Who is paying for the $25 trillion "air"?

IBM CEO issued a chilling warning earlier this month: “All these massive investments in AI and data centers will never pay off.”

If this had been said a year ago, such remarks would be regarded as a blasphemy against “future technology.” But as of December 2025, when Broadcom cannot meet market expectations and Oracle’s financial statements show rare cracks unseen in decades, market sentiment is undergoing a subtle yet deadly reversal.

This is not just a tech stock correction; it could be the collapse of the US economy’s “last fortress.”

01 End of Celebrations: From “Buy-in Faith” to “Auditing Moment”

Over the past two years, the capital markets’ attitude toward AI can be summarized as an almost religious fervor: “Buy first, ask questions later.” Everyone firmly believed this was a revolutionary technology capable of reshaping the world. As for profitability models? That was a concern for 20 years down the line.

However, as the narrative reaches late 2025, the once vague vision is forced to collide with cold financial realities.

The most typical example is Oracle. This company, once regarded as a cash cow, now shows negative free cash flow—an unprecedented scene in decades. To maintain stock prices and high dividends, Oracle has had to take on massive debt.

This is an extremely dangerous signal: tech giants are replacing growth expectations with debt.

Oracle’s CEO tried to reassure the market in a conference call, claiming they are committed to maintaining an investment-grade rating. But these statements only heightened panic: a company that has survived for decades on its own cash flow suddenly emphasizes “maintaining ratings” and “borrowing capacity,” indicating they themselves are aware of tight liquidity.

The market is starting to ask questions once deemed “inappropriate”: What if losses continue next year? If the debt market tightens, who will pay for these expensive data centers?

02 The “Vanke Moment” in Silicon Valley

This scenario evokes a strong sense of déjà vu.

Looking back at the real estate bubble era, today’s AI giants are reenacting the same scene. The current AI industry is very much like Chinese property developers before the bubble burst—such as Vanke.

Back then, the logic was: as long as you acquire enough land and build fast enough, future appreciation would cover current debts. Today’s AI companies follow a similar logic: as long as they stockpile enough GPUs and build large enough data centers, the future “AGI revolution” will erase the massive capital expenditures (Capex) of today.

What Oracle is doing is essentially “borrowing old to buy new,” trying to defer debt for another year, betting on miraculous revenue growth next year.

But the reality is, Broadcom’s backlog of orders, though as high as $73 billion, still falls short of market expectations that have been inflated by appetite. When growth no longer explodes in a parabolic manner, even a slight slowdown can trigger panic selling among investors.

Once free cash flow cannot turn positive, the debt-driven growth model will collapse instantly. This is not a technical problem; it’s basic accounting knowledge.

03 Not only the stock prices will collapse, but the illusion of class wealth

If this were merely a crisis in the tech industry, perhaps it wouldn’t be so unsettling. What’s truly frightening is that the AI bubble has become the only pillar supporting the US macroeconomy.

Currently, the US economy shows a brutal “K-shaped” divergence: the lower and middle classes are experiencing painful layoffs, with the labor market losing tens of thousands of jobs each month. Even Fed Chair Powell has had to withdraw from the phrase “robust labor market.”

So why do consumer data still look relatively stable?

The answer lies in the “wealth effect.” The consumption capacity of America’s wealthy—especially the Baby Boomers—is almost entirely linked to stock market performance. As long as the Nasdaq continues to rise, and Nvidia and Oracle stocks stay high, they dare to spend extravagantly, masking the struggles of ordinary consumers at Walmart and McDonald’s.

The AI bubble is essentially the last line of defense on the asset side of the wealthy’s balance sheet.

Once the market realizes that IBM CEO’s statement is true—that “massive investments will never be recouped”—the valuation system of tech stocks will face revaluation. When the Nasdaq dips like it did in March or April this year, the illusion of wealth among the rich will shatter.

At that point, we will see an economic hard landing with no buffers: the lower classes cannot afford to spend, the wealthy dare not spend, and companies burdened with enormous AI debts will be unable to repay.

04 Naked Swimmers After the Tide Goes Out

The Federal Reserve seems to have also sensed danger.

Although inflation still shows signs of rising on the surface, divisions within the Fed are intensifying. Officials like Austan Goolsbee have begun to hint that they are more concerned about the deterioration of the labor market than inflation.

This is a policy game of “walking on a tightrope”: the Fed must tread carefully to protect the stock market because they know that the stock market (i.e., the AI bubble) is the last means of maintaining the economy’s appearance of prosperity.

But their time is running out.

The current situation is: companies are burning cash to bet on the future, the wealthy are relying on stock prices to sustain consumption, and the Fed is betting on a soft landing before the bubble bursts.

While everyone waits for a turning point in 2026, the real problem could erupt in the coming months: when the first well-known tech company has to sell GPU or data center assets at a discount due to cash flow issues, the dominoes will fall.

Just like during the dot-com bust, people always think “this time is different,” until their stocks turn into worthless paper.

For ordinary investors, now may not be a time for greed but rather a time to think about exiting. Because when the cinema catches fire and there’s only one exit, you can’t outrun the crowd.