The Future Was Here

Global fossil fuel consumption hit an all-time high in 2023.¹ Not in 2005. Not before the Paris Agreement. In 2023. While every government on earth, every energy company, and every investor presentation in circulation was telling the same story about the transition.

In 2000, fossil fuels provided 86 percent of the world's primary energy. By 2023, after twenty-three years of the fastest renewable energy buildout in human history, that share had fallen to 82 percent.²

Four points. A quarter century. Total global energy consumption grew roughly 50 percent over the same period.³ Fossil fuel consumption did not decline. It reached a record. 505 exajoules. Oil surpassed 100 million barrels per day for the first time.

Two lines on the same chart, both going up. Renewable capacity rose. Fossil fuel consumption rose. The new energy was being added on top of the old energy. Not instead of it.

Both numbers were correct. One was on every slide deck. The other was not.

The tonnage

Solar photovoltaic capacity grew from roughly 40 gigawatts in 2010 to 1,600 gigawatts by the end of 2023. A forty-fold increase in thirteen years.⁴ Wind capacity crossed the one-terawatt milestone the same year, up fivefold from 2010. Combined, solar and wind grew elevenfold.

In percentage terms, these numbers look like an exponential curve. They are.

The number that does not appear on the same slide: the share of global primary energy provided by fossil fuels moved from 86 to 82 over that quarter century.

The puzzle is not why renewables grew so slowly. They grew faster than any energy source in history. The puzzle is why that growth barely dented the share. The answer is the denominator.

Global primary energy demand grew roughly 50 percent between 2000 and 2023. From approximately 400 exajoules to 620. Every solar panel installed and every wind turbine erected was feeding into a system that kept getting larger. The growth in total energy demand absorbed the growth in renewable supply. Renewables were not failing. The denominator was growing faster than renewables could fill it.

In Germany, the Energiewende began in the early 2000s. Two decades of subsidized renewable buildout. Solar panels on every available rooftop. The country became the global showcase for the energy transition. By 2020, Germany still needed 89 percent of its fossil-fired generating capacity. Solar photovoltaics, in northern Germany, generate on average 11 to 12 percent of the time.⁵

Global clean energy investment hit $1.8 trillion in 2023.⁶ One year. By 2024, the figure exceeded $2 trillion. For every dollar going into fossil fuels, $1.70 went to clean energy. Five years earlier, the ratio was one to one.

The reader sees 86 to 82 and thinks: not enough effort. The spend says otherwise. Trillions of dollars. The fastest buildout in history. The most concentrated energy investment since the Second World War. Four percentage points.

The spend did not fail. Without it, the four points would be zero. But the distance between the investment and the result reveals something about the problem that the slide deck does not show.

William Stanley Jevons answered this in 1865.⁷

Jevons observed that James Watt's improved steam engine made engines more efficient. Each unit of output required less coal. The efficiency gain was real. Coal consumption went up.

Efficiency made coal cheaper per unit of useful work. Cheaper energy per unit opened applications that were previously uneconomical. Factories that could not afford to run steam engines could now run them. Mines that were too deep to pump became accessible. The number of applications grew faster than the per-unit savings reduced consumption. Total coal use increased. Jevons published the finding in The Coal Question. Economists later named the mechanism after him.

One hundred sixty years later, the mechanism has not changed.

Renewables make electricity cheaper in certain applications. Cheaper electricity opens new demand categories that did not exist a decade ago. Data centers consumed roughly 260 terawatt-hours of electricity globally in 2020. By 2024, the figure had reached 415 terawatt-hours. The International Energy Agency projects 945 terawatt-hours by 2030.⁸

AI-optimized servers are projected to consume nearly five times more electricity in 2030 than they did in 2025.⁹ Each generation of chip is more efficient than the last. Total consumption accelerates. In January 2025, when a Chinese startup demonstrated dramatically cheaper AI training, the CEO of Microsoft cited the Jevons Paradox by name: cheaper AI will mean more AI consumption, not less.

By 2030, the United States is set to consume more electricity processing data than manufacturing aluminum, steel, cement, and chemicals combined.¹⁰ In Virginia alone, data centers consumed roughly 26 percent of total state electricity supply in 2023.

Electric vehicles. Heat pumps. Air conditioning. Artificial intelligence. Data centers. Each requires electricity. Each is growing. Global electricity demand is growing at 3.5 percent per year through 2030, two and a half times faster than overall energy demand.¹¹ The new demand categories do not replace the old demand. They are added on top of it.

The cheaper the new energy becomes, the more total energy the world consumes. The more total energy consumed, the more the old energy stays to fill the expanding gap.

The addition is structural, not a failure of will. Studies measuring economy-wide rebound effects across five countries found rebounds of 78 to 101 percent, implying that efficiency gains alone may not reduce total energy consumption at all.¹² The mechanism Jevons documented has been running for 160 years. The fuel changed. The paradox did not.

The boom

The compressed timeline is a product.

Nobody in the room has an incentive to say "this will take forty years." The investor deck needs the S-curve steep. The politician needs the transition imminent. The CEO needs the disruption narrative to justify the capital raise. The analyst needs returns within the fund's horizon. The journalist needs the headline.

The compressed timeline is what gets sold. And the compressed timeline creates a capital cycle that damages the thing it describes.

Solyndra received $535 million in federal loan guarantees.¹³ It filed for bankruptcy on September 1, 2011. The Department of Energy lost $528 million. Chinese manufacturers, backed by more than $34 billion in government credit lines, had driven global solar panel prices down 80 percent in five years. By 2010, China produced more than 60 percent of the world's panels.¹⁴

Solyndra was a business failure with a specific cause: a technology bet that could not compete with Chinese manufacturing scale. What followed was a political event. The House Energy and Commerce Committee held hearings. The Energy Secretary testified for more than five hours. Mitt Romney gave a speech at the shuttered headquarters. Republicans introduced the "No More Solyndras Act." Americans for Prosperity spent $8.4 million in swing states on television ads.¹⁵

"Solyndra" became an argument against clean energy investment. Not against narrative-speed investing. Against clean energy itself.

Of the $25 billion that venture capitalists poured into cleantech between 2006 and 2011, more than half was lost by 2015.¹⁶ Early-stage cleantech funding dried up. For years, the word "cleantech" was venture capital's cautionary tale. The DOE loan program that funded Solyndra ultimately turned a profit across its full portfolio. That did not make the evening news.¹⁷

The pattern repeated with electric vehicles a decade later. Capital arrived at narrative speed. The infrastructure operated at material speed. The gap produced the bankruptcies.

Lordstown Motors went public through a SPAC in 2020 at a $1.6 billion valuation. Its stock peaked above $5 billion. Bankrupt by June 2023.¹⁸ Arrival went public at $13.6 billion, the largest listing in British startup history. UPS ordered 10,000 delivery vans. Zero vehicles were delivered. Bankrupt by May 2024.¹⁹ Proterra filed for bankruptcy with municipal transit systems among its top unsecured creditors. Buses built but undelivered. Chargers ordered but uninstalled.²⁰ Nikola's founder was convicted of fraud. The company filed for bankruptcy in February 2025. Canoo had $700,000 in the bank before filing Chapter 7.²¹ Fisker went bankrupt in June 2024.

Lucid Motors peaked at roughly $90 billion in market capitalization. It trades at approximately $3.5 billion. VinFast briefly exceeded $200 billion. It recorded a $3.2 billion net loss in 2024.²²

At least seven EV companies that went public through SPACs during the 2020-2021 wave have filed for bankruptcy.²³

Every one of those failures damages public confidence in the very transition the narrative promised. The workers who were hired and laid off are real. The transit agencies that ordered electric buses and received nothing are real. The charging infrastructure that was supposed to exist by now does not.

The narrative about the speed of the transition is the thing slowing the transition.

The fabs

ChatGPT went from zero to one hundred million users in two months. No consumer application in the history of the internet had ramped faster.²⁴

Then the industry needed chips.

TSMC announced its Arizona fabrication plant in 2020. Original timeline: production by 2024. The first fab entered high-volume production in late 2024, roughly a year behind schedule. The second was pushed to 2027. A third is under construction. The total investment, initially $12 billion, has grown to $165 billion. Building a fabrication plant in the United States costs four to five times what it costs in Taiwan.²⁵

Intel announced two fabs near Columbus, Ohio. Original timeline: chip production by 2025. Current projection: the first factory by 2030 or 2031. Five to six years behind the announcement.²⁶

Samsung's Taylor, Texas fab was scheduled for completion in April 2024. Investment grew from $17 billion to $44 billion. Full production is now expected in early 2027. The delay was partly technical. It was also that Samsung could not find enough customers for the capacity.²⁷

The CHIPS and Science Act was signed on August 9, 2022. It appropriated $52.7 billion for semiconductor manufacturing and research.²⁸ By mid-2025, roughly $6 billion had been disbursed. Twenty-four of one hundred sixty-one project milestones had been completed.

NVIDIA's market capitalization went from $1 trillion to $5 trillion in roughly two years, the fastest such ascent in stock market history. Apple and Microsoft each took more than two years to move from $1 trillion to $2 trillion. NVIDIA did it in under nine months.²⁹

The market cap moved at narrative speed. A three-nanometer fabrication plant requires two to four million gallons of ultrapure water per day.³⁰ A large fab consumes as much water as a city of 60,000 people. The semiconductor industry will need 300,000 engineers by 2030. At current graduation rates, roughly 58 percent of those positions will go unfilled.³¹

You can pass a law at software speed. You can build a stock price at software speed. You pour concrete at material speed. You train a fabrication workforce at material speed. You purify four million gallons of water at material speed.

The reader is watching this one happen in real time.

The fleet

Nearly 1.5 billion passenger vehicles are on the road globally.³² The average vehicle in the United States lasts 16.6 years. In Western Europe, 18 years. In Eastern Europe, 28. Cars flow from richer countries to poorer ones, extending lifespans further.³³

In 2024, electric vehicles accounted for more than 20 percent of new car sales worldwide. A record.³⁴

Electric vehicles accounted for roughly 4 percent of the total global fleet.

Even if EV sales reach 50 percent of all new vehicles by 2035, the fleet does not cross to majority electric until the early 2040s.³⁵ The grid capacity to charge them at scale does not exist. Four countries produce roughly 90 percent of the world's lithium. Seventy percent of lithium chemical processing runs through China.

The narrative tracks the sales share. The sales share is a forward indicator. It measures what is being bought. It does not measure what is on the road, what is running, what is consuming fuel. The fleet is the number that determines the outcome. And the fleet, by the arithmetic of turnover, does not respond at narrative speed.

A billion and a half machines do not retire because a better one exists. They retire when they stop working.

Two hundred fifty years

Every transformative technology in the last two hundred fifty years followed the same cycle. The narrative compressed the timeline. Capital flooded in at narrative speed. The material constraint held. The bust came. Then the actual transition happened, slowly, funded by patient capital that understood the tonnage.³⁶

In the 1790s, more than eighty canal bills were approved by Parliament in just four years. Capital flooded in. Dozens of canals started simultaneously. The share market collapsed. War with France drained liquidity. Many schemes came to nothing. The canal network got built anyway. Over the next forty years. By operators who understood the engineering.

In the 1840s, Parliament passed 263 railway acts in a single year.³⁷ Railway shares dropped 50 percent between 1846 and 1850. Roughly a third of authorized railways were never built. Families lost their savings. The network got built. Over decades. The companies that survived the bust bought the infrastructure at pennies on the dollar.

In the United States, railroad track expanded from 30,000 miles in 1860 to 163,000 miles by 1890.³⁸ In 1893, one-fourth of all rail mileage went into receivership. Over 125 railroads failed in a single year. Fifteen thousand companies and five hundred banks collapsed in the panic. The rail network survived. Consolidated and operated profitably by the companies that bought it out of bankruptcy.

Thomas Edison demonstrated the lightbulb on December 31, 1879.³⁹ Three thousand people came to see it. Pearl Street Station, the first commercial power plant, opened September 4, 1882. The promise was immediate: electricity changes everything.

The timeline was not. In 1930, 83 percent of urban American homes had electricity. Ten percent of rural homes did. The Rural Electrification Act passed in 1936. By 1945, half of American farms had electric service. The figure did not reach 90 percent until 1953. Seventy-four years from Edison's lightbulb to near-universal rural electrification. The people selling the future in 1885 were right about the destination. They were wrong about the distance by half a century.

In 1954, Lewis Strauss, chairman of the Atomic Energy Commission, predicted that nuclear energy would produce electricity "too cheap to meter."⁴⁰ He was speaking for himself. "A serious governmental body ought not to indulge in predictions," he noted. "However, as a person, I suffer from no such inhibition."

France responded to the 1973 oil crisis by standardizing a single reactor type and building fifty-eight of them over twenty-five years.⁴¹ The Messmer Plan. One design. Framatome pressurized water reactors. Institutional commitment measured in decades. Nuclear now provides roughly 65 percent of French electricity. France succeeded because it operated at material speed with patient capital and a timeline that matched the engineering.

Three Mile Island in 1979 triggered the cancellation of more than fifty planned American reactors.⁴² No new reactor was ordered in the United States from 1979 through the mid-1980s. The narrative boom in nuclear promises had created the public confidence that the accident destroyed. The regulatory overreaction that followed did not distinguish between the technology and the timeline it had been sold with.

Every transformative technology in the last two hundred fifty years arrived. None arrived at the speed the narrative promised. And in every case, the narrative speed created a boom-bust cycle that delayed the actual arrival.

The exception

The internet is the one case where the narrative timeline was roughly right.

Radio took thirty-eight years to reach fifty million users. Television took thirteen years.⁴³ The internet reached mass adoption faster than any communication technology in history. ChatGPT reached one hundred million in two months.

The internet is almost pure information. No tonnage. No fleet to turn over. No fabrication plants to build. No grid to expand. No supply chain that depends on four countries. The marginal cost of adding a user approached zero. Adoption really was exponential, because nothing in the physical world constrained it.

This is why the belief persists.

The generation that watched the internet replace the phone book, the newspaper, the record store, the travel agent, and the encyclopedia in fifteen years learned to expect exponential adoption as the default. It was the default. For software. For information products. For anything where the product has no mass and the delivery has no floor on marginal cost.

The moment you extend that expectation to energy, transportation, semiconductors, or manufacturing, atoms intervene. Software replacing software moves at software speed. Software replacing physical infrastructure does not. The internet replaced information products. The energy transition is trying to replace 1.5 billion machines, millions of miles of pipeline, and a century of built infrastructure. The narrative has no weight. The infrastructure does.

The internet trained the intuition. The physical world does not obey it. The one time the curve was right is the reason everyone assumes it always is. Nobody went back to check whether the same math applies when the product has mass.

The future will arrive. It always does.

Electricity arrived. The railroad arrived. The automobile arrived. Nuclear arrived, in the one country that understood the tonnage and built at its pace.

But it arrives at material speed, not narrative speed.

The gap between those two speeds is filled with bankrupt companies, stranded capital, transit agencies waiting for buses that were never delivered, communities promised fabrication plants that are years from producing a single chip, and workers who were told the future was here.

The cost of the gap does not land on the people who sold the compressed timeline. The investor deck was presented. The political speech was given. The capital was raised. The cost landed on the workers at Lordstown. On the transit riders in Miami-Dade. On the communities around the Intel Ohio site, five years into a wait that was supposed to be over.

One number was on every slide deck. The other determined the outcome.

86 to 82. Twenty-three years. Trillions of dollars. Record fossil fuel consumption. The largest sustained investment of the century.

Four points.

The future is coming. It is just heavier than the story.