Summary of Key Points
This article uses the "bullwhip effect" (where small fluctuations in end-demand are amplified throughout the supply chain) to analyze the current state of the AI industry. Starting from the actual AI needs of end-users, demand is further driven by factors such as financing leverage, arms races, and long-term contracts as it moves through laboratories, cloud providers, chip manufacturers, and equipment suppliers. The supply chain is now at the "top of the whip-swinging cycle": companies closer to the end-users (such as Microsoft) are beginning to slow down, while upstream equipment manufacturers (such as ASML) are still pushing hard. The future success of the industry depends on whether end-users continue to show interest; otherwise, it could lead to serious consequences such as defaults and asset impairment.
How is the "bullwhip" in the AI supply chain created?
Let's recall the example of selling water: if a customer buys 10 more bottles, the convenience store orders 120, the distributor orders 140, and the manufacturer produces 180—demand is amplified at each stage. The same logic applies to the AI supply chain:
- End-users: Companies that use Copilot and individuals who use ChatGPT have an actual demand of approximately $150-250 billion in 2026 (although there may be double-counting).
- Laboratories和应用 companies (equivalent to convenience stores): Firms like OpenAI and Anthropic, fearing a shortage of computing power, order more than they currently need (e.g., ordering 250 bottles when only 200 are actually required).
- Cloud providers (distributors): Companies like Microsoft and Google, seeing the large orders from laboratories and competing with each other, amplify the orders to 300 bottles, leading to nearly trillion-dollar investments in equipment.
- Chip manufacturers: Manufacturers such as NVIDIA and TSMC lock in long-term contracts (e.g., selling out all of their annual production capacity for HBM4) and even squeeze out capacity for consumer-grade DRAM, increasing orders to 400-500 bottles.
- Equipment suppliers: Companies like ASML expand their production capacity for 2027 based on these chip manufacturer orders, further amplifying the overall demand.
Each link in the chain magnifies the demand due to concerns about shortages, financing competition, and long-term contracts.
Is the AI bullwhip effect more "dangerous" than the water-selling example?
Compared to the traditional water-selling supply chain, the AI bullwhip effect has two additional risks:
1. Uncertainty in end-demand: While the number of bottles sold by a convenience store is visible, it's unclear how many of the AI GPUs are actually generating revenue or sitting idle. Even Meta realized too much capacity after building its data centers and had to rent out extra capacity.
2. More severe consequences of reversals: In the water-selling example, a return of unsold goods would simply reduce inventory; in the AI supply chain, reversals result in defaults and asset impairment, which significantly impact a company's financial health (balance sheet) rather than just its profit and loss.
Current state of the AI supply chain: Some are slowing down, while others are still pushing hard
Recent financial reports reveal different trends among the various links:
- Cloud providers: Microsoft has started to stabilize its equipment investments, with positive free cash flow and rising stock prices; Google, however, is increasing its investments significantly (negative free cash flow of $5.9 billion), leading to a 7% drop in its stock price—companies closer to the end-users are slowing down.
- Chip manufacturers: Although TSMC acknowledges strong AI demand, its equipment investment has only increased moderately; HBM4 production is expected to increase in the second half of the year, indicating that the cycle may not have peaked.
- Simulator chip manufacturers: Companies like TI and NXP are still seeing growing demand, with customers placing orders for future products in advance (improving visibility).
- Equipment suppliers: ASML is the most aggressive, raising its revenue forecast for 2026 to €43-45 billion and expanding production capacity for 2027, while avoiding quarterly order disclosures to hide fluctuations.
Overall, the supply chain is at a stage where the "handle" (cloud providers) is slowing down, but the "tip of the whip" (equipment suppliers) is still moving rapidly.
Are there signs that the bullwhip is being retracted?
In the traditional water-selling example, the end-sales data provide signals for slowing down. For AI, since end-demand is difficult to measure, these signals come from secondary market stock prices:
- When a company announces increased equipment investments, its stock price falls (e.g., Google).
- When a company indicates stable cash flow despite no further investment expansion, its stock price rises (e.g., Microsoft).
The market uses stock prices to signal to the supply chain that it should stop pushing too hard.
The critical question for the AI industry: Will end-users continue to show strong interest?
All the amplified demand ultimately depends on end-users. If users continue to adopt AI widely (for example, through innovative applications like AI-powered offices, education, and healthcare), the current boom will be relatively mild. However, if interest fades suddenly (e.g., due to lack of breakthrough applications or decreased willingness to spend), the excess capacity built will lead to a bubble that could burst, causing defaults and asset impairment.
For investors in AI stocks, it might be better to pray for the emergence of new, compelling AI applications that can drive demand. This analysis clearly illustrates the "bubble risk" in the AI industry: despite the apparent enthusiasm, the growth is fueled by amplified demand that may collapse if end-users lose interest. Understanding this logic helps explain why some AI companies' stock prices rise while others fall and allows investors to avoid the most risky parts of the supply chain.