Compute Price Determination

Determine the fair market price of compute resources (e.g., GPU/CPU compute hours) when treated as a tradable commodity, specifying how demand, supply, technological advancement, and market conditions should be reflected in pricing to enable transparent price discovery.

Background

The paper proposes commodifying compute resources via the Global Compute Exchange (GCX) to address shortages, underutilization, and price volatility. Despite outlining market structures and instruments (spot, futures, options), the authors highlight that the appropriate price level for compute remains unknown.

This uncertainty stems from the unique characteristics of compute relative to traditional commodities, including rapid technological change, heterogeneous hardware, and fragmented markets. Establishing a fair market price is central to GCX’s goals of transparency and efficient allocation, making price discovery a foundational open question for the proposed marketplace.

References

What should the price of compute be? This is not known.

— Commodification of Compute  (2406.19261 - Kristensen et al., 2024) in Section 1: Introduction

Finally, our discrete-tier approximation with $K=2$ tiers yields substantial profit gains in favorable regimes (Remark~\ref{rem:few-tiers}), but characterizing optimal mechanisms with continuous tier choices remains open.

— Pricing Time, Not Just Tokens: Latency-Aware Mechanism Design for LLM Inference  (2609.40098 - McDougall et al., 30 Sep 2026) in Section 7, Conclusion, subsection “Limitations and Extensions”