Who he is
Kathryn and Shelby Cullom Davis Professor of History at Columbia and director of its European Institute; author of Crashed, Shutdown, The Deluge; writes Chartbook (Substack) at near-daily frequency, co-hosts the Ones and Tooze podcast with Cameron Abadi, and columnizes for Foreign Policy and, lately, the FT. Affiliation re-checked 3 Oct 2026: unchanged. One-line position: AI is a macro-financial event — the biggest capital-spending bet in US history, with energy, inflation, sovereign and financial-stability stakes — and, since autumn 2026, a political-economy event too: an "oligarchic arms race" run by private investors, not a state project.
Discipline & technical bet
Macro-financial historian. Non-technical, and no personal technical bet: no building, no tech equity thesis, no fund. Correction this pass: the Aug 2026 file said he had "no disclosed AI-tool practice (checked through Jul 2026)". That was wrong. Chartbook 385 (16 May 2025) is a long, hands-on account of him working ChatGPT paragraph-by-paragraph through Mao's Little Red Book, then getting it to generate Maoist-formula manifestos for the USA, and comparing DeepSeek's refusal. He is a competent, curious user of chat models — a 15-month-old fact our July check missed. Weighting rule, updated: capital-flow, energy, inflation and history claims get full weight; claims about what models can DO get a smaller discount than before (he has sat at the keyboard), but he still reads balance sheets, not loss curves, and has never seen a training run. He remains the bench's clean "non-technical, no bet" control case — just not a non-user.
Key claims (Says)
- The AI capex supercycle is a macro event — ~$650B/yr ≈ 2% of US GDP (his Feb 2026 figure); by Sep 2026 he is relaying the WSJ/Brookings estimate of $10.3T over 2025–2032, ≈3.6% of GDP a year, more than canals, railroads and the grid combined (Top Links 1239, 29 Sep 2026) and telling Foreign Policy that AI is now 36% of US business investment, $150–200B a quarter (FP, 24 Sep 2026).
- Three-bubble taxonomy (Foreign Policy, 13 Feb 2026): metaverse-type (nobody wants it), railway-type (transforms the world, finance melts down anyway), airline-type (changes society, never makes money). In Feb he refused to pick. Drifting — see Delta log.
- Energy is the binding physical constraint — grid, gigawatts, geopolitics. Still in the file; notably not where his AI writing went Aug–Oct 2026, which was finance and politics.
- The revenue gap is the central question — in Jul 2026 he flagged that global AI sales ex-China ($25B, Q1 2026) had only just caught up with estimated depreciation ($21B): "thin" is an understatement (Top Links 1159, 11 Jul 2026).
- AI is now inflationary infrastructure (Jul 2026): electronics PPI +27% (May), Apple hardware +20%, Korea's $576B chip/AI program (Chartbook, 1 Jul 2026); amplified the WSJ's "third wave of inflation" framing (26 Jun 2026).
- NEW — AI is a political-economy rupture, not a state project. "To have a trillionaire in our midst ruptures the social bargain" (Ones and Tooze, via Futurism, 1 Aug 2026); AI as "oligarchic arms race"; MAGA-meets-AI as a "crossover" of two dystopias with no coherent state behind either (Chartbook 456, 5 Jul 2026); the Manhattan Project analogy the White House uses for its Genesis Mission is "sanitised" — this is "trillionaire greed, junk bonds and jerry-rigged financial engineering" (FT op-ed 2 Oct 2026; Chartbook 477, 3 Oct 2026).
- NEW — K-shaped economy with AI as the engine. Top 10% of earners ≈ half of consumer spending; AI-driven growth is "endogenously generated" through investor confidence and credit, not a Cantillon effect (FP, 12 May 2026; Chartbook 455, 1 Jul 2026).
- NEW — Regulatory-capture hypothesis. Frontier labs ask for regulation because closed-model economics need protection from Chinese models at "95 percent of its performance for a fraction of the price"; liability law could in principle bite ("clearly negligent"), but the state lacks the talent (FP, 24 Sep 2026). Also: lab safety talk is heading for a clash with Wall Street and the White House (Top Links 1227, 17 Sep 2026).
Notable predictions — with falsifiable checks
- Capex cycle produces macro consequences regardless of AI's technical success. Check: AI share of GDP growth + AI items in PPI/trade data. HELD, strengthening: Commerce revised Q2 2026 GDP to 2.2% on 30 Sep 2026, with AI/tech capex supplying 36–53% of growth depending on import treatment (Marketplace; ING); PPI final demand +5.4% y/y in Aug 2026 (BLS, 10 Sep 2026) with memory "chipflation" named as a driver (Axios, 11 Aug 2026).
- Energy becomes binding. Check: data-center power contracts vs grid additions; utility capex; interconnection waits. HELD on third-party data: US utility plans ≈$1.4T; interconnection waits of 5+ years in major hubs; ~$64B of projects blocked across 54 local moratoriums (2026 trade reporting, secondary — primary FERC/EIA confirmation still owed). Caveat: Tooze himself has not advanced this claim since Aug; the grade rests on the observable, not on him.
- Buildout → bubble → correction pattern applies. Check: which taxonomy branch by end-2028; capex-to-revenue ratio annually; a >30% AI-equity drawdown grades the railway branch Held. OPEN, first tremor logged: Nasdaq-100 entered its second 2026 correction (−10%) on 28–31 Jul 2026 on AI-payoff worries (NBC; Fortune, 28 Jul 2026); NVDA ≈−18% from its May record by mid-Sep (TIKR). Below the 20% wake trigger and the 30% grading bar. Not gradable yet.
- NEW — The bubble will pop through the credit side, not the equity side. Implied by his Dec 2025 tech-CDS note (basket CDS volumes +90% since Sep 2025; AI ≈30% of 2025 IG net issuance — Chartbook, 30 Dec 2025) and the Oct 2026 "junk bonds and jerry-rigged financial engineering" line. Check: hyperscaler bond coverage ratios (reported falling from ~5x in Feb to under 2x by Jul 2026), tech CDS spreads, and whether a data-center SPV or private-credit vehicle defaults before end-2027. Status: not yet falsified; coverage-ratio drift is consistent.
- NEW — Political consequence: cost-sensitive households punish the incumbent party (FP, 12 May 2026). Check: Nov 2026 midterm result among the "Costco set" / lower-middle-income counties. Gradable 4 Nov 2026.
Revealed behavior (Does)
- Ships Chartbook relentlessly — Top Links 1159 (11 Jul) to 1239 (29 Sep) is 80 numbered issues in 80 days. His revealed bet is still that synthesis-at-frequency is the edge; he monetizes interpretation, not positions. Held.
- No known personal portfolio positioning on AI either way — diagnostician, not bettor; no fund, no board, no startup found this pass. (Contrast Aschenbrenner, whose 13F IS his argument.) Held.
- Steers mainstream audiences into capex-finance-and-energy framing — and now escalates: an FT op-ed (2 Oct 2026) directly contesting the White House's Manhattan Project analogy; the Musk-trillionaire remark that went viral (Aug 2026). Moving the discourse is still the action, with a sharper political edge. Held, intensified.
- Uses chat models as a research instrument when they help (ChatGPT for Maoist Chinese, May 2025) and writes it up — behavior, not a bet.
Feels
Fears watching another 2008 assemble in plain sight while consensus cheers — now layered with a historian's unease that the comfortable analogies (Manhattan Project, railways) are themselves a sedative. Wants history to be operationally useful; wants to be the one who saw the financial architecture — and the social bargain being broken — clearly.
Hears
Financial-history lineage (Kindleberger-school bubbles), FT/Alphaville and WSJ data journalism, Brookings/BIS/Deutsche Bank research notes, European political economy. Downstream of markets data and the business press, not of the labs — the FP regulation piece shows him reasoning from Chinese model pricing rather than from any lab contact.
Sees
Vantage: archives, macro datasets, finance-ministry and central-bank networks — he sees flows (imports from Taiwan and Mexico, bond issuance, CDS baskets) others miss. Blind side: has never seen a training run; inside-the-lab claims are hearsay to him.
Incentive map
Re-verified 3 Oct 2026: Columbia chair = independence; Chartbook subscriptions reward credible-contrarian macro synthesis — mild drama gradient (the "Strangelove + Gekko + Big Short" line is good copy), offset by a reputation staked on calibration. No tech equity, no lab ties, no fund, no new commercial venture found. Unusually free to call a bubble; cannot afford a botched macro call — his brand IS the track record. Growing FT/FP platform slightly raises the cost of hedging forever: editors want a verdict.
Theories aligned with
- Financial history of technology buildouts (Kindleberger, Perez — with the twist that he now warns against leaning on them)
- Macro political economy of infrastructure
- Adjacent to Mazzucato (state investment) but diagnostic, not prescriptive — and now explicitly arguing the US state is absent from this buildout
- BIS labor-share scenarios (54% → 20%) as his inequality frame
What he's reacting against
- Tech-centric AI discourse that ignores capex, energy, credit and sovereign exposure
- The assumption the buildout is self-financing
- Both hype and dismissal argued from inside the tech bubble
- New: "historic normalization" — the Manhattan Project / arms-race analogies that make a private, leveraged, oligarchic buildout sound like a disciplined public mission
Where he overlaps / splits (with Rich)
- Overlaps: picks-and-shovels/Engine-Room instincts — his capex-and-energy lens IS the infra half of the Barbell; energy-constraint realism; both-real-and-overhyped duality (kaleidoscope thinking). The credit-side pop thesis is a direct input to Rich's deflationary-spiral and token-capital threads.
- Splits: Tooze diagnoses from above; Rich builds from below at the presentation layer. Tooze has no account of what individuals DO with the technology — exactly the layer Rich creates in. His "which social contract does this belong to?" is a question Rich's §-level principles answer for one life; Tooze asks it for the polity.
- Vs Zitron/Covello: same skepticism, centuries-deep register, and now a credit-market mechanism. Vs Smil: adds finance/geopolitics to the energy physics. Vs Nadella/Huang: history says transformative AND destructive can be simultaneous — and he now adds: stop assuming history will tell you which.
Track record
- Crashed (2018): the definitive 2008 account — proven at exactly this genre of analysis.
- Called the capex/energy framing early (2024) — by mid-2026 it is consensus vocabulary; the "biggest bet in US history" framing is now WSJ/Brookings boilerplate.
- Calibration credit: flagged in Feb 2026 that AI wasn't in macro data "yet" — then published the data himself when it arrived (Jul 2026); the Q2 GDP revision (Sep 2026) vindicated him again.
- Early on credit risk: tech-CDS basket note (Dec 2025) preceded the mid-2026 coverage-ratio deterioration.
- Minus: the Feb 2026 taxonomy has not been resolved into a call; the Oct 2026 move is to reject the analogy frame rather than pick within it.
Empirical vs normative
Empirical: capex figures, PPI, GDP contributions, trade flows, bond issuance, historical analogues — heavily sourced. Normative: was mostly implicit; now explicit and louder — a trillionaire "ruptures the social bargain", the buildout is "greed", the state should be present and isn't. Still prescribes little in policy terms (the FP regulation piece ends in capacity pessimism, not a program).
Weak spots / open questions
- No technical floor: cannot independently evaluate capability claims that drive the revenue side of his own equation — though he is a hands-on user, which the Aug file wrongly denied.
- Taxonomy without a pick: three scenarios cover nearly all outcomes — honest, but hard to be Wrong. Oct 2026 adds a fourth move: "none of the analogies apply." Watch whether he ever commits to a branch, or whether "radically new" becomes the permanent exit.
- Historian's hammer: every buildout looks like railways eventually; base-rate risk if AI's revenue curve is genuinely faster. He is now aware of this — Chartbook 477 is partly a self-critique — but awareness is not a forecast.
- Moralizing gradient: "greed", "oligarchic", "Gotham" are rhetorically strong and analytically cheap; the political-economy turn risks crowding out the balance-sheet work that is his real edge.
- High-frequency publishing risks diluting the long-form authority that made Crashed; the Carbon book, long promised, is still not out.
- Energy claim is now carried by third-party data, not by him — if he has quietly demoted it, the file should know why.
Rich's take
(your synthesis here)
Delta log
HELD (9): capex-as-macro-event (Q2 GDP revised to 2.2% with AI 36–53% of growth; WSJ/Brookings $10.3T ≈ 3.6% GDP/yr); energy-binding (interconnection queues, $1.4T utility plans — third-party data, his own emphasis has moved); revenue-gap centrality (Q1 2026 sales $25B vs depreciation $21B, his Jul note); AI-inflation (PPI +5.4% y/y Aug; chipflation); prediction 1 macro-consequences; prediction 2 energy-binding; all three revealed behaviors (publishing cadence, no positions, discourse-moving — the last intensified via FT op-ed).
DRIFTED (1): "Refuses to pick — the taxonomy IS the position." Between Jul and Oct 2026 he moved from a neutral three-branch frame to (a) a strong lean toward the railway/financial-melt branch in tone ("Strangelove mashed up with Gekko and a dash of The Big Short"; "junk bonds and jerry-rigged financial engineering") and (b) an explicit argument against using historical analogy at all (Chartbook 477, 3 Oct; FT, 2 Oct). Not a formal pick, so not Wrong — but the taxonomy is no longer his position; "think the radically new" is.
WRONG (1): "No disclosed AI-tool practice (checked through Jul 2026)." Chartbook 385 (16 May 2025) documents extended hands-on use of ChatGPT and DeepSeek. Our miss, not his change; the weighting rule survives with a smaller discount.
OPEN (1): buildout → bubble → correction: Nasdaq-100 −10% correction (Jul 28–31) and NVDA −18% from May peak are below the 20% wake trigger and the 30% grading bar.
ADDED: political-economy rupture / oligarchic-arms-race claim; K-shaped-economy claim; regulatory-capture hypothesis; credit-side pop prediction with coverage-ratio/CDS checks; midterm check (4 Nov 2026); FT as a new outlet; incentive map re-verified clean. Tier moved semiannual → quarterly: eight-plus AI pieces and one frame shift in 40 days is too fast for a six-month cycle; the core balance-sheet claims are stable, the framing is not. Wake triggers widened to include a formal branch pick and any FT/FP AI column.
Template lesson: "no disclosed X" claims need a dated search of the person's own archive, not just the review window. Logged to learnings via run report.
HELD (strengthened): capex-as-macro-event (now visible in PPI +27% electronics, trade flows, Korea $576B); energy-binding (US utility $1.4T plans); revenue-gap centrality.
DRIFTED (his own, updated honestly): Feb's "not in the macro data yet" — overtaken by his own July data; calibration credit, not a strike.
WRONG: none found this pass.
ADDED: three-bubble taxonomy (Feb 2026) with per-branch checks; AI-inflation claim (Jul 2026); discipline facet graded "non-technical, no technical bet" with explicit weighting rule; empathy trio; incentive map; layer_focus=infra; decay=semiannual (framework moves slowly even though publishing is fast); wake triggers set.
Template lesson: v1 predictions were "TBD" without observables — the falsifiable-check column forced real tests. Logged to learnings.
Sources
- Tooze, primary (2026): Chartbook 477 — Against historic normalization (3 Oct) · FT op-ed, "No, AI is not similar to the Manhattan Project" (2 Oct; via FT Opinion editor) · Top Links 1239 — Biggest economic bet in US history (29 Sep) · FP — Can AI Be Regulated? (24 Sep, with Abadi) · Top Links 1227 — AI vs Wall Street and the White House (17 Sep) · Top Links 1218 — AI coming to dominate coding (8 Sep) · Top Links 1159 — revenue caught up with depreciation (11 Jul) · Chartbook 456 — MAGA meets AI (5 Jul) · Chartbook 455 — AI boom and China's walled-in wealth (1 Jul) · Chartbook — AI-inflation (1 Jul) · Top Links — data-centers driving inflation (26 Jun) · FP — K-shaped economy (12 May) · FP — Is This What an Economic Bubble Looks Like? (13 Feb)
- Tooze, primary (2025): Chartbook — Baskets of tech CDS (30 Dec 2025) · Top Links — The $3 trillion bet on AI (20 Sep 2025) · Chartbook 385 — Red AI: ChatGPT and DeepSeek hands-on (16 May 2025)
- Tooze, reported: Futurism — trillionaire remarks from Ones and Tooze (1 Aug 2026) · FP author page (column list, checked 3 Oct 2026) · Columbia European Institute (affiliation check)
- Market-level data: Q2 2026 GDP final 2.2%, AI contribution (30 Sep 2026) · ING — tech capex 44%/36% of Q2 growth · BLS PPI Aug 2026 (10 Sep 2026) · Axios — chipflation (11 Aug 2026) · WSJ/Brookings $10.3T build-out (24 Sep 2026, secondary) · NBC — Nasdaq-100 correction (Jul 2026) · Fortune — chip slump (28 Jul 2026) · NVDA −18% from May record (Sep 2026) · Goldman — ~$500B AI debt issuance 2026 · Utility $1.4T plans (secondary) · Interconnection waits (secondary)
- Carried from Aug 2026 file: BloombergNEF & utility-capex reporting, 2026 · Deutsche Bank Research Institute electronics PPI (May 2026) · prior file archived
_archives/adam-tooze.html.bak-20260824and_archives/adam-tooze.html.bak-20261003
revalidated v2 2026-10-03 · CSS v2→v2.5 · backup: _archives/adam-tooze.html.bak-20261003 · converts-from: adam-tooze.html v2 (2026-08-24) · Singularity 🧠 / People