Ep. 020 - Anthropic vs OpenAI Usage, Margins, Meta Compute, Future of MSL (Tokenomics)

Summary

The SemiAnalysis team discusses enterprise token budgeting, revealing that coding drives over 70% of API revenue and power users spend heavily with strong ROI. They compare Anthropic's high-margin API-heavy business versus OpenAI's consumer-subscription mix, forecasting Anthropic profitability exceeding $1B. Meta's rumored neocloud compute strategy provides a capex backstop, while Google falls behind with locked-in TPU deals. The group also covers token-as-a-service growth favoring AWS and Azure, and the emerging billion-dollar RL data market critical for advancing AI capabilities.

• Coding accounts for over 70% of AI lab API revenue; token budgeting often mis-targets low-usage categories.
• Anthropic is already profitable on $1B+ run-rate due to 80%+ high-margin API revenue mix, while OpenAI's consumer-heavy mix pressures margins.
• OpenAI's API business is rebounding with Codex and new models, turning the race into a two-horse competition.
• Meta's potential neocloud compute strategy with clawback clauses provides a backstop for its massive capex, de-risking AI investment.
• Google's Gemini models rank fifth and its long-term TPU deals without clawbacks signal weak AI conviction.
• Token-as-a-service through hyperscalers (AWS, Azure) is growing rapidly as enterprises buy through existing cloud commitments.
• Reinforcement learning (RL) environment data is a critical new scaling vector, with lab spend potentially exceeding $10B and doubling again.
• The episode ends with a personal bet on whether Anthropic revenue will reach $400B or less by 2027.