THE AI BUBBLE: WHEN LIMITS OF ECONOMIC COGNITION STEER CAPITAL FLOWS THROUGH GREED

 *

THE AI BUBBLE:

WHEN LIMITS OF ECONOMIC COGNITION

STEER CAPITAL FLOWS THROUGH GREED**

Author: Lê Hải

Date of completion: December 16, 2025

SUMMARY OF THE ARGUMENT

The explosive development of artificial intelligence (AI) models in recent years has triggered an unprecedented wave of investment, particularly among small and medium-sized investors. However, when the majority of this capital flows into companies that do not possess core technological foundations, do not control computing infrastructure, or lack strategic data ownership, the very speed and intensity of investment significantly increase the probability of a structural AI bubble.

This article argues that the AI bubble does not stem from a lack of intelligence among investors, but rather from large-scale limitations in economic cognition, which make individuals vulnerable to herd behavior and greed disguised in the language of technology.

I. AI IS NOT JUST TECHNOLOGY — AI IS A SYSTEM OF INFRASTRUCTURAL POWER

A common misconception today is to view AI as a form of intelligent software that can scale rapidly and cheaply, similar to traditional SaaS models. In reality, modern AI is a complex techno-economic system comprising:

High-intensity computing infrastructure (GPUs, electricity, cooling systems)

Exclusive training data

Inference costs that do not decrease linearly

Deep dependence on technological and policy power centers

Industry research from 2024–2025 indicates that:

Training a high-end large language model (LLM) can cost USD 80–120 million per full training cycle.

Average inference costs range from USD 0.3–1 per 1,000 tokens, placing heavy pressure on profit margins as user scale increases.

👉 This demonstrates that AI exhibits the characteristics of digital infrastructure industries, rather than “lightweight” technologies.

II. SURFACE KNOWLEDGE AND THE ILLUSION OF “UNDERSTANDING AI”

The growing accessibility of AI knowledge among small and medium investors is a positive signal—if it is embedded within a proper analytical framework. Risk arises when:

Knowledge remains at the level of terminology and narrative

It is disconnected from analysis of the technological value chain

Investors fail to distinguish between core AI and AI wrappers

Data from the startup ecosystem shows that:

70–80% of AI startups at the seed–Series A stage neither train their own models nor own exclusive data, relying entirely on APIs from foundational models.

From a cognitive psychology perspective, this reflects a classic case of overconfidence bias: the more information individuals encounter, the more they believe they understand deeply—while in reality, they grasp only the surface of a highly complex system.

III. HERD EFFECTS THROUGH THE LENS OF ECONOMIC SOCIOLOGY

Financial markets operate not only on data, but also on collective belief. When AI is framed as an “irreversible future,” it generates:

Social consensus pressure

Fear of missing out (FOMO)

Investment behavior that is ritualistic rather than analytical

Behavioral studies indicate that:

Individual investors overestimate their understanding of AI by approximately 35–40%.

Exposure to positive technology narratives more than doubles the likelihood of investment, while financial model scrutiny declines by nearly 30%.

👉 At this point, investors are no longer “making decisions,” but participating in a socially legitimized collective behavior.

IV. LIMITS OF LARGE-SCALE ECONOMIC COGNITION

Most small and medium investors operate within a micro-level cognitive frame, focusing on:

Individual firms

Specific products

Short-term growth

Meanwhile, AI is a macro-structural phenomenon, governed by:

Infrastructural power

National policy

Long-term technological cycles

This cognitive mismatch causes capital to flow toward entities that can tell compelling stories, rather than those that hold core structural value.

V. GREED AS A GUIDED MECHANISM

From a behavioral psychology perspective, when individuals lack the cognitive capacity to evaluate complex systems, they tend to:

Rely on social signals

Trust narrative leaders

Align personal decisions with the “general trend”

At this stage, greed ceases to be a purely individual motive and becomes an externally activated and guided mechanism.

When vision is constrained, individuals are forced to borrow someone else’s compass.

VI. AI CAPITAL FLOWS: CONCENTRATION OF POWER, NOT DEMOCRATIZATION

Global AI investment allocation data shows:

65–70% of total AI capital flows into infrastructure, foundational models, and large technology corporations

Less than 15% goes to independent AI application startups

Additionally:

AI startup burn rates are 2.3–2.8 times higher than those of SaaS startups

Average runway is only 12–16 months, even after fundraising

👉 This indicates that AI does not democratize value — it concentrates power.

VII. HISTORICAL CYCLE VERIFICATION

Comparison with previous technology cycles:

Cycle

Failure rate after 5 years

Dot-com

~78%

Blockchain

~85%

Metaverse

>90%

AI (estimated)

70–80%

AI possesses real value, but it is not immune to bubbles.

CONCLUSION

AI itself is not a bubble.

However, most current AI enterprises lack the structural capacity to bear real AI.

When investors face limitations in large-scale economic cognition, they become guided by other people’s greed, while still believing their choices are entirely their own.

The AI bubble does not burst because the technology is wrong,

but because humans place excessive expectations on structures that cannot sustain

the power, cost, and complexity of real AI.