Lo Toney on AI Inference Economics and 2026 Market Trends

AI Monetization Challenges in 2026

• 💡 Investors are cautious about software companies' ability to monetize AI effectively in 2026, leading to underperformance in software names like Adobe and Workday.
• ⚠️ A potential bottleneck for AI adoption is workflow and human culture, as existing data center structures and user habits may hinder the scaling of AI use.

Understanding AI Economics: Training vs. Inference

• 🔑 Training AI is a one-time build cost, whereas inference represents the ongoing operating cost of running AI in real-world applications.
• 📈 Inference economics studies who pays for AI each time it runs and whether revenue growth outpaces cost increases.
• 🎯 For companies, inference determines profitability, making it a critical factor to watch.

Company Strategies for AI Inference Costs

• 🍎 Apple is protected by its economics, with third-party apps handling AI execution, minimizing Apple's exposure to marginal inference costs.
• 💻 Microsoft integrates inference costs into existing models of seats, licenses, and contracts, absorbing these expenses.
• ☁️ Google utilizes vertical integration with its TPU chips, managing both training and inference within a single data center, and effectively taxing inference rather than subsidizing it.
• 📊 Meta faces exposure as inference costs can rise faster than monetization, especially since it monetizes attention; the acquisition of Manis aims to improve P&L by integrating AI into tasks and workflows.