Concerns Amid the AI Investment Boom: Return Outlook, Financial Pressure, and Macro Risks

IBM CEO’s Warning: Trillions of Dollars in Spending Difficult to Turn Profitable

In early December 2025, IBM CEO Arvind Krishna publicly questioned the sustainability of AI data center spending. He pointed out that building a 1-gigawatt AI data center costs about $80 billion. If global plans reach 100 gigawatts, total capital expenditure would amount to $8 trillion, with interest expenses alone requiring an annual profit of $800 billion to support. This is several times the combined profits of current tech giants. Krishna emphasized that the 5-year depreciation cycle for AI chips further amplifies pressure: “You must use it up within five years, or you have to discard and reset.” This statement quickly triggered market turbulence, highlighting a pivotal shift from viewing AI construction as a “technology race” to examining its “economic feasibility.”

Krishna’s calculations are not isolated. Multiple institutions estimate that hyperscalers (such as Microsoft, Amazon, Google, Meta) will have a combined Capex exceeding $315 billion in 2025-2026, with over 80% allocated to AI infrastructure. Despite strong demand, delayed returns have become a consensus: a MIT report shows that 95% of enterprise GenAI investments yield zero returns; J.P. Morgan analysis states that achieving a 10% return requires generating $650 billion in annual revenue, which is currently unattainable amid uncertainty.

Oracle Earnings: Demand Explosion vs. Negative Cash Flow Contradiction

Oracle’s fiscal Q2 2026 (ended November 30, 2025) earnings were released on December 10, with total revenue of $16.1 billion, up 14% year-over-year; cloud infrastructure revenue of $4.1 billion, up 68%; remaining performance obligations (RPO) surged 438% to $523 billion, with $68 billion of new contracts mainly from giants like Meta and NVIDIA. This reflects genuine and long-term AI demand.

However, market reaction was negative: shares plummeted over 10% after hours. The core reason was a sharp increase in Capex to $50 billion (up $15 billion from September guidance), with free cash flow turning negative by about $10 billion, and long-term debt approaching $100 billion. Management emphasized “commitment to maintaining investment-grade credit ratings,” but this exposed risks of financial tightening: the company needs to continue borrowing to expand, while delayed returns could tighten credit markets. Oracle is shifting from a “cash cow” to debt reliance, indicating bottlenecks in downstream cloud service transformation.

Broadcom Performance: Strong Growth but Marginal Pressure Difficult to Ignore

Broadcom’s fiscal Q4 2025 (released December 11) revenue was $18 billion, up 28%; AI semiconductor revenue grew 74%. The company guided that Q1 2026 AI revenue will double to $8.2 billion, showing order momentum remains strong. However, the stock fell about 11% due to gross margin warnings: rising AI business share increases component costs, and customers shifting to custom chips may weaken pricing power.

Broadcom maintains positive cash flow, but overall tech stock valuations are near dot-com bubble peaks. Any slowdown in growth (such as backlog not meeting expectations) triggers sell-offs, reflecting a market shift from “growth stories” to “profit quality” assessments.

Hyperscalers Capex Surge: Investment Scale in 2025-2026

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

• Google revised upward to $91-93 billion
• Meta $70-72 billion
• Microsoft and Amazon combined over $100 billion

This drives AI-related Capex to contribute over 1 percentage point to US GDP growth, becoming a major economic driver. But if returns fall short of expectations, the collapse of this pillar could amplify shocks: infrastructure remains valuable in the short term, but the risk of negative cash flow turns increases.

Macro Economic Divergence: AI as the Last Support Pillar

In 2025, the US labor market deteriorated significantly. ADP reports show that in November, private sector net job losses reached 32,000, with small business layoffs hitting 120,000—the largest decline since 2023. Fed Chair Powell acknowledged in December that labor market risks are rising, and official data may overstate 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 rate cut in 2026, well below market expectations. Divergence among three members—one dovish advocating larger cuts, two hawks wanting to pause—reflects split views: persistent inflation vs. weakening labor. Powell emphasized that high-income groups (wealth effects from the stock market) support consumption, but a reversal in AI stocks could sharply reduce spending. The baby boomer retirement wave exacerbates risks: sustainable when stocks rise, but a 30%-50% decline leaves no solution.

The US economy’s “K-shaped” divergence intensifies: weak consumption at the bottom (e.g., McDonald’s, Target decline), high-end reliance on AI stock market. Similar trends globally: Japanese household spending declines, European retail sluggish. If AI reverses, Capex slowdown combined with wealth effect erosion could hit consumption—already accounting for over half of economic growth in 2025.

Fed Policy Dilemma: Balancing Inflation and Recession Risks

The December Fed meeting was more hawkish: only one rate cut in 2026, reflecting concerns over rising inflation (partly due to tariffs). But weakening labor markets force “insurance” easing. Powell described the current economy as “unusual”: inflation above target, employment risks rising. If the AI bubble bursts, the Fed’s room to maneuver will be limited, making it difficult to address both issues simultaneously.

Historical Parallels and Potential Outcomes

The AI boom resembles the dot-com bubble: initial frenzy, later doubts about returns, eventual crash but with residual value (the internet). The difference is higher concentration (30% of S&P 500 supported by a few giants) and increased leverage (tech debt surging). If demand does not explode in 2026 (due to competition, self-developed chips), debt snowballing and credit tightening could trigger chain reactions.

Risks extend beyond finance: data center power demand accounts for 14% of global electricity, climate pressures intensify; employment-wise, Anthropic CEO predicts half of entry-level white-collar jobs will be eliminated, with unemployment rising 10%-20%. While Yale analysis shows overall employment has not been disturbed since ChatGPT, tens of thousands of tech layoffs are already a fact.

Conclusion: Turning Point Has Arrived, Caution Is Wise

By the end of 2025, the AI investment frenzy reaches a turning point: events involving IBM, Oracle, and Broadcom mark a shift from “buy and ask” to “ask for returns first.” Demand is real, but financial tightening and macro weaknesses amplify risks. If profitability is not achieved in 2026, valuation adjustments are inevitable; in the long term, only efficient construction can unlock transformation value. Investors should remain vigilant: AI may reshape the world, but the process will be painful. Short-term volatility will intensify, and medium-term cautious positioning is advisable.