Summary of Key Points
This article reveals the profit-making strategies of AI model companies by examining the author's own AI-powered consumer bills. The main findings are as follows:
- Surging Gross Margins: AI companies are experiencing significant increases in gross margins (rising from losses two years ago to a maximum of 80% today), yet the cost for users has not decreased; instead, it has increased.
- Diverse Profit Models: Different companies use various tactics to generate revenue (high prices, low prices with high efficiency, price hikes, subsidizing free users, etc.).
- The Purpose of Profits: The resulting gross margins are all used to fund the development of the next generation of AI models.
Detailed Explanation
1. Surge in AI Company Gross Margins: From Losses to Profit
Two years ago, AI companies were selling services at a loss—for example, Anthropic would spend $194 on推理 costs (electricity, GPU rental, etc.) for every $100 earned, resulting in a loss of $94. Today, the cost has dropped to only $20 per $100 earned, leaving an 80% gross margin.
There are two main reasons for this change:
- Increased Hardware: Partnerships like those between Anthropic and SpaceX have led to a significant increase in computing power, with an additional 220,000 GPUs, doubling the computational capacity.
- Improved Efficiency: Engineers have optimized algorithms to make better use of hardware (e.g., processing more requests with the same number of GPUs). According to Epoch AI, the cost of equivalent computing power has decreased by 50 times in a year.
These two factors combined have significantly reduced the costs for AI companies, leading to higher gross margins.
2. Rising User Bills: Why Haven’t Prices Fallen Despite Lowered Costs?
Logically, lower costs should lead to lower prices, but AI companies are either raising prices or maintaining them:
- Changing Pricing Logic: Companies no longer base their pricing on their own costs; instead, they consider the alternative cost for users. For example, if using AI saves users time (e.g., avoiding a night of work), they are willing to pay higher prices.
- Examples of Price Increases: Zhispu GLM Coding Pro’s monthly subscription fee has increased from $149 to $538 (with three price hikes in a year); Anthropic changed its pricing model for enterprise users, increasing their bills significantly; even DeepSeek has implemented tiered pricing during peak usage times.
In essence, AI companies are reaping the benefits of reduced costs without passing them on to consumers.
3. Diverse Profit Strategies Among AI Companies
Despite all earning gross margins, each company uses different approaches:
- Anthropic: Focuses on high prices and control over pricing, selling the most expensive models with an 80% margin.
- DeepSeek: Offers low prices while maintaining high efficiency, charging only $2 for generating 1 million tokens, still achieving a 70%+ gross margin.
- Zhispu/MiniMax: Increases prices and expands its scale, with Zhispu’s API gross margin rising from 4% to 18.9% (with three price hikes in a year), and MiniMax seeing its gross margin increase from 12% to 25%.
- OpenAI: Continues to subsidize free users, incurring a loss of $930 million in the first quarter despite earning $5.7 billion. This is because 1 billion free users consume $0.70 in computing power each month, effectively having paying users subsidize the free ones.
4. The Use of Gross Margins: Funding the Next Generation of AI Models
Most of the profits generated by AI companies are not used for shareholder dividends but instead go towards training the next generation of models. Training new models requires expensive resources (GPUs, data centers), which are not included in gross margins but are recorded as research and development expenses.
- Competitive Advantages: Companies with higher gross margins can afford more GPUs, hire top talent, and offer better subsidies to users, giving them a competitive edge in developing the next generation of AI models.
- Financing Needs: Companies with lower margins rely on financing to sustain their operations.
Conclusion
The money you spend on AI subscriptions not only supports the companies but also indirectly funds the development of the next generation of AI models. Although the bills may be expensive, AI saves users time and makes life more convenient. Given its practical benefits, you may continue to use these services despite the higher costs.
(After all, the convenience it provides is undeniable, and the math just doesn’t add up against its value!)