The Geography of AI for PeacebuildingOpen Data

DATASET / OPEN DATA

The Geography of AI for Peacebuilding: Compute, Training Data, and Conflict

This dataset compares three geographies of artificial intelligence at the country level: where AI compute infrastructure is physically located, where the organisations that compile AI training datasets are headquartered, and where armed conflict events occurred between 2023 and 2026. It combines three independently maintained sources: Epoch AI's compute infrastructure tracker (H100-equivalents and power capacity, CC BY 4.0), the Armed Conflict Location & Event Data Project (ACLED, aggregated country-year event counts only), and the Data Provenance Initiative's (DPI) audit of AI training datasets, aggregated to country level via DPI's institution-to-country lookup table.

Key findings

• 92.9% of AI compute capacity is located in the Global North.
• 88.99% of ACLED-recorded conflict events specifically targeting civilians (2023–2026) occurred in the Global South.
• 89.1% of AI training datasets audited by DPI were compiled by organisations headquartered in the Global North.
• Of the fifteen countries with the highest number of civilian-targeting conflict events in this period, twelve have no recorded AI compute infrastructure whatsoever.

Put simply: the places generating the data and infrastructure that AI systems are built on are, overwhelmingly, not the places where conflict is concentrated. For an Alliance built around the principle that PeaceTech should be shaped by the realities of peacebuilders and affected communities rather than by where compute happens to sit, this dataset gives that concern a concrete, citable basis.

What's in the deposit

The dataset is deposited on Harvard Dataverse under a CC BY 4.0 licence and includes a formatted Excel workbook (raw data, formula-driven summary tables, and charts), a self-contained interactive HTML dashboard, a full methodology report (PDF) documenting sources, the hybrid Global North/South classification convention used, and known limitations, plus six print-ready summary infographics and the Python build script used to compile the dataset.

Licensing note: ACLED data is included as aggregated country-year event counts only, in line with ACLED's redistribution terms. Epoch AI data is used under its CC BY 4.0 licence with attribution. See the README and Methodology Report in the deposit for full source citations.