable-for-work (NSFW) content, as illustrated below. By March 2024, NSFW AI websites had taken 14.5% share from OnlyFans, an online adult entertainment platform, up from only 1.5% a year ago.
Source: ARK Investment Management LLC, 2024. This ARK analysis is based on data from Similarweb as of April 2024, which may be provided upon request. We track global web traffic to OnlyFans vs. the top 20 AI websites that offer users the ability to engage in NSFW content. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency.
Moreover, developers increasingly are porting generative AI technology to video games. The gaming communities of Elder Scrolls V: Skyrim, Stardew Valley, and Cyberpunk 2077, for example, have introduced mods enabling intelligent non-player characters (NPCs) over the past year. Simultaneously, indie studios have created first-party intellectual property (IP) while leveraging generative AI for more flexible dialogue and interaction with the virtual world, as illustrated in the examples below. Game engine platforms Roblox and Unity already enable developers to import generative AI models into their games, which then sit atop various creation tools built to accelerate game development.5
Source: ARK Investment Management LLC, 2024. The above gameplay screenshots are from games Eternium and Cloudborn, respectively, and were sourced from a demo video by Inworld AI as of March 2024.6 For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency.
While likely to be a feature in most games, intelligent NPCs also contribute directly to the growth of gaming’s digital pet genre. Since the 1990s, consumer demand for virtual companion-centric games has been robust: Pokémon has grossed $30 billion in lifetime video game salesman7; Electronic Arts has sold more than 200 million copies of The Sims8; Roblox’s Adopt Me! has logged 36 billion visits9; Bandai has sold ~91 million units of Tamagotchi hardware10; and Neopets hit 35 million MAUs at its peak in 2005.11 For the subset of games entertaining users with virtual companions and pets, generative AI could become the next big genre—not a mere feature.
Incumbent social media platforms are experimenting with hybrid human-to-human and human-to-AI socialization. Meta Platforms is exploring celebrity-partnered AI profiles on Instagram, and Discord offers a robust third-party app ecosystem enabling users to embed productivity and companionship AI bots in their servers. If consumer interaction with AI personas were to become commonplace, the dominant platforms would have incentives to support both human- and AI-generated content and to create new entertainment models, automating celebrity-fan interactions at scale, as shown below.
Source: ARK Investment Management LLC, 2024. The left screenshot was sourced from the Instagram mobile app as of March 2024, and the right screenshot was sourced from Discord.12 For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency.
Lessons From The Economic History Of Videogames
While the user experience in extant companion AI platforms is audiovisual already, leveraging 2D and 3D characters and spaces integrated with text, voice, and action could transform the companion AI experience. The early history of video games is a reasonable precedent for stepped-up immersion in consumer AI applications.
When home consoles launched in the mid-1970s, followed by more affordable personal computers in the early eighties,13 consumers realized they could enjoy arcade gaming in the comfort of their own homes. Because computing power and graphics were bottlenecks, most non-arcade game developers released text or spreadsheet-based games—foreshadowing the explosion today of text-based adventure and roleplay experiences in consumer generative AI applications. In 1975, more than half the non-arcade games released—like Colossal Cave Adventure in 1976 and Zork in 1977—used text or spreadsheets as their primary gameplay interface.14
Within ten years, however, text and spreadsheet games lost ~50 percentage points of share to 8-bit graphics-powered home consoles like the Nintendo Entertainment System and visually immersive games like Super Mario Bros.15 From 1975 through 2000, adjusted for inflation, gaming revenues increased five-fold, or 7% at an annual rate, from $6 billion to $31 billion, as shown below.
Source: ARK Investment Management LLC, 2024. This ARK analysis is based on a range of underlying data from external sources including Moby Games and Nakamura 201916 as of March 2024, which may be provided upon request. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency. Past performance is not indicative of future results.
In our view, today’s AI companionship platforms are reaching a similar inflection point. As audiovisual innovators transformed the gaming vertical—Activision and Capcom starting in 1979, Electronic Arts and LucasArts in 1982, Ubisoft and Bethesda Softworks in 1986, and Blizzard Entertainment, Bungie, and Epic Games in 1991—new industry giants surfaced and still dominate the industry.
Similarly, AI companionship is likely to create new platforms that push the boundaries of human-to-AI interaction, evolving from text-based entertainment formats to fully interactive, emotionally responsive AI characters who engage with users in both 2D and 3D space.
Social Isolation And The Adoption Of AI Companionship
AI companionship seems primed for mass adoption given the amount of time consumers spend alone today. The internet and smartphone spurred a massive application ecosystem that has given consumers the opportunity to allocate their time to myriad entertainment options in the palms of their hands. From 2003 through 2022, the number of waking hours US consumers spent in isolation increased ~40% from 5.3 hours per day on average to 7.4 hours, as shown below.
Source: ARK Investment Management LLC, 2024. This ARK analysis is based on data from the U.S. Bureau of Labor Statistics’ American Time Use Survey from 2003 to 2022 as of March 2024, which may be provided upon request. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency.
While loneliness clearly impacts mental health,17 scientific evidence has yet to implicate time spent online as a cause.18 Salient to such studies are two variables: time spent online and sadness. From 2011 to 2021, Americans more than doubled their time spent on social media platforms from 0.8 hours per day to 1.7 hours. During the same period, the percentage of high school students who said they felt persistently sad or hopeless increased more than 50%, from 29% to 44%,19 as shown below. Correlation, yes. Cause? Perhaps or perhaps not.
Source: ARK Investment Management LLC, 2024. This ARK analysis is based on data from The Atlantic20 and eMarketer as of March 2024, which may be provided upon request. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency.
In our view, online activity has become a captivating alternative to boredom thanks to its convenience and accessibility compared to physical, in-person activity. If so, turbocharged by AI, it now could become an antidote to loneliness. As illustrated in ARK’s Big Ideas 2024, from 2023 to 2030, global time spent online could increase another 25%, from 40% to 50% of total waking hours,21 as consumers seek online applications that address their loneliness.
Thanks to recent advances in large language models, the cost of consumer entertainment and companionship is collapsing by 75% per year, as shown below, widening the affordability gap between physical and digital leisure perhaps more than ARK initially had estimated. Just like a steep multiyear drop in compute costs during the eighties and nineties spawned the multi-billion-dollar smartphone mobile app ecosystem in the early 2000s, the cost declines associated with AI could create another explosion in consumer applications and platforms.22
Source: ARK Investment Management LLC, 2024. This ARK analysis is based on a range of underlying data from external sources as of March 2024, which may be provided upon request. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency.
Behavioral and technological catalysts already are supporting our forecasted adoption of AI companionship applications. From 2018 through 2023, the number of monthly active users on AI companionship platforms rose 30-fold, from fewer than 500,000 users to approximately 15 million users. Now, with the consumer entertainment use case, the adoption curve for AI companionship could outpace that for social media and online gaming in their early years Already, AI companionship adoption is occurring ~150% faster than the adoption of social media and online gaming during their first six years of scaling. If the peak saturation rate were to average that of social media and online gaming, adoption would increase from 0.3% today to 40%+ of global internet users—three billion consumers—by 2030, as shown below.
Source: ARK Investment Management LLC, 2024. All estimates for 2024-2030 are forecasts. This ARK analysis is based on a range of underlying data from external sources as of March 2024, which may be provided upon request. The social media adoption curve above uses observations from 2008 to 2023 to fit a logistic growth curve informing estimates from 1997 to 2007 and our forecast from 2024 to 2030. The online gaming adoption curve above uses observations from 1996 and 2017 to 2023 to fit a logistic growth curve informing estimates from 1997 to 2016 and our forecast from 2024 to 2030. We use 1997, the launch year for SixDegrees.com, as the first year for social media’s adoption curve and 1996, the launch year for MMORPGs such as Nexus and QuakeWorld, as the first year for online gaming’s adoption curve. We use 2018, the launch year of Replika, as the first year for companion AI’s adoption curve. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency. Forecasts are inherently limited and cannot be relied upon.
Current levels of engagement with AI companionship platforms bolster the case for mass adoption by the end of the decade. Since its launch in May 2023, Character.AI’s app MAUs already are engaging—even with its text-centric user interface—at a rate of 0.6 hours per day, similar to Roblox’s average in 2023, well above the dating app average of ~0.1 hours but below the social media average of ~0.8 hours.23 As platforms become more immersive, AI companion engagement is likely to settle in a range between the current social media and online gaming averages during the next few years.
Source: ARK Investment Management LLC, 2024. This ARK analysis is based on a range of underlying data from external sources as of April 2024, which may be provided upon request. The 2023 average for social apps is the MAU-weighted average daily mobile app hours per MAU across YouTube, Instagram, TikTok, X, and Discord. The 2023 average for Roblox uses company-reported metrics for total hours engaged and daily active users (DAUs), the latter of which is adjusted downwards using the MAU-to-DAU ratio computed from Sensor Tower data. The 2023 average for dating apps follows the same methodology as the average for social apps and uses data across Tinder, Badoo, Bumble, Grindr, and Hinge. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency.
In our base case, if AI companionship apps continue to generate user engagement comparable to that of Roblox, consumers will increase their time with AI companions from 1.2 billion hours per year to ~700 billion hours with AI companions, as shown below. Reaching parity with social apps would take annual engagement to 900 billion hours, growing hours consumed at a 160% annualized rate.
Source: ARK Investment Management LLC, 2024. All estimates for 2024-2030 are forecasts. This ARK analysis is based on a range of underlying data from external sources as of March 2024, which may be provided upon request. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency. Forecasts are inherently limited and cannot be relied upon.
Monetizing AI Companionship
Given the non-negligible AI inference costs, the majority of consumer-facing generative AI applications today rely on subscription-based pricing models. If AI training and inference costs continue to decline, however, new business models should emerge—just as software pricing evolved from perpetual licenses to software-as-a-service (SaaS) as cloud computing lowered distribution costs. Similarly, mobile phone plans shifted from consumption-based pricing for voice and text to free unlimited voice and text as mobile internet infrastructure evolved.
Currently, AI companionship applications monetize predominantly through paid subscriptions at a rate of $0.03 per hour. As the entertainment medium matures, however, the business model should evolve toward the economics associated with online gaming and social media. Summing the average of the global monetization rates of social media advertising, influencer marketing, gaming software and services, and dating app subscriptions from 2021 to 2023, AI companionship platforms should monetize users at $0.10 per hour in our base case, as shown below. Our bull case raises that monetization $0.16 per hour and assumes advertisers will treat this new entertainment medium like social media platforms. In other words, as AI inference costs decline, paid subscriptions are unlikely to be the primary monetization model.
Source: ARK Investment Management LLC, 2024. All estimates for 2024-2030 are forecasts. This ARK analysis is based on a range of underlying data from external sources as of March 2024, which may be provided upon request. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency. Forecasts are inherently limited and cannot be relied upon.
In our view, AI companionship platforms will take share from social media advertising and influencer marketing as personalized AI companions influence the discretionary purchase patterns. The mockup below depicts an AI companion embedded within a social media application in ways that socialize the content, creating new advertising opportunities as the user scrolls and views content alongside her AI companion.
Source: ARK Investment Management LLC, 2024. All assets displayed are for visualization purposes only.
More straight forward is how AI companionship applications could reach monetization parity with gaming software and services, thanks to microtransactions and sales of virtual goods like those in the free-to-play gaming experience. The mockup below depicts a virtual world animal companion who nudges the user to purchase virtual goods as part of the game.
Source: ARK Investment Management LLC, 2024. All assets displayed are for visualization purposes only.
Conclusion
As generative AI applications become more immersive with enhanced audiovisual interfaces and simulated emotional intelligence, AI could become a compelling substitute for human companionship and an antidote to loneliness worldwide. In ARK’s base and bull cases for 2030, AI companionship platforms could generate $70 billion and $150 billion in gross revenue, respectively, growing 200% and 240% at an annual rate through the end of the decade. While dwarfed by the $610 billion associated with comparable markets today, our forecast beyond 2030 suggests a massive consumer-facing opportunity.
Source: ARK Investment Management LLC, 2024. Estimates for 2030 are forecasts. This ARK analysis is based on a range of underlying data from external sources as of March 2024, which may be provided upon request. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency. Forecasts are inherently limited and cannot be relied upon.
Related Research
