Economics

Compute pricing decomposition at SKU-region-tier grain; hyperscaler quarterly capex from SEC 10-Q / 10-K filings; bilateral semiconductor trade flow against export-control events; frontier training cost modelling; 40-year WSTS market-cycle baseline (1986–2026) with AI overlay.

What AI compute costs, where the chips flow, and what shapes access — the substrates compute governance needs underneath the politics. Four engines render the pricing index, hyperscaler capex, bilateral trade flows, and frontier training cost.

Cloud providers indexed6AWS · Azure · GCP · OCI · CoreWeave · Lambda

Bilateral trade-flow records397,705UN Comtrade + Taiwan Customs

How the four engines compose

Cost Index decomposes $/petaFLOP-day across providers, regions, and hardware tiers. Hyperscaler Capex pulls quarterly capex out of SEC 10-Q / 10-K filings with the YTD-vs-incremental gotcha handled correctly. Trade Impact runs difference-in-differences against export-control events on UN Comtrade + Taiwan Customs bilateral flows. Training Economics carries the model-scale price layer downstream of the index.

Cost Index renders $/petaFLOP-day across providers and regions. Hyperscaler Capex tracks the quarterly AI-infrastructure spend drawn from SEC filings. Trade Impact tests whether export controls actually shifted semiconductor flows or merely rerouted them. Semiconductor Cycle anchors the 40-year market-cycle baseline (1986–2026).

Cost Index

$/petaFLOP-day across 6 cloud providers, by region, hardware generation, and procurement tier (on-demand, spot, reserved-1yr, reserved-3yr).6 providers

Hyperscaler Capex

Quarterly capital expenditure across the AI-infrastructure cohort (FY2017–FY2026), sourced from SEC 10-Q / 10-K filings with the YTD-vs-incremental gotcha handled correctly.25 anchors · 781 rows

Trade Impact

Did US export controls reduce China's actual AI compute access, or redirect chip flows through Singapore, UAE, and Malaysia?397,705 flow records

Training Economics