The Real Return on AI's Massive Investments： Why Tech Giants Struggle to Show Co

BLUF

The AI industry is experiencing an unprecedented capital expenditure frenzy, but returns are extremely disproportionate. By 2027, the cumulative AI capital expenditure of the four tech giants will exceed $2 trillion, yet revenue is highly concentrated in two companies, OpenAI and Anthropic, and is only a fraction of the spending. If the commercial value of AI technology cannot be proven in the short term, the market will face a severe test of bubble burst.

Why Is 2026 a Critical Turning Point for the AI Bubble?

Answer Capsule: 2026 is the year when the disconnect between AI capital expenditure and revenue is most evident, with the four giants’ combined spending reaching $800 billion, but the revenue growth curve stagnating at a low base. This figure is not only a historic high but also represents a structural contradiction: the larger the investment, the lower the unit return.

Based on financial disclosure data, Microsoft’s annualized AI revenue in 2026 is about $37 billion, and Amazon’s about $15 billion. That sounds impressive, but note: Microsoft’s AI capital expenditure in 2026 is expected to exceed $100 billion, and Amazon’s is close to $300 billion. In other words, Amazon’s AI revenue accounts for only 0.419% of its capital expenditure. This is not investment; it’s gambling.

The Gap Between Capital Expenditure and Revenue

The data gaps at Google and Meta are themselves a warning sign. If AI truly “lights up every business segment” as CEOs claim, why not give clear numbers? The answer is simple: numbers would puncture the myth.

Who Is Actually Making Money from AI? Revenue Concentration Analysis

Answer Capsule: AI industry revenue is highly concentrated in two companies—OpenAI and Anthropic—which contribute 71% and 80% of Microsoft’s and Amazon’s AI revenue, respectively. This means the entire industry’s prosperity is actually just subscription and API revenue from two companies.

This concentration is extremely fragile. If OpenAI or Anthropic hits a growth bottleneck, the entire supply chain will be impacted. More concerning is that a large portion of this revenue comes from “circular spending”—AI companies buying computing power from cloud providers, which then return part of the revenue to AI companies. This is not a healthy ecosystem; it’s capital idling.

AI Revenue Concentration Comparison

This structure reminds me of the dot-com bubble in 2000: capital chasing concepts, not real profits. The excuse then was “the new economy needs time,” and now it’s “AI is still early.” But history shows that the key to a bubble burst is not time, but the moment investors’ patience runs out.

Why Do Meta and Google’s AI Stories Fail Scrutiny?

Answer Capsule: Meta and Google replace concrete revenue with vague metrics, such as a 3.5% increase in ad click-through rate, but never disclose how these metrics translate into revenue. This tactic works repeatedly in earnings calls but fails under deep analysis.

Take Meta as an example. In Q2 2025, its generative ad model (GEM) claimed a 5% increase in Instagram ad conversion rate and 3% on Facebook. By Q1 2026, Meta claimed that after doubling computing power, ad click-through rate increased by 3.5% and Instagram conversion rate by 1%. The two sets of data are not only incomparable but also contradictory—doubling computing power yields diminishing returns? This is not AI’s victory; it’s an efficiency disaster.

Google’s situation is even more absurd. CEO Sundar Pichai constantly emphasizes “Gemini Enterprise has 40% quarter-over-quarter growth” but never provides the base. 40% growth from $1 million to $1.4 million is vastly different from $10 billion to $14 billion. This marketing rhetoric is, frankly, an insult to the intelligence of analysts and journalists.

graph TD
    A[AI Capital Expenditure] --> B{Revenue Return}
    B --> C[Meta: Vague Metrics]
    B --> D[Google: No Specific Numbers]
    B --> E[Microsoft: Concentrated on OpenAI]
    B --> F[Amazon: Concentrated on Anthropic]
    C --> G[Ad Click-Through Rate Up 3.5%]
    D --> H[Gemini 40% QoQ Growth but No Base]
    E --> I[$37B Revenue]
    F --> J[$15B Revenue]
    G --> K[Cannot Link to Revenue]
    H --> K
    I --> L[CapEx $100B]
    J --> M[CapEx $298B]

---

Supply Chain Risks Behind $2 Trillion Capital Expenditure

Answer Capsule: By 2027, the four giants’ cumulative AI capital expenditure will exceed $2 trillion, but the hardware supply chain is already showing signs of inventory oversupply and slowing demand. NVIDIA’s GPU order lead time has shortened from 12 months in 2024 to 3 months in 2026, which is the most obvious warning sign.

When capital expenditure and revenue decouple, the first to be hit are hardware suppliers. Chip makers like NVIDIA and AMD will face dual pressure from order cancellations and inventory adjustments. Worse, cloud providers may cut prices to fill idle computing power, further squeezing profit margins.

Capital Expenditure Timeline and Risks

This is not alarmism. If no scalable killer application emerges by 2027, investors will begin to question the rationality of the AI industry, leading to a funding chain break. At that point, not only tech giants but also the entire startup ecosystem and job market will be affected.

What Happens Next? Three Possible Scenarios

Answer Capsule: The most likely scenario is a “hard landing”—a sharp reduction in capital expenditure in 2027, triggering industry restructuring; followed by a “soft landing” where AI applications finally take off; the least likely is “continued狂欢” because capital efficiency is already diminishing. I personally lean toward a hard landing because human nature never changes.

Scenario 1: Hard Landing (55% Probability)

Investor patience runs out, demanding tech giants prove AI investment returns. Capital expenditure drops sharply by 30-50% in 2027, causing suppliers like NVIDIA to miss earnings. OpenAI and Anthropic, losing cloud subsidies, are forced to raise prices, leading to customer churn. The industry enters a winter, with many AI startups shutting down.

Scenario 2: Soft Landing (30% Probability)

In the second half of 2026, some AI application (e.g., autonomous driving or medical diagnosis) achieves scalable profitability, reigniting market confidence. Capital expenditure continues to grow but at a slower pace, and the industry gradually moves toward healthy development. However, this scenario requires a miracle, not strategy.

Scenario 3: Continued狂欢 (15% Probability)

Investors continue to buy the AI story, and capital expenditure keeps rising until 2028. But this requires even larger capital injections, and only if no systemic risks (e.g., economic recession or geopolitical conflict) emerge. I consider this the least realistic assumption.

timeline
    title AI Bubble Trajectory
    2024-2025: Capital Expenditure Frenzy
    2026: Revenue and Spending Decouple
    2027: Hard Landing or Soft Landing
    2028: Industry Restructuring or Recovery

---

Conclusion: The Truth of AI Lies in the Numbers

Tech giants are reluctant to disclose AI revenue because numbers speak for themselves. When you spend $298 billion and generate only $15 billion in revenue, no marketing rhetoric can hide this fact. AI is not without value, but the current business model cannot support such massive capital expenditure.

In the next two years, we will witness an industry reshuffle. Those that survive will not be the loudest companies, but those that truly understand the meaning of “return on investment.”

FAQ

Why are tech giants reluctant to disclose specific AI revenue figures?

Because AI revenue is far lower than capital expenditure, and growth is highly dependent on a few clients; disclosing numbers would highlight the disproportionate return on investment.

When might the AI capital expenditure bubble burst?

If no scalable killer application emerges by around 2027, investor confidence could collapse, leading to a funding chain break and industry restructuring.

How significant is the contribution of OpenAI and Anthropic to tech giants’ AI revenue?

OpenAI contributes over 71% of Microsoft’s $37 billion annualized AI revenue, and Anthropic accounts for 80% of Amazon’s $15 billion, showing high revenue concentration.

Why is it difficult to quantify AI investments at Meta and Google?

Both companies use vague metrics like ad click-through rate improvements instead of concrete revenue figures, failing to link capital expenditure to actual profits, lacking transparency.

How will the AI industry develop next?

Small and medium AI companies will shut down due to funding tightening, giants will shift to vertical application integration, and hardware and cloud service providers will face inventory adjustment pressure.

Further Reading

• NVIDIA Investor Relations - Financial Reports
• Microsoft 2026 Q1 Earnings Release
• Amazon AI Services Overview
• Meta AI Research Publications
• Google AI Blog