JIAHON Research Report
The bubble is real, but that does not mean a total collapse. Transparency and real-time trading give the market a faster ability to correct itself, so a 2000-style “big burst” is far less likely than a long, noisy “slow deflation.”
Published jointly by JIAHON and NeoTech | January 2, 2026 | jiahon.com.tw
Executive summary
The AI investment frenzy of 2025 has produced a large market bubble. This report asks the central question: how will it end — with a 2000-style crash, or something new?
We propose three analytical models of our own. The Valuation Heatmap Model shows that overheating is concentrated in the application and narrative layer (generative AI startups valued at 50x revenue or more), while the infrastructure layer (the GPU supply chain) retains solid fundamentals. The Transparency Cushion Model argues that real-time information flow, algorithmic trading and widely available data give today’s market a strong capacity for continuous self-correction, replacing the traditional “single-point collapse” with ongoing “pressure release.” The Three-Tier GPU Model maps the physical bottlenecks (HBM, advanced packaging) and bullwhip effects in the AI compute supply chain, showing how hardware cycles add volatility but also act as a natural barrier against runaway overcapacity.
Our core conclusion: a 2000-style “big bang” is the less likely outcome. The 2025 AI bubble is more likely to go through a turbulent, multi-year “slow deflation” — a noisy process of startup shakeouts, sector rotation and capital rebalancing. For midstream supply chain companies, this is both a risk management challenge and a strategic opportunity to capture resilient demand that holds up across cycles.
The AI bubble is real. But the way it bursts will differ from any script history has taught us.
Global AI-related investment rose sharply in 2025, concentrated mainly in the United States and China. Capital poured into three pillars: compute infrastructure (GPUs and the semiconductor supply chain), cloud platform services (the huge capital spending of Google, AWS and Azure), and generative AI applications and the startup ecosystem. The market has every ingredient for a large bubble: striking growth, massive capital and a grand narrative.
This report builds three analytical models of its own, looking from three angles — price, market behavior and physical structure — that point to one core conclusion:
The current AI bubble is more likely a “chronic illness” with violent swings than a sudden “heart attack.” A 2000-style crash is less likely; the market is more likely to enter a multi-year “slow deflation” that is turbulent but not destructive.
Four key points:
- The style of the bubble has changed. Very high information transparency and algorithmic trading give the market a “high-frequency self-correction” ability, so risk is released through turbulence lasting quarters or even years.
- Risk is very unevenly distributed. The heat is concentrated in the “software narrative” layer (generative AI startups with unclear business models), while the “physical hardware” layer (GPU manufacturing, consumables and equipment) has more solid fundamentals.
- The supply chain is both an amplifier and a stabilizer. Bottlenecks such as HBM and advanced packaging amplify short-term swings, but the long-term capital and technology barriers also form a natural barrier against overcapacity.
- Regional strategies differ sharply. The United States (capital-driven), China (policy-driven) and Europe (regulation-driven) will play out different bubble scripts, lowering the risk of a simultaneous, resonant collapse.
1. Three analytical models
Understanding the AI bubble takes systematic tools. At the core of this report are three complementary models that together form a framework for diagnosing and forecasting the style of the bubble.
1.1 Valuation Heatmap Model — the diagnostic
Using price-to-sales, price-to-R&D and forward revenue multiples, calibrated by industry with discounted cash flow models, we draw a “heat map” of the AI market:
Figure 1: Valuation layers of the AI bubble — the “heat” is concentrated in the software narrative layer, while the physical hardware layer is relatively stable.
1.2 Transparency Cushion Model — explaining the dynamics
Traditional bubbles grow in information asymmetry; bad news stays hidden until an unsustainable total collapse. But the 2025 AI market sits in a new environment: earnings are published in real time, social media amplifies any signal around the world instantly, and algorithmic trading acts before human emotion.
The model’s core assumption is that when the speed of market information outpaces the speed of bubble inflation, the bubble loses room to “grow in secret.” Any excessive valuation gap is quickly spotted and questioned, then corrected locally and continuously through short selling, rotation or divestment. Transparency does not remove the bubble; it changes how it bursts. It works like a constantly running “pressure relief valve,” turning what could build into a destructive explosion into small releases of energy that the market absorbs step by step.
Figure 2: Transparency Cushion Model — traditional bubbles go from “hidden inflation → sudden collapse,” while the AI bubble goes from “continuous correction → gradual deflation.”
1.3 Three-Tier GPU Model — a structural telescope
The AI bubble is tightly tied to a global semiconductor supply chain with physical bottlenecks and bullwhip effects. That is the fundamental difference from a purely financial bubble.
Figure 3: Three-Tier GPU Model — the supply chain is both a volatility amplifier and a firewall against collapse.
2. Why this time is different: comparing historical bubbles
“History doesn’t repeat itself, but it often rhymes.”
Table 1: Comparing three bubbles on core dimensions
| Dimension | 2000 dot-com bubble | 2008 subprime crisis | 2025 AI bubble |
|---|---|---|---|
| Core driver | Internet adoption narrative + loose money | Financial engineering (securitization) + excessive leverage | Generative AI technology advances + massive capital |
| Valuation pattern | Non-financial metrics such as “page views” and “user numbers” | Complex financial products masking real risk | Layered: the application layer resembles 2000, while infrastructure has real orders |
| Burst trigger | Profits failed to materialize, tech overinvestment | Rising defaults on underlying assets, liquidity drying up | (Potential) technical bottlenecks, slower-than-expected commercialization, rising cost of capital |
| Market structure | Slow information flow, retail-driven, little real-time transparency | Extremely opaque information, risk hidden off balance sheet | Very high transparency + algorithmic trading, reactions in milliseconds |
| How the bubble builds | Inflates without resistance in a sealed space | Layered packaging in regulatory blind spots | Rises intermittently in a “transparent pressure cooker,” with frequent local corrections |
The key difference is the transparency gap. The previous two bubbles grew in opacity until they collapsed. This bubble is growing under the spotlight, where every flaw can be priced in real time. That makes it possible for continuous, local valuation corrections to replace an overall, concentrated destruction of value.
3. Industry view: exposure and maturity
3.1 AI exposure by industry
The impact of the AI bubble will not fall evenly. We assess it on three dimensions: technology dependence, investment intensity and revenue linkage.
- High exposure (technology, communication services): core products are built directly on AI compute and algorithms and will take the full impact of capital market swings.
- Medium-high exposure (finance, professional services): AI is deeply embedded in operations as a core efficiency tool, and valuations partly reflect expected efficiency gains.
- Medium-low exposure (healthcare, industrial manufacturing): AI is at an early stage of adoption, and these industries are more exposed to indirect effects of a burst (such as tighter financing).
- Low exposure (utilities, consumer staples): results and valuations are barely affected directly by AI market sentiment, forming a “risk buffer layer.”
This unevenness creates a risk buffer: when the highly exposed technology sector adjusts sharply, the relative stability of low-exposure industries helps absorb the economic shock and lowers the chance of a systemic crash.
3.2 Bubble differences between mature and emerging industries
Table 2: Bubble characteristics of mature and emerging AI industries
| Dimension | Mature industries (semiconductors, cloud infrastructure) | Emerging industries (generative AI applications, AI biotech) |
|---|---|---|
| Business model | Proven; leading companies have stable profits | Exploratory; little or no revenue |
| How the bubble shows | Cyclical boom; valuations rise with revenue and profit | Based entirely on narratives of future potential |
| Core risk | Judging the “cycle top” — whether today’s high profits can last | Fundamental “survival risk” — does the technology work? Does the market exist? |
| What a burst means | A deep cyclical downturn as profits and valuations revert to the mean | Large-scale company failures and destruction of capital |
The resilience of mature industries (such as orders at leading foundries) may partly cushion the shock from a collapse in emerging industries. Both scenarios may happen at the same time, but their causes and effects are very different.
4. The core weak point: expected demand vs. real-world adoption
At the heart of a bubble is a gap between expectations and reality. The main tension in today’s AI market is the gap between demand expectations, amplified without limit by capital narratives, and the reality of technical and commercial adoption.