# Sources — AI Economics Reality Check

This is the full source list for the article *"What Has to Be True for the AI
Buildout to Pay Back?"* and its companion artifact, the AI Economics Reality
Check tool. Numbering matches the article's footnotes, so you can cross-reference
directly between the two.

A disclosure caveat applies throughout: AI-specific utilisation, AI-only capex
and AI-only revenue are not cleanly disclosed by the major cloud providers.
Figures attributed to company filings use reported totals and stated
management commentary as proxies; the article deliberately avoids implying
direct utilisation measurement.

1. Brookings, *The Telecommunications Crash: What To Do Now?*, for the dot-com/telecom buildout history separating social value from investor return. https://www.brookings.edu/articles/the-telecommunications-crash-what-to-do-now/
2. Stanford HAI, *AI Index 2026*, for capability, adoption, investment and measurement context — with the caveat that reported organisational use is not the same as deep workflow transformation. https://hai.stanford.edu/ai-index
3. Microsoft FY26 Q3 earnings call, for capex, AI ARR commentary, capacity constraints and asset mix. https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q3
4. Amazon Q1 2026 Form 10-Q, for PP&E purchases, operating cash flow, AWS assets and depreciation. https://www.sec.gov/Archives/edgar/data/1018724/000101872426000014/amzn-20260331.htm
5. Alphabet Q1 2026 results, for Google Cloud growth and capex. https://www.sec.gov/Archives/edgar/data/1652044/000165204426000043/googexhibit991q12026.htm
6. Meta Q1 2026 results and 10-Q, for capex guidance, cash flow and infrastructure investment. https://investor.atmeta.com/investor-news/press-release-details/2026/Meta-Reports-First-Quarter-2026-Results/
7. Sequoia Capital / David Cahn, *AI's $600B Question* — an investment thesis to pressure-test, not a settled estimate. https://www.sequoiacap.com/article/ais-600b-question/
8. Goldman Sachs, *Gen AI: Too Much Spend, Too Little Benefit?* — an investor debate, not a neutral baseline. https://www.goldmansachs.com/insights/top-of-mind/gen-ai-too-much-spend-too-little-benefit
9. Bonney et al., U.S. Census CES, *The Microstructure of AI Diffusion*, for nationally representative firm/function/task diffusion evidence. https://www.census.gov/library/working-papers/2026/adrm/CES-WP-26-25.html
10. Arntz et al., ifo Working Paper, *Low Barriers, High Stakes*, for formal and informal workplace diffusion evidence. https://www.ifo.de/en/publications/2025/working-paper/low-barriers-high-stakes-formal-and-informal-diffusion-ai-workplace
11. Dillon, Jaffe, Immorlica and Stanton, *Shifting Work Patterns with Generative AI* (NBER), for field-experimental evidence that time savings can precede task-composition change. https://www.nber.org/papers/w33795
12. BIS Working Paper 1325, for European firm-level evidence on AI adoption, productivity and employment. https://www.bis.org/publ/work1325.htm
13. IMF, *Gen-AI: Artificial Intelligence and the Future of Work*, for exposure, complementarity and inequality mechanisms. https://www.elibrary.imf.org/view/journals/006/2024/001/article-A001-en.xml
14. *Journal for Labour Market Research*, occupational exposure and complementarity, for the distinction between exposed roles and substitutable roles. https://link.springer.com/article/10.1186/s12651-025-00418-w
15. Aghion et al., AEA Papers and Proceedings, *How Different Uses of AI Shape Labor Demand: Evidence from France*. https://www.aeaweb.org/articles?id=10.1257%2Fpandp.20251047
16. ILO Working Paper 166, *Disruption without dividend?*, for global GenAI exposure and uneven dividend risks. https://www.ilo.org/publications/disruption-without-dividend-how-digital-divide-and-task-differences-split
17. Ford technical-specialist programme, originating in Bloomberg reporting (June 2026) with on-the-record statements by COO Kumar Galhotra and VP Charles Poon; accessible secondary coverage citing Bloomberg includes Business Insider and TechCrunch. Press-mediated case study, not economy-wide evidence. https://www.businessinsider.com/ford-ai-hiring-veteran-engineers-2026-6 and https://techcrunch.com/2026/06/28/ford-rehires-gray-beard-engineers-after-ai-falls-short/
18. People, Ford statement as reported — used only to corroborate the company's framing of the programme. https://people.com/ford-hires-over-300-engineers-including-former-employees-after-finding-ai-couldnt-replicate-their-work-12010429
19. Brynjolfsson, Li and Raymond, *Generative AI at Work* (NBER), for customer-support productivity evidence. https://www.nber.org/papers/w31161
20. Noy and Zhang, *Science*, for professional-writing productivity evidence. https://doi.org/10.1126/science.adh2586
21. METR / Becker et al., *Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity* — a small randomised trial; a bounded counterexample, not a general result. https://arxiv.org/abs/2507.09089
22. METR, *We are Changing our Developer Productivity Experiment Design* (February 2026) — follow-up data on late-2025 agentic tools suggesting likely speedup, judged unreliable by the authors due to selection effects. https://metr.org/blog/2026-02-24-uplift-update/
23. Dell'Acqua et al., *Navigating the Jagged Technological Frontier*, for within-frontier gains and outside-frontier risks. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4573321
24. Cottier et al., *The rising costs of training frontier AI models*, for frontier-training cost concentration alongside falling adaptation costs. https://arxiv.org/abs/2405.21015
25. IEA, *Energy and AI*, for global data-centre electricity demand and delay risks. https://www.iea.org/reports/energy-and-ai
26. EPRI, *Powering Intelligence 2026*, for U.S. data-centre electricity demand scenarios. https://powering-intelligence.epri.com/executive-summary.html
27. U.S. Bureau of Economic Analysis, GDP/NIPA primer and the expenditure approach, for final-demand framing. https://www.bea.gov/resources/methodologies/measuring-the-economy and https://www.bea.gov/index.php/news/blog/2025-06-03/expenditures-approach-measuring-gdp
28. Acemoglu, *The Simple Macroeconomics of AI* (NBER), for modest TFP estimates and task-based macro framing. https://www.nber.org/papers/w32487
29. Funtowicz and Ravetz, on post-normal uncertainty. https://onlinelibrary.wiley.com/doi/abs/10.1002/etc.5620131203
30. "Apple Is the King of AI and Nobody Knows It," limitededitionjonathan (Substack), July 2026 — an independent, single-author opinion/analysis piece, used for its memory-bound-inference mechanism and commoditization framing. Its claim about Apple's competitive destiny is the author's opinion and is not adopted by the article; specific figures in the piece (benchmark throughput, GPU rental-price moves, exact release dates) were not independently verified and are not restated as fact. https://limitededitionjonathan.substack.com/p/apple-is-the-king-of-ai-and-nobody

This list is generated from the article's own footnotes and kept in sync with
it. If a number here ever looks out of step with the published article, the
article is the source of truth.