Summary of Key Points
This article focuses on the “bullwhip effect” within the AI industry chain. It begins by explaining the principle of the traditional bullwhip effect (how demand signals in the supply chain are amplified at each stage) and then delves into three specific distortions of this effect in the AI context: vague demand signals, hidden inventory levels, and a shift in psychological motivations from fear of out-of-stock situations to fear of falling behind. The article concludes by pointing out that the most critical link in the chain—represented by “national will”—poses the greatest risk, as it has the potential to amplify even the most ambiguous signals into projects worth hundreds of billions of dollars, with significant impacts on ordinary people. It also emphasizes the need for quantitative tools to determine one’s position within the supply chain.
Detailed Analysis
1. Understanding the Traditional Bullwhip Effect
The bullwhip effect is akin to whipping a stick: a slight movement at the wrist results in the tip of the stick traveling a long distance. In the supply chain, even minor changes in end-user demand are significantly amplified as they propagate upstream. For example:
- If the demand for baby diapers increases by 2%, retailers may order 5% more to avoid running out of stock; wholesalers, seeing this 5% increase, might order another 10%; and manufacturers could eventually expand production by 20% based on these orders.
- The pandemic-induced hoarding behavior in the United States in 2022 is a prime example: retailers’ warehouses were filled to capacity, but the Federal Reserve raised interest rates too aggressively, later realizing that this was due to the exaggerated demand created by the bullwhip effect. The reason is simple: upstream parties cannot see the actual end-user demand and can only rely on orders from their immediate downstream suppliers. Additionally, production takes time, leading each stage to order more in order to protect itself, thereby magnifying the initial signal.
2. Distortion One: The Demand for AI Is Uncertain and Based on Speculation
In the traditional bullwhip effect, end-user demand is tangible (e.g., the number of diapers needed per day by babies). However, in the AI industry chain, end-demand is speculative, as it depends on future applications that have not yet been developed. For instance, no one knows for sure whether there will be a breakthrough AI application that will drive significant demand.
- NVIDIA relies on orders from cloud service providers, which in turn depend on estimates from model companies regarding the amount of computing power required for AGI (Artificial General Intelligence).
- The capacity built by upstream players (such as chip manufacturers and data centers) is aimed at potential applications that are still hypothetical, creating a situation similar to building structures in the air without a solid foundation of actual demand.
3. Distortion Two: Hidden Inventory Leads to Sudden Price Drops
Excess inventory is easily identifiable in traditional industries, but in the AI sector, it is hidden:
- Overcapacity in computing power: Machines are produced, but they are not fully utilized due to low actual demand (e.g., idle data centers).
- Promised purchases: Model companies sign contracts for hardware, but these do not appear on their balance sheets, making it difficult for outsiders to assess the true demand.
- Financial Manipulation: Large companies offload their AI-related expenses off the balance sheet, and investment banks package private loans as “investment-grade bonds,” further obscuring the actual demand.
- When excess capacity becomes apparent, prices do not decline gradually; instead, there are sudden, unexpected shocks (e.g., a company canceling orders, causing the entire industry to cool down).
4. Distortion Three: Changing Psychological Motivations
In the traditional bullwhip effect, the motivation is to avoid running out of stock, while in the AI industry chain, the fear is of being left behind in the competitive race for resources. Companies prefer to over-order computing power to ensure they do not fall behind their competitors.
- This creates a “security dilemma”: everyone tries to expand their capacity (by buying more computing power), leading to overcapacity and increased uncertainty. No company dares to stop first, as doing so could result in being outcompeted.
5. The Most Dangerous Link: National Will Amplifies Signals
The most critical link in the AI supply chain is national policy, which can turn vague signals into massive projects with billions of dollars in investment. For example, South Korea’s Lee Jae-myeong’s government initiated a $500 billion semiconductor project located in Gwangju (an agricultural area) to balance regional economic development. However, this led to rapid housing price increases due to speculative investments by real estate developers. Once the factories started production, the industry might encounter a downturn in the semiconductor market.
- The high investment and long construction periods for chip factories mean that any excess capacity affects not just a few companies but also local residents, leading to rising housing prices, job losses, or increased taxes to fund the project.
Conclusion
While the article does not provide concrete solutions, it highlights that the bullwhip effect in the AI industry chain is more concealed and potentially more dangerous than in traditional sectors. To understand one’s position within this chain, quantitative tools are essential—such as analyzing the gap between upstream production capacity and end-user demand or tracking hidden inventory levels. For ordinary people, it is important to avoid being misled by hype about an “AI revolution” and to be cautious of projects that aim to build capacity for the future.