The analysis outlines three distinct business models: frontier labs selling tokens, compute providers renting GPUs, and physical infrastructure landlords leasing data center shells. Frontier labs aim for cheap compute and expensive tokens, while compute providers seek high compute prices and sustained demand for AI services. Physical infrastructure landlords prioritize long-term leases with credible counterparties to secure financing for their capex.
The AI infrastructure market is driven by conflicting incentives where each layer benefits from different compute pricing dynamics, creating a complex demand environment for physical assets.