(Yicai) Feb. 24 -- Artificial intelligence's biggest critique has been on identifying its business model for years, and aside from impressive technology, explosive adoption, and massive capital investment, skeptics have been dismissing it as a bubble that is built on hype rather than sustainable income.

However, AI has proven it can be monetized, making the argument obsolete. The real challenge now is not generating revenue, but building a monetization architecture that aligns with the tech's unique economic realities, a structural mismatch between how it is priced and how it actually works.

The USD20-per-month "all-you-can-eat" subscription, once hailed as a genius tool to drive the adoption of AI, lies at the heart of the mismatch. It lowered barriers, built user habits, and scaled quickly, feeling fair compared to streaming services and astonishingly cheap next to the labor it promised to replace. But buffet-style pricing only works when marginal costs are negligible, and usage is consistent, with AI breaking both rules.

Every prompt consumes computing power, with complex tasks like multi-step reasoning or code generation draining graphics processing units exponentially. Usage follows a "fat-tailed" pattern where a small minority of users devour far more infrastructure than the average.

Flat USD20 pricing forces casual users to subsidize professional workloads, squeezing margins and revealing a deeper misclassification: AI was priced like digital media, but it behaves like capital-intensive computation woven into the fabric of economic activity.

For internet platforms, unstable subscription models historically meant a pivot to advertising. But AI cannot follow that playbook due to four defining traits: computing is not a background cost, but the product itself; monetization must be integrated into reasoning, not tacked on, risking trust; AI acts like labor, which is priced by output, not unlimited access; and as AI becomes more agentic, it participates in transactions that require clear attribution without sacrificing privacy.

These traits demand a new economic design, not just tactical price adjustments, but systemic engineering. Four key solutions emerge: hybrid models -- base subscriptions plus usage-based billing -- to align revenue with compute costs; outcome-based pricing -- charging per completed task -- to match labor economics; data licensing to reward owners of scarce proprietary datasets; and vertical ecosystems that embed AI into the legal, finance, and other industries, inheriting existing billing logic and raising switching costs.

However, the most critical piece may be "economic routing," or the ability to turn conversational intent, for example, a user asking about mortgage refinancing, into real-time, privacy-preserving transactions. This is not traditional advertising, but routing embedded in AI's cognition, like partnerships between leading AI developers and performance infrastructure specialists, combining conversational intent with scalable auction and attribution systems.

This routing capability will define AI's future valuation fork, with models lacking routing becoming commodities, like electricity generators competing on cost, while those that integrate routing, having identity systems, payment rails, and attribution, platforms that coordinate markets, capturing far more value. It's a lesson from cloud computing, where raw computing becomes a commodity, but orchestration (application programming interfaces, ecosystems) drives platform-level profits.

AI's early days were defined by the monetization hype, but this year marks the start of the "how to monetize sustainably" era. The USD20 subscription stage was simple because it could afford to be, but AI's next phase requires engineering intelligence to internalize costs, preserve trust, and align pricing with real value. The next competitive frontier will not just be smarter models, but smarter monetization architectures, where routing may matter as much as reasoning.

The author of this article is Lu Duowei, an investor, researcher, and founder of FJ Insights. He previously worked at international financial institutions and sovereign wealth funds, focusing on global structural trends, AI economics, and frontier technology innovation.

Editor: Martin Kadiev