Reading list · curated, source-linked · updated as the literature grows

Research

The academic work behind this site: how compute is priced, how GPUs depreciate, and how hardware-backed credit works. Curated, dated, and linked to the primary source.

Compute futures pricing
July 2026
(Early) AI Compute Asset Pricing

The first asset-pricing framework for the announced CME/ICE compute futures: cash-and-carry arbitrage fails because compute is non-storable, term rentals give an upper bound on futures prices, and early synthetic-futures evidence points to a positive risk premium from provider hedging pressure.

AI financing structure
March 2026
Financing the AI Buildout

Maps the shift from on-balance-sheet corporate finance to leases, project finance, securitization, private credit and SPVs — the structures behind every facility in this ledger — and argues this architecture transforms rather than eliminates risk.

Compute futures pricing
June 2026
Pricing Compute Futures: Forward Curve and Volatility for a Non-Storable, Depreciating Commodity

Schwartz-Smith two-factor model with scheduled jumps at hardware launch windows. Calibrates generational repricing at roughly −20% per event and incumbent rental depreciation at 15–20% per year — direct inputs for residual-value modeling.

GPU-backed debt
June 2026
The Yield Premium of Compute: Pricing Depreciation Risk in GPU-Collateralized Debt Obligations

Models GPU-backed debt spreads as a function of borrower default probability, collateral depreciation half-life, secondary-market liquidity, and the covariance between obsolescence shocks and default — asking how much of the observed yield pays for obsolescence rather than credit risk.

Systemic risk
April 2026
The AI Circular Economy: Systemic Risk, Vendor Financing, and the Keystone Problem

Maps seven circular financing loops totalling over $1.4T in committed spending among Nvidia, OpenAI, Microsoft, Oracle, CoreWeave and others, and argues OpenAI is the keystone whose insolvency would impair the collateral valuations behind GPU-backed debt.

GPU depreciation
2022 (data updated since)
Trends in GPU Price-Performance

FLOP/s per dollar doubles roughly every 2.5 years across a large GPU dataset — the underlying technological driver of GPU collateral depreciation and used-hardware price decay.

AI infrastructure
April 2025
Trends in AI Supercomputers

Performance of leading AI clusters grew ~2.5x per year while hardware cost and power requirements doubled annually — the empirical sizing of the capex wave that GPU-backed debt finances.

Electricity analogy
2002
Equilibrium Pricing and Optimal Hedging in Electricity Forward Markets

The canonical model of forward premia in a non-storable power market, driven by which side hedges harder. Compute is closer to electricity than to oil, so this is the template the compute-futures literature builds on.

Electricity analogy
2004
Electricity Forward Prices: A High-Frequency Empirical Analysis

Documents significant, sign-switching risk premia in day-ahead electricity prices — the best empirical evidence on how premia behave in the closest existing market to rented compute.

Collateralized lending
2009
Collateral Pricing

Shows, using airline fleets, that more redeployable collateral lowers credit spreads and supports higher loan-to-value — the framework for thinking about GPU redeployability, advance rates and the two-lane price data on this site.