THE LONG READ · OCTOBER 31, 2026

If AI is the new oil, who is Rockefeller?

Whale oil, Titusville, Standard Oil, the railroads, Spindletop, OPEC, and what a century of fights over a cheap, world-changing fuel says about chips, power, models, and who gets paid when intelligence gets cheap.

~40 MIN READAN ANALOGY, NOT A PROPHECYFIGURES AS OF OCT 31, 2026NOT INVESTMENT ADVICE

If AI’s best historical parallel is oil, how does the story play out? Who gets rich, who gets squeezed, and who ends up paying for the whole thing?

It’s a good question, and the usual answer is too lazy. “Data is the new oil” has been a slogan for years. The comparison worth making is different. It isn’t about data. It’s about a cheap, general-purpose input that keeps getting cheaper, gets used in ways nobody predicted, and sets off a long, violent argument about who gets to collect the toll.

Oil is that story. It started as a lighting business that nearly drowned in its own oversupply. It made one man the richest person in the world, not by drilling, but by controlling the narrow middle of the chain and bargaining with the railroads. It produced the first great antitrust case, a century of geopolitics, and a textbook example of the Jevons paradox: the cheaper energy got, the more of it the world used.

The hunches going in are obvious ones. Elon Musk as Rockefeller. Jensen Huang as Vanderbilt, the railroad baron collecting a toll on everything that moves. Or maybe Musk is actually Vanderbilt. And since AI isn’t a natural resource but something manufactured by many different players, perhaps a lot of them are better understood as landholders and mineral-rights owners than as oilmen.

This essay takes those hunches seriously without following the analogy slavishly. Part One tells the oil story with a focus on where the money went. Part Two maps it onto AI and tests several castings. Part Three plays out the implications.

The short version: in oil, value went to whoever held the scarcest step. That step kept moving.

PART ONEThe oil story

Before petroleum, there was the price of light

To understand oil you have to forget cars. For its first half-century, petroleum was not a transportation fuel. It was a lighting business. The product that mattered was kerosene, and the market it went after was darkness.

Before kerosene, the good light in America came from whales. New England’s whaling fleet was one of the great industries of the young country; at its peak around 1846 it ran roughly 735 ships (one contemporary count says 722) and, by one estimate, something like 70,000 workers. The premium product was sperm oil, which burned bright and clean. In the mid-1850s it sold for around $1.77 a gallon, an enormous sum for something you burned at night.

The economist William Nordhaus once did a wonderful bit of accounting: how many hours of work did it take to buy 1,000 lumen-hours of light, roughly the light of a bright bulb for an evening? With a whale-oil lamp between 1815 and 1845, about 4.9 hours of labor. With a tallow candle in 1830, about 3 hours. A sperm candle, nearly 7. Then came the kerosene lamp. By 1875–85, the same light cost about 0.23 hours, around 13 or 14 minutes of work. Light got something like twenty times cheaper in a generation.

Keep that number in your head, because it is the whole story in miniature. A new source of energy did not just replace an old one. It made a basic good so cheap that people used vastly more of it, and the fight over who got paid for that abundance shaped the next hundred years.

Titusville, 1859: the first glut

“Rock oil” had seeped from the ground in western Pennsylvania for ages and was sold as patent medicine. What changed in 1859 was drilling for it, the way you’d drill for salt brine. Investors hired Edwin Drake, a former railroad conductor with an honorary “Colonel” for credibility, to try it near Titusville. On August 27, 1859, at 69 feet, the well hit oil.

What followed was not a gold rush so much as a stampede off a cliff. Crude went for roughly $20 a barrel in the last months of 1859. In 1860 it averaged $9.60. By the end of 1860 it was about $2. By the end of 1861 it was about 10 cents. Production went from around half a million barrels in 1860 to more than 2 million in 1861. Big flowing wells, like the Funk well in June 1861 and the Empire well later that year at 3,000-plus barrels a day, simply poured more oil into the valley than anyone could carry away.

At the bottom, an empty wooden barrel, which cost something like $1.75 to $2.00, was worth more than the oil you put in it. The scarce thing in the Oil Regions was not oil. It was the container, the wagon, the river barge, the railroad car. Hold onto that too.

The law made it worse. American courts applied what came to be called the rule of capture: oil belonged to whoever brought it to the surface. If your neighbor’s well and your well tapped the same underground pool, the oil went to whoever pumped fastest. So nobody could afford to wait for better prices. If you shut in your well, your neighbor would drain your oil through his. The rational move for each driller was to pump flat out, which guaranteed overproduction for all of them.

The boomtowns were the human face of this. Pithole, Pennsylvania, went from farmland to a city in months after the Frazier well came in during 1865; by September it had a reported 15,000 people (some claims go to 20,000). The Holmden Farm, where much of the town stood, sold for $2 million in September 1865. Then the wells declined, fires came through, the drillers moved on. The 1870 census counted 237 residents. The borough charter was annulled in 1877. In 1878 Venango County bought the Holmden Farm for $4.37 (some accounts say 1879). Two million dollars to four dollars and thirty-seven cents in about thirteen years.

The chokepoint was the refinery

Crude oil by itself was not much use. To make lamp oil you had to boil it, capture the right fraction, and treat it so it would not smoke, stink, or explode. Early refining was not hard to get into. A small still and a few workers could get you started, which is exactly why there were so many refiners and why the refining business, too, was brutally competitive in the 1860s.

John D. Rockefeller was a young Cleveland produce merchant, careful with money and obsessed with bookkeeping. In 1863 he went into refining with a chemist-minded partner, Samuel Andrews. In 1865 he bought out his other partners at auction and took control of the business. Over the next few years he added partners who mattered, above all Henry Flagler, who would turn out to be the deal-maker of the operation, and the financier Stephen Harkness.

On January 10, 1870, they incorporated the Standard Oil Company of Ohio with about $1 million of capital. By the commonly cited estimate, it already did roughly a tenth of American refining. Not a monopoly. Just the biggest, best-run refiner in a crowded, money-losing field.

Why refining? Rockefeller understood something the drillers did not. Producers were many and undisciplined. Consumers were many and scattered. In between, the oil had to pass through a much smaller number of refineries and a much smaller number of transport routes. That middle was where the leverage was. If you could control the narrow part of the hourglass, you could set terms for both ends.

The machine: scale, thrift, and a standard product

The Standard story is often told as pure villainy. It wasn’t. Much of its advantage was relentless, almost comically detailed cost-cutting.

Standard made its own barrels instead of buying them, at roughly half what outsiders paid (the exact numbers vary between historians). In the most famous story, Rockefeller watched cans of kerosene being sealed and asked the workers to try using fewer drops of solder on each one. Cut too many and the cans leaked; cut one and they held, and that became the new standard. The retellings disagree on the exact count, but it hardly matters: on millions of cans, a single drop was real money, and the story captures the culture.

It also used the whole barrel of crude. Rivals dumped what they could not sell. Standard learned to turn the leftovers into lubricants, paraffin wax, petroleum jelly and other by-products, each one a small extra margin that competitors didn’t earn. Ironically, one fraction it often had no use for was gasoline, a volatile nuisance that refiners frequently threw away. That detail will matter later.

And then there was the name. Kerosene from small, careless refiners was sometimes contaminated with lighter, more explosive fractions, and lamp fires were a real danger. “Standard” meant a consistent, safer product. In a market where buyers could not test what they were buying, a reliable brand was worth something.

The results showed up in prices. Sources differ on the exact numbers, but the shape is not in dispute: kerosene went from around 58 cents a gallon in 1865 to about 26 cents in 1870 and roughly 8 cents by the mid-1880s. Using figures drawn from Ron Chernow’s biography, Standard’s own price went from 23.5 cents in 1870 to 7.5 cents in 1890, and its processing cost per gallon of crude fell from 2.5 cents in 1880 to 1.5 cents in 1885. Rockefeller claimed that cheaper crude explained only about half of the price drop; the rest, he said, was Standard’s efficiency.

But efficiency alone doesn’t get you to 90 percent of an industry. For that, you needed the railroads.

The railroads: Vanderbilt, Scott, and Gould

In the 1860s and 1870s, oil moved east by rail. Three big “trunk lines” connected the oil country and the refining centers to New York and the eastern seaboard:

  • The New York Central, controlled by Cornelius Vanderbilt, “the Commodore,” who had made his first fortune in steamships and then assembled the Central system. Lines like the Lake Shore, running along Lake Erie through Cleveland, became part of the Vanderbilt network.
  • The Pennsylvania Railroad, run by Tom Scott, arguably the most powerful corporation in America at the time.
  • The Erie, the scandal-ridden road associated with Jay Gould and his allies, which had been looted and fought over in some of the most notorious financial battles of the age.

Railroads have a particular kind of economics. Building the line is enormously expensive. Running one more train is comparatively cheap. That means a railroad with empty cars will cut prices almost to the bone to fill them, and when several railroads serve the same route, rate wars break out. What every railroad craves is steady, predictable, high-volume freight.

Rockefeller and Flagler didn’t just ask for a discount. They offered what railroads wanted most: guaranteed volume.

The key deal came between 1868 and 1870 with General J. H. Devereux of the Lake Shore. According to Devereux’s own later affidavit, Flagler guaranteed sixty carloads of refined oil a day, every day, year-round. In return Standard got a through rate of $1.75 a barrel (35 cents to bring crude from the Oil Regions to Cleveland, plus $1.30 to send refined oil from Cleveland to New York) at a time when the nominal rate for refined oil from Cleveland to New York was $2.00. For the railroad, a guaranteed daily train was far cheaper to run than scattered small shipments. For Standard’s competitors, it meant paying more to ship the same product down the same tracks.

“It was equivalent to renting a railroad for their private use.”— IDA TARBELL ON THE LAKE SHORE ARRANGEMENT

Economists still argue about these rebates. Defenders say volume discounts reflected real cost savings. The legal scholar Daniel Crane, among others, argues that the structure of Standard’s rebates and especially its “drawbacks” looks less like cost savings and more like a tool for excluding rivals. The next episode makes that case hard to dismiss.

The South Improvement Company

In 1872 a group of refiners including Standard, working with the Pennsylvania, the New York Central and the Erie, set up a shell called the South Improvement Company. The idea was a cartel among the railroads and a favored group of refiners. The posted rate to ship crude to New York would be $2.56 a barrel. Members would get a $1.06 rebate on their own shipments. Then came the stunning part: members would also get a $1.06 “drawback” on every barrel shipped by non-members. Your competitors’ freight payments would literally flow into your pocket. The railroads would also hand members the waybills, the shipping details, of competitors’ shipments, a perfect intelligence feed on everyone else’s business. Oil men estimated the rebates were worth around $6 million a year.

When word leaked, the Oil Regions exploded. Producers held mass meetings, organized boycotts and embargoes of their own, and called it an “oil war.” By around March and April of 1872 the railroads backed away and cancelled the contracts. The South Improvement Company never actually operated.

But it had already done its job in Cleveland. In February and March 1872, while the scheme was hanging over everyone, Standard approached its Cleveland competitors and offered to buy them out, for cash or Standard stock. Most of them sold. In about six weeks, Standard absorbed 22 of its 26 Cleveland rivals. People in the industry called it the Cleveland Massacre. Rockefeller would always argue that he was rescuing an overbuilt industry from ruinous competition and that sellers who took Standard stock got rich. Many of the sellers remembered it as an offer they were not allowed to refuse.

When the railroad tried to refine

The most revealing fight of the era came in 1877, and it is the one most worth remembering for the AI story.

The Pennsylvania Railroad had a subsidiary, the Empire Transportation Company, run by Joseph Potts. Empire owned pipelines, tank cars and terminals, and it began moving into refining itself. In other words, the railroad, the owner of the infrastructure, decided it didn’t just want to carry the oil. It wanted to own the refining business that its biggest customer ran.

Standard’s response was to pull its freight off the Pennsylvania and shift it to the competing roads, while cutting kerosene prices in the markets where Empire’s refineries sold. The Pennsylvania was a giant, but it was also caught in a rate war and battered that summer by the great railroad strike of 1877. It blinked. In October 1877 Standard bought Empire’s pipelines, refineries and tank cars for $3.4 million.

The lesson: in that era, the customer who controlled the volume beat the infrastructure owner who tried to move up the chain. The railroad needed Standard’s freight more than Standard needed any single railroad.

Pipelines: routing around the rails

Railroads were not the only possible route. Short gathering pipelines had existed since the 1860s; the decisive moment came in 1879, when a group of independent producers and refiners built the Tidewater Pipe line, about 110 miles of six-inch pipe from Coryville, in the Bradford field, over the Alleghenies to Williamsport, Pennsylvania. They finished it in under ninety days. Oil arrived around June 4, 1879, and was loaded onto the Reading Railroad to reach the coast. It was the first long-distance crude pipeline, and it proved crude could move to the sea without paying tribute to the trunk lines.

Standard’s response was characteristic. It didn’t fight pipelines; it became the pipeline company. It built its own trunk lines and, by 1883, had taken a stake in and reached a settlement with Tidewater. The bypass did break the railroads’ leverage over crude. It did not break Standard. By around 1880, by the commonly cited figure, Standard controlled roughly 90 percent of American refining.

The Trust, Tarbell, and the breakup

Standard was by then dozens of companies in different states, and state laws made it awkward for a company in one state to own companies in another. So in January 1882 the shareholders handed their stock to nine trustees, who controlled the whole group. That was the Standard Oil Trust, and it gave the English language the word we still use for monopoly regulation: antitrust.

Congress passed the Sherman Antitrust Act in 1890. For years it had little bite. What changed the politics was a journalist.

Ida Tarbell grew up in the Oil Regions. Her father was one of the independent producers and businessmen hurt during the South Improvement era. Between 1902 and 1904 she published The History of the Standard Oil Company as a serial in McClure’s magazine. It was meticulous and devastating precisely because it didn’t need to exaggerate: the rebates, drawbacks and waybills, laid out in the company’s own paper. It turned Standard from a business story into a moral one.

On May 15, 1911, the Supreme Court upheld the government’s case and ordered Standard broken into 34 companies, under the “rule of reason” standard that still shapes antitrust law. The successors included Standard Oil of New Jersey (later Exxon), Socony (later Mobil), Standard of California (later Chevron), Standard of Indiana (Amoco) and Standard of Ohio (Sohio).

Here is the twist that every investor should remember. Rockefeller didn’t lose. He kept his proportional stakes in all 34 companies. Once they were separate, investors could value them individually, and the shares rose. Estimates vary, but his fortune grew to something on the order of $900 million by around 1913. The breakup made the founder richer.

And by 1911, Standard’s grip was already slipping. Its share of refining had gone from 88 percent in 1890 to roughly 64 percent. The market, new oil fields, and a new product were doing as much as the courts.

The gasoline moment

The single biggest threat to Standard’s core business didn’t come from a competitor or a regulator. It came from Thomas Edison. His Pearl Street power station opened in Manhattan in 1882, the same year the Trust was formed. Electric light was cleaner, safer and more convenient than any lamp, and as cities wired up, the kerosene illumination market that had built Standard started to erode.

Standard could have become a footnote, a great company that dominated a dying product, like the whalers. Instead, a different invention saved the industry. The internal combustion engine and then the automobile turned gasoline, that waste fraction refiners used to dump, into the most valuable product in the barrel. Henry Ford’s Model T arrived in 1908, and its price fell steadily as the assembly line took hold. Around 1911 (some sources say 1910), U.S. gasoline sales passed kerosene sales for the first time.

The original use case, lighting, was killed by a better technology. A new one, mobility, turned out to be vastly bigger. People didn’t burn more oil because their lamps got better. They burned more because oil let them do something entirely new.

Gushers, Baku, and the navy

At the same time, the supply map was blowing open. On January 10, 1901, a well at Spindletop, near Beaumont, Texas, blew in with a gusher estimated at about 100,000 barrels a day, more than anyone had seen. Spindletop gave rise to Gulf Oil, backed by Mellon money from Pittsburgh, and the Texas Company, later Texaco. Standard had dominated the Pennsylvania supply chain. It never controlled Texas the same way. The new oil came with new owners.

Abroad, the industry had already gone global. In Baku, on the Caspian, the Nobel brothers, Ludvig and Robert (siblings of Alfred, of dynamite and prize fame), built an integrated oil business in the 1870s. They were engineers first. They pioneered the oil tanker; their Zoroaster sailed in 1878, carrying oil in bulk rather than in barrels. The Rothschilds financed a railway from Baku to the Black Sea port of Batum to get Russian oil to world markets, and for a few years around the turn of the century Russia was the world’s largest oil producer. A London merchant named Marcus Samuel sent the tanker Murex through the Suez Canal in 1892 to sell Russian kerosene into Asia; that business became Shell. In 1907 Shell combined with Royal Dutch, under Henri Deterding, to create a true rival to Standard.

Then oil became a matter of state. Winston Churchill, First Lord of the Admiralty from 1911, pushed the Royal Navy to switch its ships from coal to oil, a faster, longer-range fuel that Britain, unlike coal, did not have at home. In June 1914, weeks before the First World War, the British government bought 51 percent of the Anglo-Persian Oil Company, the ancestor of BP, for about £2.2 million. A government had decided that the supply of a fuel was too important to leave to the market.

International oil then settled into the hands of seven companies the Italian oilman Enrico Mattei nicknamed the Seven Sisters: Jersey (Exxon), Socony (Mobil), Socal (Chevron), Texaco, Gulf, Anglo-Persian (BP) and Royal Dutch Shell. Through concessions across the Middle East, Venezuela and beyond, they set the terms for decades.

Too much oil, again: East Texas and the Railroad Commission

In 1930 wildcatters found the East Texas field, one of the largest ever discovered in the lower 48, and the rule of capture produced the same result it had produced in 1861. Everyone pumped as fast as they could. Crude fell from around $1.15 a barrel to about 10 cents in 1931. On August 17, 1931, Governor Ross Sterling declared martial law and sent in troops to shut in more than 1,600 wells.

Out of that chaos came a peculiar American institution. The Texas Railroad Commission, a body originally set up to regulate railroads, got the job of “proration”: setting how much each well could produce. Bootleg “hot oil” that exceeded the quotas kept leaking into interstate markets until the Connally Hot Oil Act of 1935 brought in federal enforcement. For decades after, the price of oil in America was effectively managed by a state commission in Austin limiting supply.

Note the cure for the rule of capture: not a smarter company, but a government body controlling permission to produce.

The owners take over

Through the middle of the century, the countries that sat on the oil pushed for a bigger cut. Venezuela won a 50/50 profit split in 1943; Saudi Arabia followed in 1950. Iran nationalized its oil industry in 1951. In September 1960, in Baghdad, a group of the major producing countries founded OPEC, explicitly modeling it on the Texas Railroad Commission. The Seven Sisters’ share of production in OPEC countries slid from 86 percent in 1963 to 77 percent in 1970.

Then came 1973. In October, during the Yom Kippur War, OPEC raised the posted price from about $3 a barrel to $5.12, and Arab producers embargoed the United States and others. On January 1, 1974, the posted price went to $11.65. Through the 1970s, producing countries nationalized their fields or took “participation” stakes; by 1980 Saudi Aramco was fully state-owned.

A century after Titusville, the rent finally went to the people who owned the ground. Not the private landowners of Pennsylvania, who had mostly collected modest royalties. The sovereigns, who controlled enough supply to withhold it.

Jevons, in practice

In 1865, the English economist William Stanley Jevons published The Coal Question. Britain was worried about running out of coal, and many people assumed more efficient engines would help by reducing consumption. Jevons said they had it backwards:

“It is wholly a confusion of ideas to suppose that the economical use of fuel is equivalent to a diminished consumption. The very contrary is the truth.”— W. S. JEVONS, THE COAL QUESTION (1865)

Oil is the Jevons paradox played out over a century. Kerosene made light roughly twenty times cheaper, and people lit their homes far longer and brighter. Refining got cheaper, kerosene fell to pennies, and consumption grew. Gasoline and cars made travel cheaper per mile, and people traveled more, lived farther from work, and built an entire economy around it. Every time a unit of energy got cheaper, the world found more ways to use it. Total consumption didn’t shrink. It exploded.

THE OIL STORY, COMPRESSEDDates and figures from the source notes; some estimates vary
YearWhat happened
~1846The American whaling fleet peaks at roughly 735 ships. Sperm oil is the premium lamp fuel.
1859Drake’s well, Titusville, Pennsylvania, strikes oil at 69 feet on August 27. Crude sells for about $20 a barrel.
1861Flowing wells flood the market. Crude falls to roughly 10 cents a barrel; the empty barrel is worth more than the oil in it.
1863–65Rockefeller enters refining in Cleveland with Samuel Andrews, then buys out his partners at auction.
1865Pithole, Pennsylvania, booms to a reported 15,000 people. By the 1870 census it has 237.
1868–70Flagler’s Lake Shore deal: a guaranteed 60 carloads a day in exchange for a secret through rate.
1870Standard Oil (Ohio) incorporated January 10 with about $1M of capital and roughly a tenth of U.S. refining.
1872South Improvement Company scheme with the three trunk railroads. The “Cleveland Massacre”: Standard absorbs 22 of 26 local rivals in about six weeks.
1873Panic of 1873, set off by Jay Cooke’s railroad financing.
1877Standard beats the Pennsylvania Railroad’s Empire Transportation Co. and buys its assets for $3.4M.
1879Tidewater’s ~110-mile crude pipeline reaches Williamsport. Pipelines break the railroads’ hold on crude.
~1880Standard controls roughly 90% of U.S. refining.
1882Standard Oil Trust formed. Edison’s Pearl Street station opens the electric-light era.
1890Sherman Antitrust Act.
1901Spindletop gushes near Beaumont, Texas. Gulf and the Texas Company follow.
1902–04Ida Tarbell’s History of the Standard Oil Company runs in McClure’s.
1907Royal Dutch and Shell combine.
1908Ford’s Model T.
1911Supreme Court orders Standard broken into 34 companies. Gasoline sales pass kerosene around this time.
1914British government buys 51% of Anglo-Persian to fuel a navy switching to oil.
1930–35East Texas glut; martial law in 1931; Texas Railroad Commission proration; Connally Hot Oil Act.
1951Iran nationalizes its oil.
1960OPEC founded in Baghdad, modeled on the Texas Railroad Commission.
1973–74OPEC posted price goes from about $3 to $11.65; Arab embargo.
1980Saudi Aramco fully state-owned.

The economics: who actually got paid

Strip the personalities out and the oil story is really a story about where the money sat in the chain, and why that spot kept moving.

Producers (the wildcatters) created the supply but, as a group, captured remarkably little of the value. The work was open to anyone, the product was a commodity, and the rule of capture forced everyone to overproduce. A few struck it rich. Pithole is the more typical outcome. Producers only got pricing power when a government restricted supply on their behalf, first the Texas Railroad Commission and later OPEC.

Transport (railroads, then pipelines) was a chokepoint in theory, but the railroads couldn’t hold it. They had huge fixed costs, competed with each other on the same routes, and needed volume desperately. That made them vulnerable to whoever controlled the freight. And once pipelines proved you could route around them, their leverage over crude collapsed. Building infrastructure and capturing value from it turned out to be two different things.

Refining was where Rockefeller planted his flag, because it was the narrow part of the hourglass. Then he integrated outward: into barrels, tank cars, pipelines, and finally the tank wagons that delivered to stores. The point of integration wasn’t that every layer was profitable on its own. It was that controlling the whole flow let him extract the margin wherever it was easiest, and starve competitors wherever they were weakest. Chernow estimates that Standard priced unprofitably low in roughly 9,000 of the 37,000 or so towns its tank wagons served, specifically to discipline local rivals.

Distribution and brand mattered more than people remember. “Standard” was a quality promise; later the branded service station became the face of the industry. But retail fuel itself was a thin-margin business of location and volume.

Resource owners captured the most in the end, but only when they were big enough to act as a bloc. Private American landowners mostly got royalties. Sovereign states with concentrated reserves got the rents of the 1970s.

End-use industries and consumers are the part most people forget. The real fortune created by oil wasn’t only Rockefeller’s. It was the consumer surplus: cheap light, then cheap mobility. Ford and the auto industry, trucking, suburbs, aviation, petrochemicals, all built enormous businesses on top of a fuel that kept getting cheaper. Most of the total value created by oil was captured by people who never drilled or refined a drop.

WHERE THE CHOKEPOINT SAT, ERA BY ERAAnalysis
EraChokepointWho captured valueWho got squeezed / who benefited
1859–1865Nothing. Crude is a free-for-all.Landowners who leased early; a few lucky drillers; barrel makers and teamsters.Consumers (light gets far cheaper). Most wildcatters lose money.
1865–1879Refining + rail transportRockefeller, by bargaining refining volume against railroad rate wars.Producers squeezed; railroads fight each other for Standard’s freight.
1879–1911Pipelines, refining, marketing (the integrated whole)Standard Oil — about 90% of refining at peak, plus pipes, tank cars and tank wagons.Kerosene falls to pennies; consumers win big. Rivals and producers resent it.
1901–1930sUpstream discovery shifts; the car creates a new marketIntegrated majors (Standard successors, Gulf, Texaco, Shell); Ford and the auto industry.Gasoline replaces kerosene; mass mobility is the giant consumer surplus.
1930s–1960sSupply disciplineThe Texas Railroad Commission at home; the Seven Sisters abroad via concessions.Host countries push from royalties toward 50/50 splits.
1960s–1980The resource itselfSovereign owners — OPEC states and national oil companies.Rent moves to whoever owns the ground and can withhold it.

Why did the chokepoint move? Three forces, again and again.

  1. Bypass. Whenever one layer extracted too much, someone built a way around it. Pipelines went around railroads. Texas and Baku went around Pennsylvania.
  2. New demand. The car didn’t just grow the market; it changed which product, which assets and which companies mattered. Standard had to reinvent itself around gasoline, and new integrated majors grew up alongside it.
  3. Politics. When concentration became visible and painful, the public reacted. Tarbell, the Sherman Act, the 1911 breakup, the Railroad Commission, nationalizations, OPEC. Every era’s winner eventually ran into a political force that changed the rules.

Value didn’t stay put. It flowed to whoever controlled the scarcest step at that moment, until someone built around it, a new use changed the map, or politics took it away.

PART TWOMapping the oil story onto AI

The new value chain

The tempting move is to start with the characters. The oil story says start with the chain, because characters only matter relative to where the scarcity sits. Here is the AI chain, from the ground up.

Energy, power and land. Every model runs on electricity. Lawrence Berkeley National Laboratory estimated that U.S. data centers used 176 terawatt-hours in 2023, about 4.4 percent of U.S. electricity, and projected 325 to 580 terawatt-hours by 2028, or 6.7 to 12 percent. In Texas, ERCOT had large-load requests of about 474 gigawatts as of June 2026, more than five times its record peak demand, with about 90 percent of that from data centers. This layer is the land and mineral rights of AI: you can’t refine without it.

Chips. Nvidia designs the dominant AI accelerators. In its quarter ended July 26, 2026, it reported $96.2 billion of revenue, up 106 percent from a year earlier, $89.0 billion of that from data centers, a 75.0 percent GAAP gross margin, and $59.7 billion of net income. It guided to $108 billion for the next quarter, assuming no China data-center compute revenue at all. Under Nvidia sits TSMC, which manufactures most of the world’s leading-edge AI chips, including Nvidia’s, and under TSMC sits ASML, the sole supplier of the extreme-ultraviolet lithography machines needed to make them.

Data centers and cloud. This is where the rails get laid. Amazon, Alphabet, Microsoft, Meta and SpaceX (now including xAI) plan up to $828 billion of capital spending in 2026: Amazon about $220 billion, Alphabet up to $205 billion, Microsoft about $170 billion, Meta up to $145 billion, SpaceX up to $88 billion. Oracle and specialist “neoclouds” add more. Banks such as Evercore and Bank of America see AI capex around $1 trillion in 2027.

Model labs. The refiners. OpenAI, Anthropic, Google DeepMind, xAI and Meta at the closed and frontier end; DeepSeek and a wave of mostly Chinese open-weight labs on the other. They take raw inputs (compute, data, electricity) and turn them into a product people will pay for. OpenAI was at an annualized revenue run rate of about $50 billion at the end of September 2026. Anthropic’s IPO filing showed $16.29 billion of revenue in the first half of 2026, against $1.16 billion a year earlier.

Distribution and applications. The tank wagons, gas stations and cars. Search, phones, browsers, productivity suites, and the growing layer of applications built on top of models. ChatGPT alone reported 1.2 billion weekly users on September 29, 2026.

Data owners. The landholders. Publishers, authors, Reddit, X, and all of us who wrote the web. For years, AI treated the web like a shared underground pool under the rule of capture. That is shifting: Anthropic’s $1.5 billion settlement with book authors in Bartz v. Anthropic, approved July 20, 2026, works out to about $3,000 per work; Google reportedly pays Reddit around $60 million a year; Reddit disclosed $203 million in aggregate licensing contracts. Real money, and small next to the rest of the chain.

THE VALUE CHAIN, THEN AND NOWAnalysis; AI figures as of Oct 31, 2026
Oil layerAI layerWho held itWhere profits went
Land, leases, mineral rightsEnergy, power hookups, sites, permitsThen: farmers and, later, sovereign states. Now: utilities, grid operators, power-rich states like Texas, gas-turbine and land owners.Then: modest royalties until OPEC-scale owners took the rent. Now: rising; ERCOT has hundreds of gigawatts of requests waiting.
Crude and the underground poolData: the open web, books, posts, user activityThen: whoever pumped it first. Now: publishers, Reddit, X, authors, users.Then: rule of capture, almost nothing to the landowner. Now: shifting toward licensing and settlements, still small next to the chain.
Drilling rigs, barrels, tank carsChips and the tools that make them: Nvidia, TSMC, ASMLThen: many suppliers. Now: a few near-monopolies.Then: a modest equipment trade. Now: the fattest margins in the chain (Nvidia at 75% gross margin).
Railroads and pipelinesData centers and cloud: Microsoft, Amazon, Google, Oracle, SpaceX/xAI, neocloudsThen: Vanderbilt, Scott, Gould, then Standard’s own pipes. Now: hyperscalers plus SpaceX.Then: railroads overbuilt, fought rate wars, and lost leverage to pipelines. Now: huge capex, cloud revenue growing fast, free cash flow negative at four of five big spenders.
RefiningModel labs: OpenAI, Anthropic, Google DeepMind, xAI, Meta, DeepSeek and the open-weight labsThen: Standard, eventually ~90%. Now: a handful of frontier labs plus a swarm of open-weight labs.Then: Standard’s fortune. Now: fast-growing revenue but compute-heavy costs; open weights compress prices.
Kerosene and brand (“Standard”)The model product: chat, API, coding agentsThen: Standard. Now: ChatGPT (1.2B weekly users), Claude, Gemini, Grok.Then: brand plus cost let Standard win at pennies. Now: brand and trust matter; per-token prices fall fast.
Tank wagons, stations, and the carDistribution and apps: search, phones, browsers, enterprise software, vertical appsThen: Standard’s wagons, branded stations, Ford. Now: Google, Apple, Microsoft, Meta, and thousands of app builders.Then: the car industry and consumers captured the giant surplus. Now: whoever owns the user relationship.
Governments: TRC, OPEC, the AdmiraltyGrid regulators, export controls, Taiwan, courtsThen: Austin, Riyadh, London. Now: Austin again, Washington, Beijing, Taipei.Then: rent and rules. Now: permission to connect, permission to sell chips abroad.

Two things stand out. First, oil had one narrow spot at a time; AI has several stacked up: lithography, foundry, accelerators, power. Second, AI is manmade all the way down. The scarce inputs are factories, machines, designs, power plants and permits, so the “landholders” are many different players, some of them governments.

Casting the characters

Now the fun part: the main candidates, with a verdict on each. The point isn’t one perfect mapping. It’s to see what each casting predicts, and which predictions you’d actually bet on.

Elon Musk as Rockefeller

The case for: Musk is the most vertically integrated figure in AI. X supplies data and distribution. xAI builds models. SpaceX, which acquired xAI on February 2, 2026 (at valuations of $1 trillion for SpaceX and $250 billion for xAI) and listed on June 12 at a first-day market cap of roughly $2.1 trillion, provides capital, launch, and Starlink connectivity. Tesla brings robots, cars and energy storage. The Colossus data centers in Memphis come with their own gas turbines. No one else owns so many layers at once, and nobody else is talking seriously about putting data centers in orbit and linking them with Starlink’s laser mesh.

The case against: Rockefeller’s integration was built on top of being the best and cheapest refiner. Integration was the moat around a dominant core. xAI isn’t the dominant refiner. It lost $2.4 billion in the first quarter of 2026 and spent heavily on infrastructure ($12.7 billion in 2025, $7.7 billion in Q1 2026). Its models are competitive, but nothing in the numbers says it is the cheapest or most-used refiner. And the SpaceX filing itself calls the orbital data-center idea “novel and untested.”

VERDICTRockefeller’s ambition, not yet Rockefeller’s position. If xAI ever becomes the low-cost, high-volume refiner and the integration locks it in, this casting gets much stronger.

Elon as Vanderbilt

This is the casting the facts lean toward right now. According to SpaceX’s filing, Anthropic pays SpaceX $1.25 billion a month, roughly $15 billion a year, to rent about 300 megawatts of xAI data-center capacity through May 2029, with a 90-day termination right. Cursor rents capacity too. And SpaceX says it “expect[s] to enter into additional similar services contracts for compute capacity with third parties.” In October 2026, SpaceX closed $40 billion of debt to buy GPUs.

Read that again in oil terms. A company that owns a model lab is renting its rails to a competing refiner, on a big, volume-committed, multi-year contract, financed with debt. That’s the Lake Shore and Flagler, except the railroad owner also has a refinery.

There is a darker reading, too. In 1877, the Pennsylvania Railroad tried to move into refining through Empire, and Rockefeller crushed it, because the refiner controlled the volume. Today the roles may be reversed. The infrastructure owners are vastly stronger than the labs that rent from them, and several of them (Google, Meta, SpaceX through xAI) also run their own frontier models. The Empire play might actually work this time.

VERDICTStrong, and getting stronger. Musk may be Vanderbilt and Tom Scott at once: a rail baron whose railroad also refines.

Elon as the Nobel brothers

A casting people miss. The Nobels built an integrated empire in Baku, far from the center, and invented the tanker to make it work. Formidable and innovative, they never became Standard Oil, just one strong player in a world that ended up with several.

Swap Baku for Memphis and the tanker for rockets and Starlink, and the parallel holds: an engineering-first integrator that builds its own logistics and is powerful without being dominant.

VERDICTPerhaps the best current fit for Musk. It predicts a major independent power, not a monopolist.

Jensen Huang as Vanderbilt

Everything moves over Nvidia’s tracks. A 75 percent gross margin is not what an equipment maker earns. It’s what a toll road earns when there is no alternate route. Jensen has framed the economics in railroad-like terms himself: “Now, compute is revenue.” Nvidia has also moved into financing its customers, partnering with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize more than $500 billion of third-party capital for AI infrastructure, and joining AMD to invest a total of $4 billion in Meta alongside Meta’s planned raise of up to $95 billion. That is a railroad helping its biggest shippers buy more rail cars.

And just like the railroads, the toll invites a bypass. Google’s TPUs, Amazon’s Trainium, Meta’s MTIA and other custom chips are the Tidewater pipelines of this era: built specifically so the biggest shippers don’t have to pay the toll forever. Nvidia’s response looks a lot like Standard’s response to pipelines. It is absorbing alternatives (it acquired or licensed Groq’s technology, and Groq 3 LPX is now in production according to Nvidia) rather than letting them grow up independently.

The DeepSeek episode showed how fast a toll road can look vulnerable. On January 27, 2025, Nvidia lost about $589 billion of market value in a single day, a record, on fears that cheaper models would need fewer chips. Satya Nadella’s reply on X: “Jevons paradox strikes again!” Cheaper AI would mean more AI, which would mean more chips. So far, he’s been right.

VERDICTStrong. The best answer to “who is Vanderbilt?” is Jensen, with the warning that railroads were also where the overbuilding and the bypasses happened.

Jensen as the rig-and-barrel maker

The alternative view: Nvidia is a pick-and-shovel supplier, like the barrel makers and drilling-rig builders of the Oil Regions. During a rush, suppliers do well. When the rush calms, their margins compress toward normal manufacturing margins.

The problem: nobody in oil’s equipment trade earned anything like Nvidia’s margins or had its software lock-in. But keep it in your back pocket. In 1861 the barrel was worth more than the oil, and that scarcity didn’t last.

VERDICTRight position in the chain, wrong margin profile. Becomes the right casting if custom silicon and competitors catch up.

Sam Altman and OpenAI as Rockefeller

OpenAI refines compute into a product people want, and it owns the strongest consumer brand in the field. ChatGPT, with 1.2 billion weekly users, is the closest thing AI has to “Standard” kerosene: the name people reach for. OpenAI has also made giant volume commitments for compute, including the Stargate buildout, which is the Flagler move: guarantee the railroad its sixty carloads a day, and get priority and price in return. Anthropic has done the same; its filing shows more than $265 billion of non-cancelable minimum purchase commitments, mostly for compute.

But Rockefeller was the low-cost producer and controlled the transport. OpenAI is neither. It rents its rails from Microsoft, Oracle and others. Open-weight models undercut it on price. And unlike Standard, the labs are not the narrowest point in the chain; the chips and power are.

VERDICTBrand and distribution like Standard; economics more like a big independent refiner that pays the railroad. The “Netscape” risk from this site’s thesis lives here.

Google as Standard Oil

If you ignore the personalities and look only at structure, Google is the most Standard-like company in AI. It has its own chips (TPUs: its own pipelines, built to avoid the toll). It has its own cloud (its own rails), and Google Cloud revenue more than doubled last quarter. It has a frontier lab in DeepMind (the refinery). And it has the tank wagons: Search, Android, Chrome, YouTube, delivering AI to billions of people without needing anyone’s permission. Google processed about 3.2 quadrillion tokens a month as of May 2026. And like Standard, it has already lost major antitrust fights.

VERDICTThe strongest structural fit. Not as dramatic as any of the Musk castings, but if you want to know which company could quietly integrate from silicon to the consumer, this is the one.

The other casts

Hyperscalers as railroads. Amazon, Microsoft, Google and Oracle look a lot like the trunk lines: enormous fixed costs, borrowed money, big volume deals, and growing revenue (Azure up 48 percent, AWS about 39 percent). They are also, increasingly, refiners. Custom silicon is their Tidewater.

TSMC and ASML as the narrow pass. This is the one thing oil never really had: a single, physical, technical chokepoint that nobody can currently route around. China is barred from ASML’s EUV machines, and proposed U.S. rules in April 2026 would restrict deep-ultraviolet sales and servicing too. If there is a “Persian Gulf” in this story, it’s a small number of factories in Taiwan and one company in the Netherlands.

Texas and the utilities as the Railroad Commission. In 1931 the Texas governor sent troops to shut in wells. In 2026 the governor of Texas sent auditors. On August 3, Greg Abbott ordered the state’s utility commission and ERCOT to audit all data centers in the interconnection process before any more advance; noncompliant projects “must be denied” a grid connection. A 2025 law, SB 6, already lets large loads be curtailed in emergencies and makes them pay for interconnection. As of late October, the queue stood around 440 gigawatts across 1,797 requests. That is proration by another name: a state body deciding who gets to produce. The difference is that this time Texas is rationing the input, not the output.

Open weights as Spindletop. Spindletop didn’t kill Standard. It created new owners, more supply and lower prices, and Standard was never as dominant again. The open-weight wave fits. On OpenRouter, a large developer routing service, the open-weight share of tokens among the top 50 models went from 26.8 percent in October 2025 to a peak of 77.5 percent in July 2026, before settling around 71.7 percent in late September. Chinese-developed models reached about 72 percent of OpenRouter tokens in the week ending October 25, 2026. (OpenRouter is a developer sample, not the whole market, so treat this as a signal rather than a census.) Just as Texas came with Gulf and Texaco, the new supply here comes with new owners, many of them in China.

Data owners as landholders. Sobering. Private landowners in Pennsylvania and Texas mostly got modest royalties; only sovereign-scale owners who could withhold supply captured serious rents. Unless data owners act as a bloc (an OPEC of content), oil suggests they’ll get paid, but not much.

CASTING SCORECARDAnalysis — a judgment call, not a finding
CastingFitWhy
Elon as RockefellerWeak–mediumIntegration is real; refining dominance isn’t. xAI is not the best or cheapest refiner.
Elon as VanderbiltMedium–strongSpaceX now rents compute to a rival lab. That is a railroad business.
Elon as the NobelsStrong (for now)Engineer-integrator with a new transport technology, a big regional base, and no monopoly.
Jensen as VanderbiltStrongToll on everything that moves; 75% gross margins; bypasses under construction.
Jensen as the rig-and-barrel makerMediumRight role in the chain, wrong margins. Barrel makers never earned 75%.
OpenAI as RockefellerMediumHas the brand and the distribution; doesn’t own the rails or the crude.
Google as Standard OilStrongest structural fitOwn pipelines (TPUs), own rails (cloud), refiner (DeepMind), wagons (Search, Android, Chrome, YouTube).
Hyperscalers as railroadsStrongFixed costs, debt, volume deals, rate pressure, and custom silicon as their Tidewater.
TSMC / ASML as the narrow passStrong, and newOil never had a single point like this. Nobody routes around them today.
Texas / grid as the Railroad CommissionStrong and risingThe 2026 data-center audit is the first proration-like act.
Open weights as SpindletopStrongNew supply, new owners, falling prices, and the incumbent’s grip loosens.

Where the analogy breaks

The analogy is useful. It is also wrong in important ways, and those matter as much as the ways it’s right.

AI is manufactured, not extracted. Oil was in the ground, and its supply was ultimately set by geology. You could find more, but you couldn’t make more. AI’s inputs (chips, data centers, power plants, model weights) are built. Supply responds to investment, not discovery. That means the oil-era logic of owning scarce land is less powerful. Anything people can build, other people can build too, given enough capital and time.

Costs fall with learning, not rise with depletion. Oil fields deplete. The easy oil gets pumped first, and the next barrel is harder to find. AI runs the opposite way. Every generation of chips does more per dollar, models get more efficient, and the cost of a given level of capability collapses. Stanford’s AI Index found that the cost of running a model at GPT-3.5’s level (64.8 percent on the MMLU benchmark) fell from $20 per million tokens in November 2022 to $0.07 in October 2024, a drop of more than 280 times. Epoch AI estimates that for a fixed level of performance, prices fall anywhere from 9x to 900x a year depending on the task, with a median around 50x a year. No commodity in history has ever behaved like that.

Software copies at near-zero marginal cost. A barrel of oil can only be burned once. A model, once trained, can be copied endlessly. That changes everything about refining economics. Standard’s refineries were physical assets that took years to replicate. A model’s weights can be downloaded in an afternoon.

Open weights and distillation. In oil, a rival had to find its own field. In AI, a rival can take an open model and build on it, or “distill” a smaller, cheaper model from the outputs of a bigger one. On OpenRouter, popular open models are served by 20 to 31 competing providers; DeepSeek V4.1 Flash, for instance, has 31, with input prices from $0.03 to $0.60 per million tokens. Token-weighted list prices for open models were $0.28 per million input tokens and $1.62 per million output tokens, against $1.27 and $6.36 for closed models, roughly four times higher. That is what a commodity market looks like.

Quality still matters, at least at the top. On the other hand, AI isn’t perfectly fungible like crude. The best closed model on the Artificial Analysis index (Claude Opus 5.5, at 57.6) still scores about eleven points above the best open model (MiMo-V2.6-Pro, at 46.3). For hard problems, buyers pay up. But most volume goes to cheap, fast models. That is the kerosene pattern: a premium brand at the top, a commodity everywhere else, and the volume in the commodity.

Many more “landholders.” Oil had one critical natural resource. AI has a stack of manmade ones, owned by different players in different countries. That makes a single Rockefeller less likely. The chain is too long and too varied for one company to control all of it, even if Google comes closest.

Speed. Oil took a century to get from Titusville to OPEC. AI went from ChatGPT’s launch to a ~$828 billion annual capex budget in about four years. Booms, busts and backlash will be compressed alike.

ANALYSISNet: oil is a good guide to how value moves (to the chokepoint, then around it) and a weaker guide to where it ends up. In AI, the scarce things are built, so scarcity is temporary by design. Expect chokepoints to move faster, and expect every one of them to attract a bypass.

PART THREEWhat it implies

Jevons for intelligence

If there is one part of the oil story that maps onto AI almost perfectly, it is Jevons.

The price of a given level of AI capability has fallen by orders of magnitude in a couple of years. More than 280x for GPT-3.5-level performance between late 2022 and late 2024, by Stanford’s measure; around 50x a year at the median, by Epoch AI’s. If AI followed the intuition Jevons argued against, that would mean the total AI bill would shrink. It hasn’t.

Google processed about 9.7 trillion tokens a month in May 2024. By May 2025 it was 480 trillion. By May 2026 it was 3.2 quadrillion. That is roughly a 330-fold increase in two years, at one company. On OpenRouter, open-weight token volume rose more than 60x over the past year as prices collapsed, and closed-model volume rose about 10x. The cheaper the unit, the more units the world consumes, and total spending goes up, not down. That is why hyperscalers are spending up to $828 billion this year, and why Anthropic’s IPO filing says plainly: “We have historically operated in a compute-constrained environment in which demand for compute has exceeded the supply available to us.”

ANALYSISThe useful Jevons question isn’t “will usage grow?” It’s “who gets paid for the growth?” In oil, cheaper energy mostly enriched the people who used it to do new things, plus whoever owned the narrowest pass at each moment. The same thing is likely here.

The chokepoint will move

Today, the narrowest pass in AI is chips: Nvidia’s accelerators, TSMC’s fabs, ASML’s machines. That is why Nvidia’s margins look like a toll road’s.

But oil says chokepoints don’t stay put. Here is a plausible sequence for the next decade. It is a guess, and labeled as one.

  1. Chips (now). Scarce, high-margin, concentrated. Custom silicon, competitors and capacity expansion are the pipelines being built around it.
  2. Power and permits (next, and already starting). Data centers need electricity, and the grid can’t be built at chip speed. ERCOT’s queue (around 440 gigawatts) and the Texas audit are the early signs. xAI’s gas turbines in Memphis, and the lawsuit over them, show the local politics. Whoever controls interconnection becomes the new Railroad Commission.
  3. Distribution and trust (later). As models commoditize, value moves toward whoever owns the customer relationship: the default assistant on the phone, the search box, the enterprise software seat, the brand people trust with their money, health and family. That’s the tank-wagon and service-station layer, and in oil it was where Standard locked in its final advantage.

Models may be the layer where oil predicts the least durable profit: 1860s-style open entry plus software’s near-zero copying costs. The winning labs will likely own something scarce below them (compute, power) or sticky above them (users, workflows, trust).

The railroad bubble

This site is about bubbles, so let’s be direct. In the oil story, the overbuilt, debt-financed boom wasn’t oil. It was the railroads. They were laid ahead of demand with borrowed money, competed brutally once built, and in 1873 the collapse of Jay Cooke’s railroad financing helped set off one of the worst panics of the century. The railroads were enormously valuable to the economy. That didn’t save the people who financed them at the wrong price.

The data-center buildout looks a lot like that. Up to $828 billion of capex this year, perhaps $1 trillion next. Four of the five biggest spenders had negative free cash flow in the third quarter of 2026, Microsoft being the exception. Amazon’s free cash flow over the past year was negative by more than $42 billion. Microsoft sold a $25 billion bond, its first in more than nine years. Meta plans to raise up to $95 billion through new shares, with Nvidia and AMD investing a total of $4 billion alongside — suppliers financing customers. SpaceX closed $40 billion of debt to buy GPUs. On October 30, 2026, the Nasdaq 100 lost more than $1 trillion in market value, partly on capex fears.

Andy Jassy’s answer on Amazon’s earnings call is worth quoting, because it is exactly the kind of thing railroad builders would have said:

“Would it be more convenient if this type of acceleration and opportunity happened over 25 years and you could spread the CapEx over 25 years... Yeah, it might be more convenient, but that’s not the world that we live in, and that’s not the opportunity.”— ANDY JASSY, AMAZON, OCTOBER 2026

He may be right about the opportunity. So were the railroads. The bubble question is different: is the capital priced as if every mile of track will earn a toll forever? Oil says some rails get bypassed, some get overbuilt, and the long-run winners are often the shippers using them at falling prices.

ANALYSISThere’s a real difference, though. The 1870s railroads were mostly standalone companies living on debt. Today’s biggest builders have enormous, profitable core businesses (search, ads, software, retail) standing behind the spending. That makes a 1873-style collapse less likely at the core, and pushes the risk toward the edges: the debt-financed neoclouds, the vendor-financed customers, and the investors paying peak multiples.

The Tarbell moment

Every era of concentration in oil eventually produced backlash: the Oil Regions’ revolt in 1872, Tarbell, the breakup.

AI’s Tarbell moment hasn’t come, but the raw material is there: copyright fights, power-bill fights, the NAACP lawsuit over xAI’s 27 methane turbines in Memphis, job worries, and extreme concentration of market value. Tarbell’s power came from documentation, not outrage, and the S-1s now being filed are the kind of paper trail she would have loved.

A 1911-style breakup is possible, especially of whichever company ends up most integrated. But remember the punchline: Rockefeller got richer after the breakup. Breakups can unlock value for shareholders. They don’t necessarily hurt the people who own the pieces.

Taiwan as the Persian Gulf

Churchill switched the navy to a fuel Britain didn’t have, so the government bought control of Anglo-Persian, and oil-producing regions became the center of geopolitics for a century. The AI version is already here. The most advanced chips are made largely in one place, Taiwan, with machines from one supplier, ASML. Export controls are the embargo of this era: China is barred from EUV lithography, the U.S. has proposed restricting deep-ultraviolet sales and servicing as well, and Nvidia’s own guidance assumes no China data-center compute revenue at all. Meanwhile, Chinese open-weight models now account for a large share of the tokens flowing through OpenRouter. The embargo has not stopped the oil from flowing. It has changed who pumps it.

In 1973, the people who owned the resource discovered they could withhold it. The AI version of that risk is not a cartel of model labs. It’s a disruption at the narrowest physical point in the chain.

The gasoline moment

For oil, the decisive event wasn’t a better lamp. It was the car. A new use for the product turned out to be much bigger than the original, killed the original’s importance, and reshuffled who got rich.

AI’s original use case, chat, is already big: 1.2 billion weekly ChatGPT users. But chat might be AI’s kerosene. The candidates for the gasoline moment are the uses that consume vastly more tokens per person and do work rather than answer questions: agents that run for hours on their own, coding systems that write and test software, and physical AI in cars, robots and factories. An agent grinding through a task all day consumes far more tokens than a person typing questions into a chat box. If that happens at scale, today’s token counts will look like 1880s kerosene sales.

The lesson from 1911: the biggest winners of the gasoline era weren’t refiners. Standard adapted, but the car created Ford and a whole new industry built on cheap fuel.

Society and labor

Oil reached ordinary people in waves: cheap light, then cheap mobility, suburbs, trucking and global trade. The whalers lost their jobs; many more jobs appeared that nobody in 1859 could imagine. The costs were real too, from boomtown busts to pollution to wars over supply.

AI is likely to follow a similar shape, faster. Cheap intelligence makes expertise available to people who couldn’t afford it before: legal help, tutoring, health information, bookkeeping. It will also displace work that is mostly routine information processing, the way kerosene displaced whale oil. And like oil, it will concentrate some of its gains in a few hands while spreading most of its benefits as consumer surplus that never shows up on anyone’s income statement.

The honest part: oil’s transitions took decades; AI’s may take years. The backlash will likely be about labor and local costs (power bills, water, land) as much as monopoly.

Four scenarios

No one knows how this plays out. Here are four distinct versions, each with something specific to watch for. These are analysis, not predictions; the real outcome will probably mix more than one.

SCENARIO 1

Standard Oil 2.0

One or two integrated players (Google is the obvious candidate; SpaceX/xAI the wild card) end up owning chips, cloud, models and distribution. They get a lasting cost advantage, use it to undercut rivals, and capture a large share of the value. Eventually they face a Tarbell moment and an antitrust case.

WATCH FOR

  • Custom silicon (TPU, Trainium) taking a growing share of AI workloads.
  • Labs consolidating into hyperscalers or being starved of compute.
  • Bundling of AI into search, phone and productivity defaults.
  • New antitrust cases aimed at AI integration.
SCENARIO 2

Spindletop Glut

Open-weight models and cheap inference keep getting better and keep falling in price. Tokens become a commodity, like crude in 1861 or 1931. Model labs struggle to earn returns on their training spend. The big winners are the users and application builders who get nearly free intelligence, and whoever owns distribution.

WATCH FOR

  • The gap between the best closed and best open models (about eleven points today) shrinking.
  • Open-weight share of tokens rising past its July 2026 peak.
  • Price cuts at the frontier labs; falling revenue per token.
  • Labs pivoting from selling tokens to selling applications.
SCENARIO 3

OPEC of the Grid

Power and permits become the binding constraint. States, utilities and grid operators decide who gets to build, effectively rationing AI capacity like the Railroad Commission rationed oil. Value shifts toward energy owners, power-rich regions, and whoever already has a hookup. Chip margins compress because chips sit idle waiting for power.

WATCH FOR

  • What comes out of the Texas audit, and whether other states copy it.
  • Data centers paying premium rates or building their own generation.
  • Political fights over power bills and turbines.
  • Rental prices for existing, powered capacity rising.
SCENARIO 4

Gasoline Moment

Agents and physical AI create demand that dwarfs chat, the way cars dwarfed lamps. Token use explodes far beyond current forecasts. Today’s capex turns out to be too little, not too much. The winners include whoever builds the “cars”: the agent platforms, robot makers and businesses built on top of them.

WATCH FOR

  • Token consumption per user rising sharply as agents run for hours.
  • Revenue from agents and coding systems overtaking chat subscriptions.
  • Robotaxis and robots moving from pilots to scale.
  • Persistent compute shortages despite the capex boom.

The short version

Oil’s history is not a story about one great man. It’s a story about a chokepoint that kept moving: from the barrel, to the refinery, to the railroad, to the pipeline, to the integrated giant, to the government quota, to the sovereign owner. Each time, whoever held the narrowest pass captured the rents until someone built around it, a new use changed the map, or politics took it away.

AI is following the same pattern, faster, with more players and a different physics. Nvidia looks most like Vanderbilt. Musk, for now, looks more like a rail baron crossed with the Nobels than like Rockefeller. Google looks the most like Standard. Open weights look like Spindletop. Texas is acting like the Railroad Commission again. And the biggest value, as with oil, will probably go to the people who use cheap intelligence to do something new.

Don’t sell the fuel. Build the car, own the corner, and keep the customer’s trust.

This essay is analysis and opinion built on dated, sourced figures (as of October 31, 2026). Some historical figures vary by source and are flagged where they do. It is not investment advice; mentions of companies are not recommendations.

DON'T TAKE MY WORD FOR IT

Research trail

Nordhaus — The History of Lighting (price of light) ↗Petroleum History Institute — whale oil prices ↗Springer — The American whale oil industry ↗Hamilton — Historical Oil Shocks (NBER) ↗Petroleum History Institute — early barrel scarcity ↗Oil & Gas Journal — A U.S. oil story ↗AAPG Explorer — Pithole: rise and fall of an oil boomtown ↗Tarbell — History of the Standard Oil Company, Vol. 1 Ch. 2 ↗Tarbell — Appendix 3 (Devereux affidavit, Lake Shore deal) ↗Tarbell — Appendix 5 (South Improvement Company contract) ↗Tarbell — Vol. 1 Ch. 6 (the 1877 Empire fight) ↗Daniel Crane — rebates and exclusion (Michigan Law) ↗UH Digital History — South Improvement Company lesson ↗FEE — John D. Rockefeller and the oil industry ↗H-Net — review of Chernow’s Titan ↗Smethport History — the Tidewater pipeline ↗Scientific American — The Tide-Water oil pipe line ↗Gilder Lehrman — the life and fortune of John D. Rockefeller ↗Titan Agonistes — the Standard Oil breakup ↗ExxonMobil — company history ↗EIA — history of gasoline ↗TSHA — Spindletop oilfield ↗Hansard — Anglo-Persian Oil Company, June 17 1914 ↗TSHA — East Texas oilfield ↗NPR — The Texas origins of OPEC ↗Federal Reserve History — Oil shock of 1973–74 ↗Fortune — Big Tech’s AI capital spending spree (Oct 31 2026) ↗Nvidia — Q2 FY2027 results ↗Fortune — Nadella on DeepSeek and Jevons paradox ↗Stanford AI Index 2025 — research and development ↗Epoch AI — LLM inference price trends ↗Business Insider — Google’s AI token usage surges ↗Techmeme — Google monthly tokens ↗The Verge — ChatGPT at 1.2 billion weekly users ↗Investing.com / FT — OpenAI annualized revenue ↗CNBC — Anthropic S-1: revenue and compute ↗SEC — Anthropic PBC Form S-1 ↗DCD — SpaceX filing: Anthropic pays $1.25B/month for compute ↗Reuters — SpaceX to merge with xAI ↗Nasdaq — SpaceX IPO ↗OpenRouter — open-weight vs closed-weight models ↗Dror Poleg — Chinese models on OpenRouter ↗LBNL — 2024 U.S. data center energy usage report ↗Akin — Texas pauses data center interconnections ↗ERCOT — Senate panel: assessing the grid (July 2026) ↗Reuters — proposed U.S. chipmaking export restrictions ↗CACM — The ASML Monopoly ↗Reuters — judge approves Anthropic’s $1.5B copyright settlement ↗AP — Anthropic copyright settlement (Bartz) ↗