Everyone is calling for a rise, but no one talks about the biggest social risk brought by AI development.

Author: Old Finance Master of Politics and Economics; Source: X, @PolEcoGuru

Along with the nationwide frenzy over AI investment, everyone thinks that simply buying AI-related stocks is the key to wealth freedom, but why has Buffett not stepped in yet, and Peter Thiel cleared all his holdings at the end of last year? Because they see AI’s capabilities and also the macro risks AI might bring to human society:

AI is not simply replacing “workers,” but replacing the skill premiums of the middle class.

Over the past 40 years, the middle class has survived not through physical strength, but through several scarce skills: reading and writing, calculation, information organization, report writing, coding, analysis, process management, client communication, and implementing company policies. The problem is, generative AI is first consuming exactly these “white-collar middle-class skills.”

IMF estimates that AI will impact about 40% of jobs worldwide, around 60% in developed economies; some people will be augmented, while others will face job displacement, wage compression, or reduced hiring. WEF also mentions that 40% of employers expect to reduce staff where AI can automate tasks.

The real danger lies in the fact that AI is not replacing physical labor like during the Industrial Revolution, but first attacking the “entry-level” and “middle-tier” office jobs. A 2026 WEF article states that entry-level positions in the U.S. have decreased by 35% over the past 18 months, with many basic clerical, data entry, customer service, junior coding, and report preparation tasks being performed by AI.

This will create a troublesome gap: in the past, companies trained people through entry-level positions, young workers started with low-value jobs, and after a few years, grew into managers, analysts, lawyers, engineers, or investment researchers. But if AI directly eats these foundational jobs, companies will see short-term efficiency gains, but in the long run, the pathway for the middle class to advance will be cut off.

Therefore, the macro paradox of AI is: The more individual companies use AI, the higher their profit margins; the more society as a whole uses AI, the more middle-class income collapses.

From a corporate perspective, this makes sense. A department that once had 20 people can now be operated by 6 people plus AI tools, increasing profit margins, boosting stock prices, and earning rewards from capital markets. But on a macro level, wages, consumption capacity, and social stability are being squeezed. The middle class is not just an economic ornament but the main buyers of mortgages, cars, education, insurance, travel, dining, electronics, and retirement finance.

If AI suppresses middle-class wages, companies will face a backlash: They save on wages but lose their ultimate consumers.

This is why the judgment that “increased AI use might accelerate economic collapse” makes sense. Not because AI itself is bad, but because the benefits will be heavily skewed toward capital, platforms, computing power companies, and a few high-skilled individuals, while costs are borne by ordinary white-collar workers, small- and medium-sized city service industries, and young graduates. McKinsey also admits that productivity gains from generative AI depend on worker reallocation and skill transformation; if automation occurs without re-employment, the dividends won’t naturally translate into universal income.

Eventually, a “K-shaped society” will emerge:

The top layer consists of those who control capital, data, platforms, computing power, and AI tools, with incomes continuing to rise; the bottom layer includes those unable to gain skill premiums, with stagnant or even declining wages; and the middle white-collar layer is gradually squeezed out.

This is not a short-term stock market issue but an economic structural problem.

In one sentence: The real danger of AI is not unemployment, but that many people will “still have jobs but no longer have bargaining power as the middle class.”

Without tax redistribution, educational restructuring, shorter working hours, universal basic income, or asset distribution mechanisms, the faster AI spreads, the more the demand side of society may be hollowed out. At that point, companies’ efficiency on paper will improve, but macroeconomic resilience will weaken.

In the short term, AI-beneficiary stocks rise because the capital market is buying into a straightforward logic: Cloud providers, model companies, and corporate clients are pouring into AI, so chips, servers, memory, electricity, and data centers are the first to profit.

The most benefited at this stage are Nvidia, Broadcom, Micron, SK Hynix, TSMC, Dell, Vertiv, optical modules, power equipment, and data center REITs—“shovel sellers.” Currently, major tech companies’ AI capital expenditure has reached historic levels; Reuters reports that the four largest tech firms’ AI investments in 2026 could approach $600 billion, and MarketWatch mentions that leading hyperscalers’ capital spending this year might reach $755 billion.

But the long-term issue is: If AI truly compresses middle-class income, demand will weaken, and AI stocks cannot forever rely solely on “cost savings for companies” to rise.

Because corporate profits ultimately come from two sources: one is cost reduction, and the other is selling more. In the early stages, AI boosts profit margins by cutting costs, layoffs, and efficiency improvements, which naturally drives stock prices up; but if all companies do this, middle-class income declines, consumption weakens, and this will eventually backfire on demand industries like advertising, e-commerce, software subscriptions, automobiles, finance, travel, and education.

Therefore, long-term AI stocks will fall into three categories.

The first, potentially long-term winners, are infrastructure monopolies. These are companies that control GPUs, AI chips, cloud platforms, networks, EDA tools, semiconductor equipment, and data center power. They are like “railroads, power grids, and oil companies in the AI era,” charging as long as everyone is building AI. But the risk is that valuations are too high; if capital expenditure slows, stock prices will crash first.

The second, mid-term winners but long-term uncertain, are AI application companies. For example, software, customer service, design, office tools, coding tools. If they can turn AI into a real pricing power, they will do well long-term; but if AI commoditizes software capabilities, many SaaS companies may see their valuations compressed.

The third, most dangerous, are companies relying solely on AI concepts without a profit cycle. These stocks might surge most in the short term but are likely to resemble many internet companies during the 2000 bubble—initially hyped by capital, then crushed by cash flow realities.

The key to long-term success is: Can AI create enough new demand to offset the middle-class income it destroys?

Industry development must first consider politics, and this applies not only to China.

If AI leads to new industries, new drugs, increased energy efficiency, robotic productivity, widespread education and healthcare, and accelerates scientific discovery, then long-term AI leaders will remain strong; but if AI mainly becomes a “company layoffs tool,” it will be a typical scenario of short-term profit margin increases coupled with long-term decline in societal purchasing power, ultimately leading to a bubble burst.

My view is:

AI-beneficiary stocks may remain strong over the next six months, but over three years, they won’t rise across the board and may even lead the stock market toward collapse, leaving only a few companies controlling infrastructure and cash flow to survive and re-emerge as oligopolies.

In one sentence: Buying AI in the short term is buying into the capital expenditure cycle; long-term, it’s about buying monopolistic cash flow.