On a chilly October morning in 1887, in his Paris offices, a metals merchant named Pierre-Eugène Secrétan sat down to write a letter to the directors of Europe's largest bank.

His proposal was simple. The world, he argued, was about to run out of copper. He wanted to own most of what remained.

Secrétan ran the Société industrielle et commerciale des métaux, the biggest copper-fabricating firm on the continent. For five years he had watched the price of copper slide - £60 a ton, then £50, then £39 - while his order book grew fatter than anyone in the industry could remember. Every week brought new contracts: wire for Edison-style central stations sprouting in Berlin and Milan, cable for the arc-light systems lining the Champs-Élysées, bus bars and dynamo windings for the electric streetcars that were suddenly everywhere. Secrétan saw what his competitors didn't. Copper prices were falling only because traders hadn't yet understood what electricity was about to demand of the world.

The Comptoir d'Escompte, one of France's four great banks, extended him an initial line of £2.5 million, a staggering sum for a single trade. The Rothschilds followed. So did Crédit Lyonnais and the Banque de Paris et des Pays-Bas. With their backing, Secrétan sent agents to the copper mines of Montana, Spain, Chile, and Lake Superior and signed three-year exclusive purchase contracts with thirty-seven producers. Within eight weeks his syndicate controlled roughly 80 percent of the world's new copper supply. The London Metal Exchange price doubled: £39 in early October, £80 by Christmas.

The copper writers of the London papers called it a triumph. The electricians called it a catastrophe.

Because here is what was actually happening in the last weeks of 1887: Thomas Edison, George Westinghouse and Elihu Thomson were in an arms race to wire the world's cities, and the single biggest line item in every one of their plans was copper. A central generating station in 1885 might budget more than half its capital outlay for copper wire alone. Edison, who thought in copper the way some men think in money, had already spent two years quietly re-engineering his distribution system to use less of it. His "feeder and main" layout - sub-trunks connected in rings to give every bulb multiple paths back to the generator - stripped enormous quantities of wire out of his distribution system. His three-wire system, patented on 20 March 1883, cut the copper needed in his conductors by a further sixty per cent, on Edison's own accounting. By the time Secrétan was buying Montana by the wagonload, the engineers were making his cartel obsolete one transformer at a time.

Secrétan didn't see it. He saw only the order book, and the order book kept growing. Every few weeks his agents had to buy more to hold the price, because the producers he had locked in were opening new mines as fast as they could raise the capital - Butte, Montana went from a dusty silver camp to the leading copper-producing district in the United States in roughly half a decade. The syndicate was trying to drink a river. By early 1889 it was sitting on tens of thousands of tons of physical copper, much of it stacked in warehouses along the Channel ports. The carrying cost alone was ruinous.

On 5 March 1889, the corner broke. Copper on the LME collapsed from near £100 to around £35. That evening, Eugène Denfert-Rochereau, the chief executive of the Comptoir d'Escompte and the man who had authorised the first £2.5 million, walked into a private room and shot himself.

The Comptoir faced ruin the next morning. A run began; depositors queued down the boulevard. What saved it is often described as one of the first modern bank bailouts in European history: the Banque de France, pressured by the finance ministry, organised an emergency guarantee syndicate of the other Paris banks - the same Rothschilds and Crédit Lyonnais who had ridden Secrétan's trade up - to backstop the Comptoir's obligations. The bank was reconstituted as the Comptoir national d'escompte, a direct ancestor of today's BNP Paribas. Secrétan himself died bankrupt. The copper kept flowing. The lights kept coming on.

Now, in April 2026, we are living inside a structurally identical story, and almost nobody is naming it.

The International Energy Agency projects global data-centre electricity use will hit roughly 1,000 terawatt-hours this year - the consumption of Japan, a country of 125 million people, routed into a few thousand buildings running AI. The US grid is the bottleneck, and the bottleneck inside the bottleneck is a single piece of industrial equipment: the high-voltage transformer. Lead times that were 24 to 30 months in 2019 now stretch as long as five years. Fortune reported last month that roughly half of planned US data-centre builds have been delayed or cancelled. Utilities in Virginia, Texas and Arizona are quoting grid connection dates in the 2030s for hyperscale sites that corporate roadmaps had slated for 2027.

And so the same instinct that seized Paris in 1887 is seizing the capital markets today. Hyperscalers are signing multi-decade power purchase agreements for every available megawatt - solar, gas, geothermal, and now, strikingly, nuclear. Microsoft has contracted to restart Three Mile Island. Amazon and Google have each signed large commitments with small-modular-reactor developers whose first units are not yet built. Private equity is rolling up transformer manufacturers. Copper itself, the same metal that broke Secrétan, is trading near record prices, and Wood Mackenzie now warns that the electrification build-out will create structural shortages well into the 2030s. The scramble for the scarce input is back.

What is different, and what the copper story should whisper in the ear of every AI executive right now, is that the real solution to a commodity bottleneck almost never comes from controlling the commodity. It comes from the engineers. Earlier this month researchers at UC San Diego published a new chip architecture that reportedly cuts AI energy use by up to 100x while improving accuracy. Nvidia's latest parts deliver roughly an order of magnitude more inference per watt than the chips that trained GPT-4 two years ago. Liquid-cooled racks have doubled the compute you can fit behind a given transformer. This is Edison's three-wire system, arriving on a different schedule.

If the pattern holds - and patterns like this have held remarkably well since the first factory smokestack - the people who get rich from the AI power crunch will not be the ones who cornered the power. They will be the ones who made less of it necessary. The speculators will end up, as speculators usually do, sitting on warehouses full of something the engineers quietly made cheap again.

Secrétan died believing the world had run out of copper. It hadn't. It had just temporarily run out of cleverness, and cleverness was restored faster than he could liquidate his inventory.

The grid of 1889 was built on copper. The grid of 2026 is being built on transformers and training silicon. The metal changes. The story doesn't.