(Early) AI Compute Asset Pricing Research

This study proposes an early asset pricing framework for compute, pointing out that compute capital expenditures and service flows have exceeded 1% of US GDP and are growing rapidly, with compute prices reflecting the uncertainty of AI application adoption. The non-storable nature of compute invalidates the direct no-arbitrage relationship between futures and spot prices, and synthetic futures prices constructed based on existing term lease contracts are most likely an upper bound for true futures prices. Preliminary evidence from the study shows a positive compute risk premium, indicating hedging pressure on the supply side of compute.