A key challenge for agentic AI products is their business model. Unlike chatbots that incur costs per request, agentic systems that run continuously in the background have non-zero marginal costs, making freemium or low-cost models difficult to sustain.

Scott says only about 3% of U.S. households paid for an AI subscription as of February, and OpenAI had 50 million paid subscribers against roughly 1 billion users—about 5%. He argues the larger opportunity, and why Anthropic pulled ahead, is enterprise, making consumer subscription revenue look too thin to support the valuations.

While AI agents will be used personally, their high token costs make the return on investment far greater in enterprise settings. An agent's ability to generate output that directly impacts GDP means business use cases will receive development priority over consumer or personal automation.

Today's high AI inference costs feel prohibitive for consumer apps. However, this is a short-term challenge. The market is heading towards an intersection where dramatically cheaper models meet a consumer base increasingly willing to pay for valuable digital services, enabling sustainable business models.

Unlike traditional software's zero marginal costs, AI-powered apps incur significant inference expenses that scale with users. One founder estimated needing $25M just for 100k monthly actives, challenging the classic VC model for consumer startups.

Initial AI business models based on per-seat subscriptions ($20-$200/mo) could not justify trillion-dollar infrastructure spends. The market's revenue explosion only occurred after shifting to an agentic, usage-based paradigm, where per-person economics can reach thousands of dollars, unlocking a vastly larger Total Addressable Market (TAM).

Concerns about massive AI capex are countered by a powerful bottom-up signal: millions of businesses and consumers are independently choosing to pay for AI services. This widespread, rational economic behavior provides strong evidence of tangible ROI, justifying the large-scale infrastructure investment.

Despite intense hype, only 2% of US households pay for AI, lagging far behind legacy services like SiriusXM (25%). This signals that the primary monetization models will likely be ad-supported or B2B, as most consumers don't yet see sufficient value in paid personal AI tools.

Consumer AI development is slow because investors fear competing with giants like OpenAI. Furthermore, a viable consumer business model for AI has not yet emerged, as subscriptions hit ceilings and inference costs are high. Enterprise offers a clearer, less risky path to monetization.