How much AI can earn depends on how much it can take from the human wage pool.

How much money can AI large model companies actually make? A recent report from Guolin Securities provides a disruptive answer: Stop staring at the software market—look at people's paychecks instead.

Companies aren't buying AI to follow trends; they're doing it to save money. Replacing some human labor with AI, boosting efficiency, and cutting costs—that's the real reason businesses are willing to pay. So the true ceiling for AI revenue isn't the size of the software market—it's the size of the wage pool that AI can reprice. Guolin Securities calls this the "wage pool that can be repriced by AI."

A recent research report by Guolin Securities crunched the numbers: Of the approximately $10.83 trillion in annual total wages in the U.S., $1.45 trillion is already exposed to the impact of AI—meaning the job content in these positions can be done by AI, or at least heavily assisted by it.

So how much of that money have AI companies captured? Take the leading company Anthropic, with an annualized revenue of around $47 billion—that's only 3.2% of the $1.45 trillion. Put another way, what they've taken is barely a rounding error.

The Wage Pool, Not the Software Market, Is the Valuation Anchor for ARR

The Guolin Securities report points out that the most intuitive way to understand the "epic growth ceiling" of AI revenue in this cycle is to calculate exactly how large the "wage pool that can be repriced by AI" really is.

The report matches the AI exposure of different occupations to 830 job categories from the Bureau of Labor Statistics (BLS) 2025 Occupational Employment and Wage Statistics (OEWS 2025) for estimation. The results show that among the approximately $10.83 trillion in total U.S. wage income, using Anthropic's actual observed exposure metric, about $1.45 trillion in wage costs are already within the scope of AI exposure—accounting for 13.4%. Under the OpenAI/Eloundou theoretical exposure metric, the potential impact range could be about $5.68 trillion—or more than 52%.

In terms of employment numbers, among the approximately 156 million employed people in the U.S., the actual number exposed is about 18.35 million, accounting for 11.8%; the theoretical number exposed is as high as about 68.3 million, accounting for 43.9%.

The report emphasizes that the $1.45 trillion in wage costs should be understood as the "ideal upper limit for ARR revenue under current penetration rates and technical capabilities," and this ceiling faces a discount—companies may need only $10k in AI spending to equivalently replace $100k in labor costs. Even so, the current ARR of major large model companies, in the tens of billions of dollars, still represents extremely low penetration relative to the size of the aforementioned wage pool.

AI Impact Tends to Favor High-Wage Positions, Knowledge-Based Roles Hit First

Unlike past automation, which mainly impacted manufacturing and repetitive manual labor, this round of AI more directly touches high-wage, knowledge-intensive, and service-sector positions.

The report's data shows that the theoretical exposure of occupations to AI technology has a clear right skew relative to average annual salary distribution—high-income groups face significantly higher AI exposure than middle- and low-income groups. Taking specific occupations as examples, the lowest income quantile groups (such as laundry workers, bakers, tire technicians) generally have low AI exposure; while among high-income groups, financial product managers (income quantile 96.6%, exposure 78.6%), HR managers (income quantile 95.3%, exposure 76%), and aerospace engineers (income quantile 92.5%, exposure 89.3%) all face high replacement risk.

By industry, the three sectors with the highest theoretical exposure are, in order: Computer and Mathematical (87.6%), Business and Financial (78.2%), and Legal (78.0%). However, the observed actual exposure rankings differ from the theoretical ones. The industries with the highest actual exposure are Computer and Mathematical (35.3%), Office and Administrative Support (33.2%), and Sales and Related (24.6%).

This gap reveals that AI's replacement of labor is not determined solely by model capabilities but is also constrained by job attributes, accountability, and organizational processes. The legal industry involves stakeholder coordination, litigation strategy judgment, and lifelong liability; financial services rely on client relationships and non-standardized information judgment. In contrast, programming roles, with their clear work objects and short feedback loops, are seeing faster actual replacement.

The Computer Industry: "Equal Treatment for All"; The Financial Industry: Significant Divergence

Among the top 20 occupations with the highest actual exposure, 8 belong to the Computer and Mathematical category, involving about 1.59 million employed people—30.2% of that industry's total workforce. The report points out that within the computer industry, there is no necessary correlation between salary level and AI exposure—facing the AI shock, the entire industry is treated "fairly alike," highlighting the industry's overall vulnerability under technological iteration.

The financial industry presents a starkly different landscape of divergence. Because some positions require accountability (e.g., auditing, accounting) and the standardization of work output varies across roles, the financial industry's overall actual exposure is relatively low, but internal divergence is pronounced. Among them, Market Research Analysts have an actual exposure of 64.8%, and Financial and Investment Analysts have 57.2%, facing significant replacement risk. Meanwhile, other roles requiring client relationship maintenance and non-standard judgment have relatively lower exposure.

In terms of total exposed wages, the $1.45 trillion in actual exposed wages is largely concentrated in five industries: Office and Administrative Support ($289.6 billion), Business and Financial Operations ($247.4 billion), Management ($221.7 billion), Computer and Mathematical ($215.2 billion), and Sales and Related ($199.5 billion). The report argues that this provides directional guidance for the development of specialized large model to-B business: those seeking certainty can go deep into industries like administration, computer, and finance where significant replacement has already appeared; those aiming for "zero-to-one business breakthroughs" see potential in industries like education and medical diagnostics.

Replacement Does Not Equal Unemployment, But Wage Restructuring Is Underway

The report clearly distinguishes between "exposure" and "replacement": exposure means tasks may be assisted, automated, or reorganized by AI, but it does not mean these wage incomes will disappear proportionally. What truly determines the economic impact of AI remains the pace of enterprise adoption, model capability boundaries, organizational process transformation, and regulatory constraints.

However, the report also points out that the macroeconomic impact of AI will not simply manifest as a linear decline in employment numbers. A more likely path is: some single-responsibility positions are replaced, while many multi-responsibility positions are restructured; some wage costs are compressed, and more labor processes are repriced. Notably, AI Agents have the characteristic of "higher wages leading to higher replacement rates," which could make AI's potential impact on income and consumption more far-reaching.

For investors, the report's core conclusion is: the medium-term ceiling for AI revenue should not be understood solely from the size of the software market, but should find its valuation anchor in the larger labor cost pool. The current ARR penetration of large model companies is still at extremely low levels, but the other side of this coin is that the human wage structure is undergoing a systemic restructuring that has yet to be fully priced in.

Risk Warning and Disclaimer

        Market risk exists, and investment requires caution. This article does not constitute individual investment advice and does not consider the specific investment objectives, financial situation, or needs of any particular user. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific situation. Investment based on this article is at your own risk.