Big Tech, AI Power Constraints, and the Specter of 1929

The three initial theses are maintained, but with increased robustness:

Quality of Big Tech Profits: The non-economic share of the aggregate net profit of the five largest tech companies is estimated at 42% in the central scenario (range 35–52%), instead of the single point estimate of 48%. The revision accounts for the sensitivity of EPQS weights, the treatment of depreciation extensions, and a dynamic circularity coefficient.

Energy Constraint of AI: The massive deployment of data centers requires, in the central scenario, 55 to 60 nuclear reactors of 1.1 GW by 2030 (weighted range 24–103 reactors depending on scenarios). The expression “physically impossible” is replaced by “severely constrained and delayed,” except in the extreme scenario where the 2030 horizon becomes unrealistic.

1929-type Valuation: The central Shiller CAPE is 42.5 (range 41–44), the equity risk premium is -1.85% (range -2.2 to -1.0). The probability of a drawdown of the S&P 500 greater than 20% over 12 months is estimated at 55% (range 48–63%).

Unchanged strategic recommendation: Underweight US growth stocks, overweight nuclear energy, utilities, short-term bonds and gold; hedge with asymmetric options; issue SPCI and SPQI indices as priority products; use SVRI as a regime signal and not as a permanent exposure.

THESIS 1. QUALITY OF BIG TECH PROFITS: ROBUST ESTIMATION

Indicators and formulas: Cash Conversion Score (CCS = FCF reported / NI reported). Accruals Ratio (AR = NI – CFO / Avg Total Assets). Stock-Based Compensation Ratio (SBC/NI = SBC / NI). Economic Value Added (EVA = NOPAT – WACC * Capital). Dynamic Circular Revenue Coefficient (αt = α_0 + sum of w{j,t} * ΔR_interco,j,t / R_total,t). Economic Profit Quality Score (EPQS v1.1) = w1 * (FCF_adj / NI_reported) + w2 * (1 – |AR|) + w3 * (1 – SBC_NI) – α_t – β_amort. Parameters include w1=0.40, w2=0.30, w3=0.30, FCF_adj = FCF_reported – circular CFO – incremental maintenance capex, and β is the effect of extended depreciation. Aggregate calibration (low/central/high): FCF/NI declared: 0.91/0.87/0.82. Accruals/Avg Assets: 0.07/0.10/0.13. SBC/NI: 0.10/0.12/0.15. Circularity α: 0.05/0.08/0.12. Amortization extensions β: 0.08/0.12/0.15. EPQS score: 0.65/0.58/0.48. Non-economic share: 35%/42%/52%.

THESIS 2. ENERGY CONSTRAINT OF AI: REVISED MODEL

Energy demand: E_AI = sum of N_i * P_i * U_i * 8760 * PUE_i * η_eff,t. The efficiency factor η_eff,t decreases from 1.0 to 0.80 between 2026 and 2030. Reactor count: N_reactors = ΔE_DC / (η_grid * 8760 * CF * P_avg) where η_grid=0.95, CF=0.90, P_avg=1.1 GW. Each reactor provides 8.231 TWh annually. Scenarios: Low/gas pivot: 200-300 TWh increase, 24-36 reactors (30% probability). Central: 400-550 TWh, 49-67 reactors (50% probability). High/extreme: 650-850 TWh, 79-103 reactors (20% probability). Expected number of reactors: ~53, rounded to 55-60 in central. Constraints: Gen III+ construction timelines (10-15 years). SMR requirement (300-400 units, prototypes only). Interconnection queue (2.6 TW, 5-7 year delays). Transformer lead times (2-4 years). Gas turbine supply constraints. Uranium structural deficit (2 years of commercial stocks).

THESIS 3. 1929-TYPE VALUATION. ESTIMATION WITH INTERVALS

Valuation indicators: Shiller CAPE: 42.5 (range 41-44). Buffett Indicator: 2.10 (1.95-2.20). Equity Risk Premium: -1.85% (-2.2 to -1.0%). Top 10 S&P concentration: 35% (32-37%). Monte Carlo simulation (100,000 trajectories): S&P expected return 12 months: -12.5% (-18.0 to -7.0%). Probability of drawdown >20%: 55% (48-63%). CVaR 95%: -34.2%. Expected shortfall 97.5%: -41.8%. Conditional probability of drawdown >20% given CAPE > 40: 55-65%.

GAME THEORY

The hyperscalers situation represents a prisoner’s dilemma, where competition drives the Nash equilibrium to build, justifying overinvestment individually. AI capex signals technological credibility. The underinvestment in grid capacities represents an equilibrium paradox.

The quantitative models currently indicate a strong risk-off regime, with Hidden Markov Model (HMM) filtered probabilities showing a 67% likelihood of a Risk-Off/Bubble state, versus 12% for Risk-On and 21% for Neutral. A GARCH-X model, incorporating an energy shock parameter (δ=0.02), estimates high implied volatilities of 22% for the S&P 500 and 35% for the Nasdaq 100. A quantum-classical hybrid optimization framework is used for portfolio construction, minimizing a combined objective of risk, return, sparsity, and cardinality constraints, with a net exposure limit of 1.5.

The strategic probabilistic allocation targets a portfolio with a central weight of 35% in US 2-year bonds, 20% in US Utilities, 15% in Uranium/Nuclear, 15% in Gold, 20% in Short Nasdaq 100, and a -5% cash/T-Bills position. This results in an ex-ante Sharpe ratio of 0.68 (range 0.60–0.75), a target volatility of 8%, and a maximum expected 12-month drawdown of -12%. Key liquidity pools include futures on E-mini S&P 500/Nasdaq 100, equity and VIX options, DP for uranium and utilities, Treasury futures, uranium equities/ETF (URNM) with squeeze risk, US electricity futures, and crypto perpetual swaps.

Institutional flows suggest potential CTAs liquidating 2.4 million long contracts if the S&P 500 falls below 5,200, pension funds rebalancing towards bonds, and hedge funds de-grossing megacap tech while accumulating uranium/nuclear assets.

The estimated 12-month Sharpe ratios for individual assets range from 0.15 (Short Nasdaq) to 0.60 (US 2-year bonds), with the full portfolio achieving a 0.68 Sharpe. Version 1.1 of the analysis reinforces the core thesis that quantitative finance, applied to AI’s energy constraints, validates a 1929-style market risk scenario, but with robust ranges and weighted scenarios. It highlights that 42% of Big Tech profits are non-economic, the central energy need is for 55-60 nuclear reactors, and there is a 55% probability of a S&P 500 drawdown exceeding 20%.

Recommended actions include underweighting US growth equities, overweighting uranium, utilities, short-term bonds, and gold, hedging with asymmetric options, issuing the SPCI and SPQI indices, using the SVRI as a regime signal, and deploying a hardened OKX strategy with a kill switch.

SOURCES

1. Cambridge Analytica (2016). Behavioral Microtargeting Methodology.
2. WEF Davos (2026). Global Risk Report.
3. BIS Basel (2026). Annual Economic Report.
4. Heston, S. (1993). Closed-Form Solution for Options with Stochastic Volatility.
5. Selten, R. (1975). Perfectness Concept for Equilibrium Points in Extensive Games.
6. Spence, M. (1973). Job Market Signaling.
7. Markowitz, H. (1952). Portfolio Selection.
8. Harry Dent, H.S. Spending Wave Model.
9. Amihud, Y. (2002). Illiquidity and Stock Returns.
10. Avellaneda, M. & Stoikov, S. (2008). High-Frequency Trading in a Limit Order Book.
11. Glassnode, CryptoQuant, Santiment, Coinglass, Arkham, Dune Analytics.
12. CFTC (2026). Commitments of Traders Reports.
13. Planet Labs (2026). Satellite Imagery and AIS Detection.
14. SaaS Capital, Bessemer, OpenView Partners (2026). SaaS Metrics Benchmarks.