Considering recent developments, it could be argued that consumer-focused AI is experiencing a revival.
Meta’s personal AI, Muse, along with its charming mascot Jolly, has emerged as an unexpected success. Just yesterday, OpenAI introduced Dots, seemingly pursuing a similar concept of a playful personal assistant. Concurrently, the burgeoning Instinct assistant secured a $10 billion valuation, driven by its impressive agentic capabilities for tasks such as travel bookings, restaurant reservations, or subscription cancellations.
The optimistic argument is straightforward. Agentic AI has achieved sufficient reliability to manage routine tasks effectively. Businesses are progressively offering these services to the general public, who are finding real utility in them. For investors, this scenario strongly echoes the 2022 ChatGPT debut—where AI’s immense potential unlocked an entirely new product sector. It’s an opportunity many would be keen to capitalize on.
However, there’s a clear explanation for why leading AI research labs have become cautious regarding consumer AI, and it’s not due to technological limitations. Even immensely successful tech offerings are encountering limits on what consumers are prepared to spend, and it remains uncertain whether more advanced models are translating into a more lucrative consumer market. This has prompted a sector-wide pivot towards an Anthropic-like strategy, emphasizing enterprise agreements and incremental expansion across specific industries.
Should products such as Muse and Instinct defy this pattern, it’s likely because their primary focus isn’t immediate monetization. Yet, the fundamental economic challenges of consumer AI are not improving, and any new entrant in this sector will inevitably confront them.
These economic realities were underscored in Andreessen Horowitz’s biannual State of Markets report, which sourced its data from a PNC research study published this summer. The report presented two charts illustrating the gradual increase in the proportion of consumers subscribing to AI services, alongside the incrementally rising average expenditure. By May, only 2.2% of consumers were paying for AI, with an average monthly spend of $31.
Andreessen interprets this favorably, suggesting that “it’s still so early when it comes to mature AI adoption and utilization.” They posit considerable potential for expansion. However, the growth rate depicted in both charts appears distinctly linear. Despite significant advancements in AI models, there’s little substantial change in either the volume of customers prepared to pay for AI or their willingness to spend more. For example, the substantial performance leap from GPT-5.2 to Astra is hardly discernible in the data.
While the per-consumer figures are not particularly impressive, they remain significantly beneath the typical breakeven threshold. Using Netflix as a benchmark for mature online services (with 325 million subscribers), a $34 per customer average would only generate $11 billion in annual revenue, which is less than a third of OpenAI’s operational expenses.
Even if you believe PNC’s figures on adoption are conservative, comparable statistics are available from Bank of America. In March, that institution reported approximately 3% of U.S. consumers paying for AI, a 40% increase year-over-year. Meanwhile, a Menlo survey conducted in September presents a somewhat more optimistic outlook, indicating that a quarter of adults engage with AI daily, and half of these users are paying subscribers.
The core challenge with a consumer-centric strategy lies more in operational costs than in revenue generation. AI stands out as an exceptionally expensive technology to maintain, especially when contrasted with less resource-intensive predecessors like social networking platforms or cloud services. Even acquiring hundreds of millions of paying users offers no assurance of reaching profitability.
Commendably, OpenAI appears to have effectively adjusted to these circumstances. The company’s well-publicized shift towards enterprise solutions has largely proven fruitful, with enterprise revenue reportedly doubling since July. Even the introduction of Dots incorporated a significant enterprise focus, demonstrating the new personal agent’s potential utility for software engineers and creative agencies. A long-established strategy for profiting from widely used yet affordable consumer services involves selling them to businesses at a premium, a tactic OpenAI seems to be employing.
The implications for Muse and Instinct are less clear. Muse benefits from the immense power of Meta’s personalized advertising capabilities, affording it additional monetization pathways and extended time before revenue generation becomes a critical concern. Significantly, Meta is already investigating enterprise applications.
Instinct operates with a distinct strategy that includes earning a commission on purchases facilitated by its agent, potentially increasing its revenue potential. It’s also likely to circumvent the considerable expense of training a foundational model, which would be a substantial advantage.
Nevertheless, the challenging economics inherent in consumer AI impose a strict limit on how much a company can realistically scale without leveraging enterprise income. This is a truth already recognized by prominent research labs, and it stands as one of the few constants within the rapidly evolving industry.