- Markets - Around The World
- Posts
- Costo Effect
Costo Effect
& Japan's Steel Price is Cheeper than Water
Costco's Impulse Buying Blueprint
Costco has mastered generating unplanned purchases, a phenomenon known as the "Costco effect," where shoppers frequently buy significantly more than intended, sometimes including high-value items like hot tubs or precious metals. This strategy has helped make Costco the third-largest global retailer, often turning planned small trips into large spending sprees.
The retailer employs strategic store layouts, placing essentials deep inside the warehouse while greeting customers with high-end merchandise and seasonal items to maximize product exposure and slow shoppers down. Combined with a frequently changing inventory and limited-time offers, this creates a "treasure hunt" atmosphere that encourages exploration and urgent purchasing decisions.
Impulse buys account for about 25% of Costco's revenue and carry higher profit margins than many staples, making them crucial to the company's financial success. While customers often appreciate the value and unique finds, recent reports indicate shoppers may be becoming more selective about discretionary spending amid economic pressures.
Steel Prices Undercut Water in Japan
In Japan, steel products have become cheaper than bottled water when compared by weight, a result of intense competition among distributors driving down prices. The average price for a one-liter bottle of mineral water was reported at 156 yen ($1.09) in March, highlighting the unusual price dynamic.
This price pressure from a large number of wholesalers is actively undermining attempts by steel manufacturers to increase rates following consolidation efforts. The Japan Domestic Corporate Goods Price Index for Iron and Steel reflects this trend, showing a decline both month-over-month and year-over-year as of March 2025.
Despite the current price drops caused by distribution competition, the overall Japanese steel market was valued at USD 84.3 billion in 2024 and is projected to grow in the coming years. This situation coincides with recent decreases in Japan's crude steel output and falling prices for steel scrap.
Surging Costs Define AI Model Training
Training state-of-the-art AI models now frequently surpasses $100 million, with Google's Gemini 1.0 Ultra estimated at $192 million and OpenAI's GPT-4 at $79 million. Other costly examples include Meta's Llama 3.1-405B ($170M) and xAI's Grok-2 ($107M).
These high expenses stem from extensive cloud compute rentals and significant R&D investment, where staff salaries can constitute up to 49% and hardware like AI accelerator chips 23% of the total cost for models like Gemini Ultra. While some newer models like DeepSeek-V3 report much lower training costs ($6 million), these figures are sometimes disputed.
The cost of training frontier models has grown dramatically, approximately 2-3 times per year recently, raising concerns about accessibility for organizations without massive funding. Projections suggest that if this trend persists, the largest AI training runs could cost over a billion dollars by 2027.