The AI Ledger

Compute, Capital, Chips, Code

Where AI is made, funded, powered, staffed and governed. Live World Bank research indicators joined with Stanford AI Index model counts, Epoch AI release history, SEMI foundry capacity, IEA data-centre energy, MacroPolo talent flows and the 2023-2025 global policy timeline.

Updated 11 mo agosource: Curated
curated
US AI investment
$109.1B
2024 private-market (Stanford AI Index)
TSMC leading-edge share
68%
of global sub-7nm wafer capacity
Frontier releases YTD 2025
8
tracked in Epoch registry
Data-centre energy 2024→30
+166%
IEA Electricity 2025 base case
Live: World Bank IP.PAT.RESD (patent applications), IP.JRN.ARTC.SC (scientific articles), SP.POP.SCIE.RD.P6 (researchers per million), GB.XPD.RSDV.GD.ZS (R&D as % GDP). Curated: Stanford AI Index 2024/25, Epoch AI model registry, SEMI + TrendForce, IEA Electricity 2025, MacroPolo Global AI Talent Tracker, 16-item AI policy timeline.
Chapter 1

The Compute Race

Stanford AI Index counts "notable" ML models — those with ≥100M parameters or comparable compute. The US leads on absolute count; China leads on year-over-year growth. Everyone else is a smaller order of magnitude.

Scroll horizontally through the ticker above to see notable-model counts by country of origin. Below, the frontier chart in Chapter 2 anchors every release by date and training compute — the two chapters read as one continuous view of the compute race.
Chapter 2

Model Releases

Every notable frontier LLM 2018-2025, plotted by release date and training compute. The step-function from GPT-3 (2020) to GPT-4 (2023) captures the scaling-hypothesis era. DeepSeek-R1 (Jan 2025) proved the open-weight Chinese labs could keep up.

Chapter 3

Patents Frontier

World Bank IP.PAT.RESD tracks resident patent applications. China's domestic patent surge post-2011 is one of the largest quantitative shifts in the entire economic-indicator library — even if quality-adjusted counts are less lopsided.

Chapter 4

Chip Manufacturing

The AI compute stack runs on a physical bottleneck. TSMC alone runs 68% of global leading-edge wafer capacity. Everything on this page — model training, inference, agent workloads — flows through a handful of fabs in Taiwan, South Korea and the US.

Chapter 5

AI Investment

Stanford AI Index private-market investment. The US pulled away from everyone else after ChatGPT: $109B in 2024 vs China's $9.3B and the EU's $8.9B combined.

Chapter 6

Energy Footprint

IEA projects data-centre electricity to more than double 2024→2030 — from ~550 TWh to ~1,450 TWh. Google, Meta, Amazon and Microsoft have collectively signed for >20 GW of nuclear capacity to keep up.

Live research data via World Bank API proxy. Curated snapshots refreshed 2025-10-15. Historical model releases from Epoch AI + lab announcements. Fab capacity from SEMI + TrendForce Q2-2025. Data-centre energy projections from IEA Electricity 2025. Talent flows from MacroPolo Global AI Talent Tracker. Regulation timeline compiled from Stanford HAI AI Index Chapter 7, OECD.AI, and the EU AI Act official journal. Loading…