Hidden Concerns Amidst the AI ​​Investment Boom: Return Prospects, Financial Pressure, and Macroeconomic Risks

IBM CEO's Warning: Trillion-Dollar Spending Unlikely to Be Profitable

In early December 2025,IBM CEOIn early December 2025,IBM CEOpublicly questioned the sustainability of AI data center spending. He pointed out that building a 1-gigawatt AI data center costs approximately $80 billion, and if the global plan reaches 100 gigawatts, the total capital expenditure will reach $8 trillion, requiring $800 billion in profits annually just for interest expenses. This is equivalent to several times the current combined profits of tech giants.

Hyperscalers Capex Soars: Investment Scale in 2025-2026

In 2025, the Capex of the four major hyperscalers (Microsoft, Amazon, Google, Meta) is expected to exceed $315 billion, a significant jump from 2024:

• Google's forecast is raised to $91-93 billion

• Meta Microsoft and Amazon combined account for over $100 billion. This has driven AI-related Capex to account for over 1 percentage point of US GDP growth, becoming a major economic driver. However, if returns fall short of expectations, the collapse of this pillar will amplify the impact: the legacy value of infrastructure is high, but the risk of negative short-term cash flow increases. Macroeconomic Divergence: AI Becomes the Last Pillar of Support

  Macroeconomic Divergence: AI Becomes the Last Pillar of Support

  In 2025, the US job market will deteriorate significantly. The ADP report shows that the private sector lost a net 32,000 jobs in November, with small businesses laying off 120,000, the largest drop since 2023. Federal Reserve Chairman Powell acknowledged at the December meeting that risks to the labor market have increased and that official data may overestimate monthly growth.

  Macroeconomic Divergence: AI Becomes the Last Pillar of Support

  In 2025, the US job market will deteriorate significantly. The ADP report shows that the private sector lost a net 32,000 jobs in November, with small businesses laying off 120,000, the largest drop since 2023. Federal Reserve Chairman Powell acknowledged at the December meeting that risks to the labor market have increased and that official data may overestimate monthly growth.

  Historical Mirror and Potential Outcomes

  The AI ​​boom is similar to the dot-com bubble: initial frenzy, later questioning of returns, and eventual collapse but leaving behind value (like the internet). The difference lies in higher concentration (30% of the S&P 500 is supported by a few giants) and heavier leverage (soaring tech debt). If demand doesn't explode by 2026 (due to competition and self-developed chips), the snowballing debt and tightening credit will have a chain reaction.

  The AI ​​boom is similar to the dot-com bubble: initial frenzy, later questioning of returns, and eventual collapse but leaving behind value (like the internet).