Selling Tokens or Selling Outcomes: Several Paradoxes of the AI Business Model
"The Token vs. Outcome Sale: Key Paradoxes in the AI Business Model
By mid-2026, the AI industry shows rapid growth in revenue and token usage, yet the underlying business models differ significantly. This article analyzes four structural paradoxes defining the current landscape, all pointing to the commoditization of intelligence and the concentration of profits in few segments.
**The Cost Paradox: Cheaper Tokens, Heavier Bills**
Despite a >95% price drop for equivalent AI capability since 2023, total spending has skyrocketed due to the Jevons Paradox: lower prices expand usage into previously uneconomical tasks. Furthermore, the shift to autonomous agents operating 24/7 multiplies consumption. However, efficiency gains often remain unrealized due to unchanged organizational workflows (the Solow Paradox). The focus is shifting from optimizing token price to optimizing the task itself.
**The Hierarchy Paradox: The App is King vs. The App is Dead**
While conventional wisdom holds that value accrues at the application layer, the AI stack is inverted. Infrastructure (chips) captures ~70% of industry revenue and ~80% of gross profit, while application-layer margins are thin (0-30%). Fast-evolving base models threaten "thin" apps. Sustainable applications are those that embed intelligence into specific contexts, possessing private data, workflows, or delivery capabilities that become more valuable as the base model improves.
**The Responsibility Paradox: Profit Follows Accountability**
Growth rates alone don't guarantee profit. A key differentiator is a company's willingness and ability to take responsibility for specific outcomes. Selling by the token competes for IT budgets; selling by the outcome (e.g., a resolved support ticket) taps into larger human labor budgets. Low-responsibility, high-volume tasks (e.g., generic客服) face commoditization. High-stakes, regulated domains (e.g., law, healthcare) where vendors assume heavier liability for results command higher margins, as seen with companies like Harvey in legal tech.
**The Open-Source Paradox: Open Wins Traffic, Closed Wins Revenue**
Open-source models dominate in usage share and developer adoption, often being 5-20x cheaper. However, closed-source models still capture the majority of enterprise spending (~89%). Enterprises pay a premium for closed-source reliability, support, compliance, and accountability. The total cost of ownership (TCO) is converging as closed-source prices fall faster than open-source builds trust, leading to hybrid deployments. Profit is migrating from the model layer itself to upstream (compute) and downstream (orchestration, data, services)."
marsbit18 min fa
