AI's Economics Don't Make Sense
Key Points:
- Microsoft announced that GitHub Copilot will switch to usage-based pricing starting June 1, 2026, charging users based on the actual compute costs rather than a fixed subscription, due to unsustainable subsidies covering heavy AI usage.
- The economics of generative AI services are fundamentally broken, as AI companies have been heavily subsidizing compute costs, leading to massive losses and inevitable price hikes or restrictive usage limits for users.
- Large AI data centers, such as Oracle’s Stargate Abilene, involve enormous capital expenditures and operational costs, requiring sustained, high revenue from customers like OpenAI to be viable, but current growth and profitability projections appear unrealistic.
- AI startups like OpenAI and Anthropic face significant financial challenges, with token-based billing revealing extremely high per-user costs that many companies struggle to justify, raising doubts about the long-term sustainability of the generative AI industry.
- The article argues that generative AI products have been marketed deceptively through subscription models that mask true costs, creating an unsustainable bubble dependent on continuous venture capital and debt, with looming risks of collapse affecting major players like Oracle and OpenAI.