Tokenomics Is Dead. The Lie Of Affordable Inference
Summary
OpenAI is projected to incur substantial losses, burning approximately $9 billion against $13 billion in revenue in 2025, with an estimated $14 billion in losses for 2026. Inference costs alone reached $12 billion with Microsoft by November 2025. Anthropic's gross margin is expected to be around 40% for 2025, a decrease from its 50% target. The current "cheap inference" environment is largely sustained by venture capital and hyperscaler cross-subsidies, with major players like Microsoft, Amazon, Nvidia, and SoftBank simultaneously investing in, supplying, and purchasing from the same AI labs. This financial model suggests that the industry is vulnerable to a significant repricing, potentially 10x, once these subsidies cease, with planned IPOs potentially impacting retirement funds.
Key takeaway
For CTOs and VPs of Engineering evaluating AI adoption, recognize that current "affordable inference" models are financially unsustainable and heavily subsidized. Your long-term cost projections for AI services should account for a potential 10x price increase once these subsidies inevitably end. Prioritize vendors with transparent cost structures and avoid over-reliance on services whose pricing is artificially deflated by investor capital.
Key insights
Current "cheap inference" is a VC and hyperscaler-subsidized illusion, masking significant AI industry losses.
Principles
- Subsidized pricing distorts true market costs.
- Interconnected funding creates systemic risk.
In practice
- Analyze AI service costs beyond current pricing.
- Diversify AI vendor relationships.
Topics
- AI Industry Economics
- Inference Costs
- Hyperscaler Subsidies
- Tokenomics Critique
- Market Repricing Risk
Best for: CTO, VP of Engineering/Data, Director of AI/ML, Investor, Executive, Consultant
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by AI Advances - Medium.