Entering the 'Agent-Based Monetization' Stage

"Efficiency Matters More Than Model Performance"

Benefits Spreading Across 'AI Infrastructure Bottleneck' Sectors

There is an analysis that the AI industry has shifted from a "competition of intelligence"—focused on which model is smarter—to a "competition of efficiency and monetization," measuring who can complete real work at a lower cost. In other words, investors are now opening their wallets for "cost-effective worker AIs" rather than just smart AI.

"It’s not about who is smarter, but who can do more work at a lower cost"

Competition Paradigm Shifts from "Smarter AI" to "Cost-Effective Worker AI" [Weekend Money] View original image

According to Meritz Securities, the AI market is entering a stage that goes beyond simple performance competition, moving into "monetization"—the stage of generating revenue. In particular, "AI agents" (AI that performs tasks on behalf of users) are considered the key to AI monetization. The explanation is that AI has evolved from simple "conversational AI" into a stage where it actually performs business tasks for companies and creates revenue.


Hwang Suwook, a researcher at Meritz Securities, said, "The focus of the new AI model competition seems to be shifting from comprehensive intelligence competition to who can perform work more efficiently," and added, "For AI to be truly profitable, it is more important for it to do the job better with less inconvenience and at a lower cost for users, rather than just being vaguely intelligent."


In fact, recently released models such as GPT-5.5 have higher per-token costs (costs per unit of data used by AI) compared to previous versions, but the number of tokens required to complete the same task has decreased by 40–70%. As a result, overall cost efficiency has actually improved. This change indicates that the criteria for AI competition are shifting from "performance" to "productivity relative to cost."


Within developer communities (such as Reddit and Hacker News), even if intelligence scores are similar, models are now evaluated based on "task completion capability"—the ability to see complex implementation tasks through to the end—and "controllability," which refers to reducing the need for user intervention.


Researcher Hwang explained, "Anthropic is estimated to have surpassed OpenAI in annualized revenue, with enterprise sales growing rapidly based on AI coding capabilities." He continued, "Based on this, Anthropic is being recognized with a corporate valuation similar to OpenAI, and new investment rounds are being discussed."

"Investment opportunities should be sought in 'bottlenecks,' not 'models'"

This shift is changing the very focus of investment. Currently, profits are appearing first in related infrastructure industries rather than AI service companies.


Researcher Hwang pointed out, "AI investment opportunities do not lie in chasing performance advantages, but rather in the 'bottleneck' segments where demand is surging and supply cannot keep up." He added, "Due to the increasing volume of AI computation, bottlenecks are emerging across memory, networks, and substrates, and the introduction of 800VDC (high-voltage direct current) in the process of next-generation server architecture transition is creating new investment opportunities."

Competition Paradigm Shifts from "Smarter AI" to "Cost-Effective Worker AI" [Weekend Money] View original image

He analyzed, "As AI performs work, not only simple calculations but also functions for dividing and coordinating overall tasks are increasingly necessary, and in this area, the role of the CPU (central processing unit) is growing."



Power supply issues are also a burden. Researcher Hwang reported, "The supply of power facilities needed for data centers is insufficient, causing construction delays, and on-site power generation facilities, which can be built in a shorter period, are being considered as an alternative."


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing