History does not repeat, but it has structural patterns. After studying five eras — the California Gold Rush, the Great Depression, the Post-WWII Tech Boom, Nuclear Deterrence, and the current AI-driven economy — three patterns recur across every transition. They describe who wins, who loses, and why the actors best positioned for the last era are usually the least prepared for the next.

Supply Side Always Wins

The California Gold Rush drew three hundred thousand people to the Sierra Nevada. Almost none got rich from the gold itself. The people who built enduring wealth — Levi Strauss, John Studebaker, Henry Wells and William Fargo, the Big Four railroad barons — were selling jeans, wagons, banking, and rail transport to the miners. Sutter and Marshall, the men who discovered the gold and knew more about extraction than anyone, both died broke.¹

This pattern — the shovel seller’s advantage — follows from a structural asymmetry. Early entrants in any gold rush compete on a winner-take-all basis; most lose. Suppliers of inputs — tools, compute, logistics, training — face diversified demand that grows with the boom and does not collapse when individual prospectors fail. James J. Hill built the Great Northern Railway without a single government land grant. His customers were thousands of independent farmers and miners. His revenue did not depend on any one of them succeeding.²

The same logic applies to NVIDIA’s position in the AI boom. OpenAI, Anthropic, Google, Meta, and a thousand startups all buy NVIDIA GPUs. If one fails, the others still need chips. The compute vendor faces diversified demand; the model-makers face winner-take-all competition.³ The question of who will win the AI race is less useful than the question of where second-order bottlenecks will appear. In the Gold Rush they were jeans, wagons, and banking. In the 1990s they were routers, databases, and logistics. Today they are compute, training data, verification tools, and regulatory compliance tooling. The picks and shovels outlast the miners every time.

Near-Misses as Institutional Data

Near-misses teach differently than successes. Success admits multiple interpretations. Failure tends to have a single verifiable cause. When a nuclear war does not happen, it is hard to prove deterrence worked rather than both sides being lucky. When the Able Archer 83 NATO exercise nearly triggered a Soviet launch alert, the lesson was unmistakable: military exercises with nuclear-release components can be misread as the real thing.⁴

Each nuclear near-miss produced institutional reform. The Cuban Missile Crisis (1962) produced the Moscow-Washington hotline within a year. The 1979 NORAD false alarm — a training tape loaded into the live early-warning system — produced improved verification protocols. Able Archer 83 produced Risk Reduction Centers and exercise transparency agreements. The Stanislav Petrov false alarm (1983) produced redundant early-warning verification chains.⁵ In each case, the near-miss was followed by structural change.

The pattern has a weakness: reforms are episodic, not continuous. Institutional memory decays as bureaucrats rotate out. By the time Able Archer occurred, the lessons of the Cuban Missile Crisis had been diluted by twenty years of turnover.⁶ The last major nuclear near-miss was 1983 — over forty years ago. The architects of deterrence who lived through those events are effectively gone. Meanwhile, near-misses can produce overconfidence. The 2023 regional banking crisis was contained by swift emergency lending, but the success of that response may have created moral hazard. Bankers who watched the Fed absorb SVB’s failure may take larger risks next time.⁷ Near-misses are valuable data, but only if you ask whether the lucky outcome was due to skill.

The Bind of Expertise

The people most damaged by a transition are the people who were best at the old system. Expertise in a previous equilibrium becomes a liability when the rules change, and deeper expertise means a harder fall.

The bind has three components. Depth of investment — twenty years mastering a skill set about to become irrelevant. Cognitive framing — seeing the new world through the lens of the old one, interpreting the internet as a better fax machine. Identity attachment — self-concept built around being the expert, so admitting the rules have changed feels like admitting a career was a mistake.⁸

The examples are consistent. Sutter and Marshall were the world’s best gold extraction experts, and that expertise kept them focused on mining while the money moved to logistics. Studebaker dominated wagon manufacturing, entered automobiles early (1902 electric, 1904 gasoline), but never escaped its carriage-maker DNA — it operated like a wagon company that happened to make cars, and failed while less experienced upstarts consolidated the industry.⁹ IBM dominated mainframes with the System/360, the most successful product in computing history. That expertise made it nearly blind to personal computing; the PC launched in 1981 but was managed by mainframe executives who could not understand why anyone would want a computer on their desk. By the time IBM understood, the architecture belonged to Intel and Microsoft.¹⁰

What breaks the bind? Three traits appear in actors who successfully navigate regime changes. Identity separation — treating expertise as a tool, not an identity; Levi Strauss was a dry goods wholesaler before the Gold Rush, not a miner trying to pivot. Heterogeneous information sources — talking to people outside the industry and paying attention to anomalies that do not fit the expert narrative. And willingness to cannibalize — the move that separates Bezos (who cannibalized Amazon’s retail margins to build AWS) from Borders (whose book-retailing expertise prevented seeing the internet as anything but a threat). The people who survive regime changes are the ones most willing to learn the new rules from scratch.