jakiro
IBM CEO warns: Trillions of dollars spent will make it difficult to turn a profit
In early December 2025,CEO of IBMArvind Krishna publicly questions the sustainability of AI data center spending.He pointed out that the cost of building a 1 GW AI data center is about 80 billion U.S. dollars. If the global plan reaches 100 GW, the total capital expenditure will reach 8 trillion U.S. dollars, and interest expenses alone will require annual profit support of 800 billion U.S. dollars.This is equivalent to several times the combined profits of the current technology giants.Krishna emphasized that the five-year depreciation cycle of AI chips further amplifies the pressure: “You must use it within five years, otherwise you have to throw it away and reset it.” This statement quickly caused a shock in the market, highlighting the core transition of AI construction from “technical competition” to “economic feasibility” examination.
Krishna’s calculations are not isolated.Several institutions estimate that the total Capex of hyperscalers (such as Microsoft, Amazon, Google, and Meta) will exceed US$315 billion in 2025-2026, of which AI infrastructure accounts for more than 80%.Despite strong demand, delayed returns have become a consensus: an MIT report shows that 95% of corporate GenAI investments have zero return; J.P. Morgan analysis states that achieving a 10% return requires generating $650 billion in annual revenue, which is out of reach under the current uncertainty.
Oracle financial report: The contradiction between demand explosion and negative cash flow
Oracle’s financial report for the second quarter of 2026 (as of November 30, 2025) was released on December 10. Total revenue was US$16.1 billion, a year-on-year increase of 14%; cloud infrastructure revenue was US$4.1 billion, an increase of 68%; remaining performance obligations (RPO) surged 438% to US$523 billion, with the additional US$68 billion mainly coming from contracts with giants such as Meta and NVIDIA.This reflects that the demand for AI is real and bound to the long term.
However, the market reaction was negative: the stock price plummeted by more than 10% after hours.The core reason is that Capex has been significantly raised to US$50 billion (an increase of US$15 billion from September guidance), free cash flow has turned negative by about US$10 billion, and long-term debt is close to US$100 billion.Management emphasized that it is “committed to maintaining an investment-grade credit rating,” but this instead exposes the risk of financial tightening: the company needs to continue to borrow money for expansion, and delays in returns may cause the credit market to tighten.Oracle’s shift from a “cash cow” to debt dependence marks the emergence of transformation bottlenecks for downstream cloud service providers.
Broadcom performance: Strong growth but unable to conceal marginal pressure
Broadcom’s fiscal fourth quarter of 2025 (released on December 11) revenue was US$18 billion, a year-on-year increase of 28%; AI semiconductor revenue increased by 74%.The company guides AI revenue to double to US$8.2 billion in the first quarter of 2026, showing that the order momentum continues unabated.However, the stock price fell by about 11% due to a gross profit margin warning: the increase in the proportion of AI business has led to component cost pressure, and customers’ shift to custom chips may weaken pricing power.
Broadcom maintains positive cash flow, but overall technology stock valuations are close to the peak of the dot-com bubble.Any growth slowdown (such as backlog missing expectations) triggers a sell-off, reflecting the market’s shift from “growth story” to “earnings quality” assessment.
Hyperscalers Capex surges: investment scale in 2025-2026
In 2025, Capex, the four major hyperscalers (Microsoft, Amazon, Google, Meta), is expected to exceed US$315 billion, a significant jump from 2024:
-
Google raises forecast to $91-93 billion
-
Meta $70-72 billion
-
Microsoft and Amazon combined exceed $100 billion
This has pushed AI-related Capex to account for more than 1 percentage point of U.S. GDP growth, becoming the main driver of the economy.However, if the returns are not as good as expected, the collapse of this pillar will amplify the impact: the legacy value of infrastructure is high, but the risk of short-term negative cash flow increases.
Macroeconomic differentiation: AI becomes the last pillar of support
In 2025, the U.S. job market will deteriorate significantly.The ADP report shows that the private sector lost 32,000 net jobs in November, and small businesses laid off 120,000 people, the largest decline since 2023.Fed Chairman Powell acknowledged at the December meeting that labor market risks have risen and official data may overestimate monthly growth.
The December FOMC meeting cut interest rates by 25 basis points to 3.5%-3.75%, but the dot plot only hinted at one interest rate cut in 2026, which is far lower than market expectations.The dissent of three members (one dovish wanted a deeper reduction, two hawks wanted a pause) reflected a split: stubborn inflation vs. a weakening labor force.Powell emphasized that high-income groups (stock market wealth effect) support consumption, but the reversal of AI stocks will sharply reduce spending.The wave of retirements of the baby boom generation has exacerbated the risk: the stock market is sustainable when it rises, but there is no solution when it falls by 30%-50%.
The “K-shaped” differentiation of the U.S. economy is intensifying: consumption at the bottom is weak (McDonald’s and Target are declining), and the high-end relies on the AI stock market.Similar trends around the world: Household spending fell in Japan and retail sales in Europe were sluggish.If AI reverses, Capex slows down and the wealth effect fades, it will hit consumption, which already accounts for most of economic growth in 2025.
Fed Policy Dilemma: Balancing Inflation and Recession Risks
The Fed’s December meeting was more hawkish: it cut interest rates only once in 2026, reflecting concerns about rising inflation (partially affected by tariffs).But the weakening labor market forced “insurance” easing.Powell called the current economy “abnormal”: inflation is above target and employment risks are rising.If the AI bubble bursts, the Fed will have limited space to respond simultaneously.
Historical Mirroring and Potential Endings
The AI craze is similar to the dot-com bubble: initial enthusiasm, later doubts about returns, and eventual collapse but leaving behind value (the Internet).The difference is that there is greater concentration (30% of the S&P 500 is supported by a few giants) and more leverage (tech debt has exploded).If demand does not explode in 2026 (competition, self-developed chips), debt will snowball and credit tightening will have a chain reaction.
The risks go beyond finance: data center power demand accounts for 14% of the world’s electricity demand, and climate pressure is intensifying; in terms of employment, Anthropic CEO predicts that half of entry-level white-collar workers will be eliminated, and the unemployment rate will rise by 10%-20%.Although Yale analysis shows that overall employment has not been disturbed since ChatGPT, hundreds of thousands of technology layoffs have become a reality.
Conclusion: The turning point has arrived, prudence is the priority
At the end of 2025, the AI investment boom reached a turning point: IBM, Oracle, and Broadcom incidents marked the market’s shift from “buy and ask later” to “ask first for returns.”The demand is real, but financial constraints and macro weaknesses amplify risks.If profitability conversion is not achieved in 2026, valuation adjustments will be inevitable; only in the long run can efficient construction unlock the value of transformation.Investors need to be wary: AI may reshape the world, but the process will be accompanied by severe pain.Short-term volatility has intensified, and it is advisable to make prudent allocations in the medium term.
