AI, Markets, and the Economy
AI is changing organizations, markets, and the economy
Independent research on how AI changes organizations, markets, and the economy. Public data and primary sources, with stated assumptions, and published code when valuable to the examiner.
Stated magnitudes across 22 of the 35 ledger entries, 2025 to 2026-08-17. Announced, not deployed.
Every institutional AI action, dated
35 entries · 2025-01-03 → 2026-08-17
01Institutional AI Investment
Open the imprint →- Analysis · No. 3
The Substation
The buildout isn't short of money or chips. It's short of megawatts.
474 GWin the Texas queue, Aug 2026 - Analysis · No. 2
Sovereign Compute
Governments have put more than a trillion dollars of AI infrastructure under their names since January 2025.
0.3 of 10 GWStargate live vs pledged, Apr 2026 - Analysis · No. 1
The Capex Ledger
Five companies will spend about $800 billion on equipment this year, more than they make in operating cash.
$802B2026 capex guidance, five companies, mixed basis
02Quantitative Markets & Institutions Lab
Source on GitHub Open the lab →Portfolio construction · Estimation risk
Portfolio optimization under estimation error
The fitted max-Sharpe portfolio weakened materially on later data.
1.12MVO Sharpe · fitSharpe 1.12 → 0.44Market risk · Forecast calibration
Historical VaR under coverage and independence tests
Breaches arrived too often and in clusters; the forecast failed its calibration tests.
135Observed breaches135 breaches · 107 expectedDerivatives · Stochastic assumptions
Constant-volatility dynamics under empirical tests
Observed volatility and option prices contradicted the tested constant-volatility dynamics.
≈3%–93%Realized vol range≈3% → 93% realized volatilityEmpirical dossier · Measured relationship
Stock–bond correlation regimes and their portfolio consequences
One sign change, three portfolio consequences.
−0.35SPY–TLT · 2015–19−0.35 → +0.37 correlationFactor models · Product inference
ETF factor exposure and alpha after multiple testing
Advertised exposures appeared; no tested alpha survived multiple-testing correction.
0.34MTUM momentum β0 alphas at 5% FDRNumerical methods · Simulation error
Monte Carlo convergence, bias, and variance reduction
Sampling error followed the N⁻¹ᐟ² law; enough paths is defined by the decision's tolerance.
N⁻¹ᐟ²Error rateError ∝ N⁻¹ᐟ²
Quantitative model · Derivatives
Option Valuation Models
The Black-Scholes-Merton closed form and the Cox-Ross-Rubinstein lattice priced side by side, with every Greek computed live in the browser.
Quantitative model · Stochastic processes
Stochastic-Process Simulation
Five stochastic processes — GBM, jumps, Heston, mean reversion, regime switching — simulated live as 50 seeded one-year paths.
03Capital expenditure
01Amazon, Alphabet, Microsoft, Meta, and Oracle bought about $150 billion of property and equipment in fiscal 2023.
02Fiscal 2024 took the same five to about $226 billion. Roughly nine percent of it was funded by new debt.
03Fiscal 2025 reached about $381 billion — more than double the 2023 figure.
04Guidance for 2026 puts the five at about $802 billion, roughly a third of it funded by new debt — more than they generate in operating cash.
Capital expenditure, five companies
USD billions · cash purchases of property and equipment unless noted
Capital expenditure data
USD billions. Historical columns show cash purchases of property and equipment. The guidance column uses different bases: Microsoft and Meta include finance leases, Oracle is net of customer reimbursement, and Microsoft and Oracle refer to fiscal 2027. Guidance is not reported expenditure.
| Company | FY2023 | FY2024 | FY2025 | 2026 guidance |
|---|---|---|---|---|
| Amazon | 52.7 | 83.0 | 131.8 | 220.0 |
| Alphabet | 32.3 | 52.5 | 91.4 | 200.0 |
| Microsoft | 28.1 | 44.5 | 64.6 | 175.0 |
| Meta | 28.1 | 39.2 | 72.2 | 137.0 |
| Oracle | 8.7 | 6.9 | 21.2 | 70.0 |
04Papers
All papers →Research Paper
AI, Power, and Strategic Infrastructure
The U.S. Grid Buildout, Global Competition, and an Investor Framework
Analysis of AI electricity demand, U.S. grid constraints, U.S.–China competition, and the investment risks across power infrastructure.
05Ledger, most recent
All entries →| Date | Actor | Action | Magnitude | Source |
|---|---|---|---|---|
| 2026-08-17 | NvidiaChipmaker | Nvidia agreed to provide up to $105 billion in financing for an OpenAI Ohio data center with 4.25 GW initial capacity, built by SB Energy. | Up to $105B; 4.25 GW | Bloomberg |
| 2026-08-02 | European Commission (AI Office)Regulator | AI Act transparency and high-risk obligations became enforceable, with penalties up to €35 million or 7% of global turnover for banned practices. | Fines to €35M or 7% turnover | European Commission |
| 2026-07-30 | European CommissionSovereign | The Commission launched a tender for up to seven AI Gigafactories backed by up to €10 billion public funding, targeting over €30 billion investment. | €10B public; €30B total | European Commission |
| 2026-07-30 | AmazonPlatform | Amazon raised 2026 capital expenditure guidance to about $220 billion from about $200 billion, attributed by CEO Andy Jassy to higher memory costs; second-quarter cash capex was $53.1 billion. | ~$220B capex 2026 | CNBC |
| 2026-07-22 | AlphabetPlatform | Alphabet raised 2026 capital expenditure guidance to $195–205 billion from $180–190 billion after quarterly capex of $44.9 billion. | $195–205B capex 2026 | SEC (Alphabet Form 8-K) |
| 2026-07-13 | MetaPlatform | Meta said its Hyperion data center in Louisiana will expand to 5 GW at a cost above $50 billion, up from about $27 billion. | >$50B; 5 GW | CNBC |
Independent research by Kyle Wisniewski. For information and education; not personalized investment advice, and not the position of any employer or university.