Sources & Methods#
This book makes strong claims. This page explains how those claims are sourced, and how to read the parts that are deliberately illustrative.
The standard we hold ourselves to#
Every quantitative claim in this book — every percentage, dollar figure, gigawatt, and date — is meant to trace to a primary or authoritative secondary source: government and regulator data (EIA, Federal Reserve, Bank of England/FCA, CFPB), standards and lab bodies (LBNL, NREL, IEEE), multilateral and NGO research (IEA, FSB, Amnesty International, UNICEF, Ellen MacArthur Foundation), peer-reviewed journals, and named companies’ own disclosures.
Where a widely-circulated figure turned out to be wrong or unsupported, we corrected it — even when the correction made the number less dramatic. A few examples of what changed in this edition:
- Rack compute-density growth is described as ~9x in three years (8-GPU servers → 72-GPU GB200 NVL72 racks), not the internally-inconsistent “12x.” The 576-GPU/600 kW configuration is NVIDIA’s Rubin Ultra NVL576, expected in H2 2027 and spanning eight racks — a future data point, not a current one.
- Renewable vs. fossil energy-return-on-investment is presented with harmonized figures (solar PV ~11–12:1; fossil useful-stage ~8.5–14:1, per Brockway et al., Nature Energy 2024) — a tightening margin, not the cherry-picked “2.4:1 vs 80:1.”
- Regenerative-agriculture profit figures are cited per hectare with their study of origin, alongside BCG’s US-farm profitability range.
- Financial-system claims are grounded in the Bank of England/FCA 2024 survey and FSB analysis, not a precise-sounding “87% of banks use three models” for which no source exists.
If you find a figure you believe is wrong, tell us — corrections make the argument stronger, not weaker.
How to read the scenarios#
Parts of this book imagine the future. The 2027 Scenario in particular walks month-by-month through how the convergence could unfold. These passages are explicitly illustrative — a plausibility narrative built on real, cited trends, not a dated forecast of specific events. They are labeled as scenarios wherever they appear. The underlying drivers (grid lead times, chip concentration, model homogenization) are sourced; the specific dated events are a storytelling device.
Key sources by domain#
AI compute & energy
- Goldman Sachs Research — data-center power demand
- Lawrence Berkeley National Laboratory — Queued Up: interconnection queues & lead times
- U.S. EIA — utility-scale capacity additions
- NVIDIA — GB200 NVL72 specifications
- Epoch AI — energy per AI query
Semiconductors & critical minerals
- CSIS — rare-earth & supply-chain concentration
- IEA — Global Critical Minerals Outlook
- Amnesty International — This Is What We Die For (cobalt/DRC)
- TechInsights — EUV lithography energy use
Finance & algorithmic risk
- Financial Stability Board — AI & financial stability (2024)
- Bank of England / FCA — AI in UK Financial Services 2024
- INET Oxford — Value-at-Risk pro-cyclicality
- U.S. Federal Reserve — 2024 Small Business Credit Survey
- The Markup — bias in mortgage-approval algorithms
- Frank Pasquale, The Black Box Society (Harvard University Press, 2015)
Regenerative agriculture
- Frontiers in Sustainable Food Systems (2025) — Tanzania conservation-agriculture trial
- Boston Consulting Group (2023) — Regenerative Agriculture Profitability for US Farmers
- LaCanne & Lundgren (2018), PeerJ — regenerative vs. conventional corn
- Rodale Institute — Farming Systems Trial
Microgrids & circular economy
- U.S. Army — 14-day installation energy security
- OSTI / NREL — valuing resilience for distributed energy resources
- Casa Pueblo / Adjuntas Pueblo Solar — Canary Media coverage
- Kalundborg Symbiosis — official site
- Ellen MacArthur Foundation — circular-economy resilience
This edition was fact-checked claim-by-claim against the sources above. Full per-chapter citations appear in each chapter’s “Sources” section.