What this page is and isn’t
Most catastrophic AI scenarios do not kill anyone directly. The scenario literature — nuclear winter modeling, engineered pandemics, infrastructure collapse — converges on an uncomfortable point: the mechanism of mass death, in nearly every branch, is hunger.
This page separates two kinds of claims. The structure of the modern food system (inventories, concentration, fertilizer dependence) is documented and observed. The cascade sequence — how fast shelves actually empty when the inputs stop — is part documented emergency behavior and part scenario analysis, and this page labels which is which. Nothing here is a preparation instruction; the household-level companion is water, power, and food security.
The system runs on tight margins
Modern food supply is a just-in-time system. Supermarkets typically hold roughly three to five days of stock at normal sales rates; the system is designed to be restocked continuously, not to store. Food processing is heavily concentrated: in the United States, a small number of very large plants handle the majority of beef and pork slaughter, so the loss of a handful of facilities idles a large share of national capacity — a concentration USDA’s Economic Research Service documents in its cattle and beef sector analysis (USDA ERS, cattle and beef sector) and that CISA treats as a critical-infrastructure dependency (CISA, Food and Agriculture Sector).
Three more dependencies sit underneath:
- Synthetic nitrogen fertilizer. Roughly half the world’s population is fed by calories made possible by the Haber-Bosch process. Natural gas is its main feedstock, which is why fertilizer prices track energy prices — and why the 2021–2022 energy shock became a food-price shock within months.
- Refrigeration. The cold chain handles most meat, dairy, and produce from plant or port to shelf. It requires uninterrupted power at thousands of nodes.
- Diesel. Planting, harvesting, trucking, and most farm machinery run on it. It has no near-term substitute at scale.
None of these is exotic. All of them fail together if the wrong two or three inputs stop at once.
The cascade, step by step
Scenario analysis. Suppose regional power is out for weeks. Each step is observed individually in emergencies; the full sequence at national scale is the modeled case, not a documented event.
- Power out. Refrigerated warehouses and stores begin warming within hours.
- Cold chain lost. Perishables become waste within days even where the food is physically present.
- Processing halted. Slaughterhouses, dairies, grain mills, and packaging plants stop with their power and their sanitation water; live animals back up on farms.
- Distribution narrows. Fuel priority goes to emergency services; trucking thins; urban stores restock slowly or not at all.
- Shelves empty in days, not weeks. Emergency planners assume grocery stock depletes fast under disrupted resupply; the FAO’s food-security frameworks treat access and logistics, not just production, as the binding constraint in crises (FAO, State of Food Security and Nutrition in the World).
Panic buying accelerates every step. Disaster studies consistently find that consumer hoarding converts a supply disruption into an immediate local shortage, amplifying the very scarcity it responds to. The shelves empty because everyone correctly predicts they will empty — a self-fulfilling run on a system with no buffer to absorb it.
Rehearsals: partial cascades we have already watched
COVID-19 (2020). A demand shock, not a supply failure — and still, meatpacking plants closing for outbreaks constrained US meat supply within weeks. CDC documented tens of thousands of cases among meat and poultry plant workers, and the closures showed how much national capacity lives in how few buildings (CDC MMWR, COVID-19 among meat processing workers).
The 2021–2022 fertilizer and price crisis. Natural gas prices roughly tripled; fertilizer prices followed; IFPRI estimated that the fertilizer-price surge alone added several percentage points to global food price inflation and pushed tens of millions toward food insecurity (IFPRI, high fertilizer prices and food insecurity).
Sri Lanka, 2022. A government ban on synthetic fertilizer — intended as a rapid transition to organic farming — cut crop yields sharply within a single growing season. Rice and tea production fell, foreign exchange earnings dropped, and the country slid into its worst economic crisis on record, with food inflation over 90% at the peak and a government that fell. IFPRI analyzed the ban as the shock’s origin (IFPRI, organic agriculture in Sri Lanka); the World Bank documents the depth of the crisis that followed (World Bank, Sri Lanka overview).
Each rehearsal stopped short of national famine — because the failures were regional, the rest of the world kept producing, and logistics recovered. The extinction-adjacent scenario removes exactly those stops.
Why food is the extinction hinge
A population can absorb a 10% production shock. Historically it does — through prices, substitution, and imports — painfully but survivably.
It cannot absorb 60% for two consecutive years.
That is the arithmetic behind nuclear-winter famine modeling. Robock and colleagues’ simulations of a full-scale United States–Russia nuclear exchange — soot injected into the stratosphere, sunlight and temperatures falling across major growing regions — project crop declines of that order or worse, with global calorie production falling far below what remains after feeding livestock (Robock et al., 2008). Later work in Nature Food modeled the food system under such a scenario and found mass famine in almost every country, including ones untouched by war, with several billion deaths from food shortage as a plausible central estimate (Xia et al., 2022, Nature Food).
The same logic transfers to non-nuclear branches. An engineered pandemic (see engineered pandemics and bioweapons risk) kills or incapacitates the workforce that farms, trucks, and staffs power plants. A crisis that breaks grid and fuel together (see AI, nuclear command, and crisis stability for one such path) does the same. In every branch, the dying happens in kitchens.
What breaks the cascade
Three things interrupt the sequence, and all three are policy choices rather than household ones:
- Strategic reserves of grain and fuel, sized for multi-year disruption, with distribution plans written in advance. Grain reserves exist in some countries; almost none are sized for a 60% shock.
- Decentralized production: regional grain, smaller processing, local food webs. Sri Lanka’s crisis and the COVID plant closures are the same lesson at different scales — concentration converts small failures into large ones.
- Priority logistics protocols: pre-agreed rules for keeping fuel, power, and transport flowing to food under stress. These are exercises and statutes, not improvisations.
This is where the manual’s chapters meet: averting the scenario (chapter 05 territory) and surviving it (chapter 06, including core household reserves) are different jobs. A household buffer buys weeks. Only policy buys seasons — which is why the cascade argument belongs in front of governments, not just in pantries.