Diesel Is the Anvil, AI the Hammer—and the U.S. Economy Is Caught Between Them

A single click on the "buy" button creates the illusion of an almost weightless economy. But in the continuing absence of teleportation, the package must still pass through a warehouse, travel along a highway and cross someone’s doorstep. Much the same is true of artificial intelligence: the answer appears in a second, but behind it stand chips, copper, power plants, cooling systems, vast data centres and ever-growing amounts of borrowed money.

Those two forces are now tightening an economic vise around the United States. At the pump, Americans spend today’s income, on the stock market, investors trade in the "future." Diesel is the anvil driving up the cost of the physical world, while the thinking machine has become a hammer whose value depends on whether the promised world will ever generate enough revenue. Caught between them are consumers, businesses, retirement accounts and a political establishment now just weeks away from the congressional elections.

A tax Congress never passed

The national average price of diesel reached $6.51 a gallon on September 21, a staggering 76 percent higher than a year earlier.^[1] Gasoline rose 41 percent over the same period. According to a model developed by researchers at Brown University, Americans have spent $112 billion more at the pump since the war against Iran began than they would have without the resulting energy shock. Remarkably, diesel alone accounts for $51 billion of that sum, even though far less of it is consumed than gasoline.

The paradox is particularly cruel: the world’s largest oil producer cannot simply produce enough of the fuel it needs. The problem is not merely the supply of crude oil, but the bottleneck between the barrel and the fuel tank. US refineries are currently operating at an extraordinary 97 percent of capacity. The last major refinery of comparable scale began operating in 1977, while building a new one could take a decade. Record refining margins are delivering windfall profits to Marathon and Valero, but even spectacular earnings cannot conjure a new distillation column overnight.

Inventories, meanwhile, are disappearing. In early September, they fell to their lowest seasonal level since comparable records began in 1982. The US Energy Information Administration expects them to fall below 100 million barrels and remain below the five-year average through the end of this year and for much of the next.^[3] Empty storage tanks are now being offered for lease because traders have nothing to put in them. Shortages are rarely this literal.

The consequences become clearest when the vast national figures are reduced to a single journey. The average cost of operating a truck last year was about $2.34 per mile.^[4] At six miles to the gallon, the current increase in diesel prices adds roughly 47 cents to every mile. Over a thousand-mile journey, that means an additional $470 before warehousing, retail and last-mile delivery costs are even counted. Major freight carrier Knight-Swift recorded an operating margin of just six percent in the latest quarter. It cannot absorb a cost like that. It has to pass it on.

Even Amazon, the ultimate symbol of the supposedly immaterial internet economy, cannot escape physics. In a single quarter, it spent $27.9 billion on fulfilment—more than its retail division earned in operating profit.^[5] As we already know, diesel powers a large share of freight transport. What matters more is what happens next: the carrier sends the bill to the retailer, the retailer builds it into the price, and the customer pays it at the checkout. Diesel has become a tax Congress never passed.

The Fed cannot build a refinery with interest rates

This is where the central bank’s problem begins. The Federal Reserve can raise interest rates and suppress demand, but an interest rate cannot build a refinery or fill an empty storage tank. Higher transport costs simultaneously push up prices and erode purchasing power, producing precisely the stagflationary combination that is hardest to treat. In September, the Fed raised its benchmark rate to a range of 3.75 to four percent^[6], while yields on ten-year Treasury bonds climbed above five percent for the first time since 2007.

On paper, the economy still appears resilient. Retail sales rose 1.2 percent in August, while third-quarter growth estimates have been revised towards three percent. But some of that increase is simply the result of higher fuel prices. Consumer spending is increasingly being sustained by rising stock-market wealth, falling savings and the depletion of previously accumulated cash reserves, even as real wages weaken. In other words, the American consumer is still walking—but increasingly on something other than his own strength.

The thinking machine as an engine of statistical growth

Now the hammer joins the anvil. According to Bloomberg Economics, investment related to artificial intelligence accounted for roughly half of the approximately two percent growth in US GDP over the past year.^[2] But, as you might suspect, that does not mean AI is already generating enormous gains in productivity. GDP counts the concrete as soon as it is poured, the chip as soon as it is purchased and the data centre as soon as it is built. The statistics do not wait to discover whether, five years from now, that facility will be generating profits or gathering dust. For now, Nvidia alone can confidently count its winnings: it is selling components by the millions into the AI boom. Everyone else in the chain still has something to prove.

Wall Street, meanwhile—and this would hardly be the first time—has already booked tomorrow’s profits as today’s wealth. Since the launch of ChatGPT in late 2022, the market value of companies in the S&P 500 has increased by nearly $33 trillion. Now for the crucial detail: roughly three-quarters of the index’s total gains came from just twenty or so companies, most of them tied to AI. What does that tell us? AI has become an engine of the US economy, yet AI may also be a bubble. No lengthy explanation is needed to imagine what would happen if both propositions prove correct—and the first one already has.

In 2022, Alphabet, Amazon, Meta and Microsoft were collectively investing about $150 billion a year. Today, the figure is roughly five times higher, and their combined capital expenditure is expected to exceed $1 trillion in 2027. Their aggregate free cash flow, which stood at a positive $230 billion in 2024, could turn negative by roughly $50 billion, according to analysts’ estimates. On top of that are nearly $2.4 trillion in multiyear contractual obligations for energy, leases and infrastructure. In other words, the great cash-generating machines once built on virtually weightless software are becoming the most capital-intensive industry of the digital age.

When expensive money collides with expensive energy

The two stories therefore converge in a deeply unpleasant chain reaction. More expensive diesel fuels inflation. More persistent inflation keeps interest rates high. High interest rates make the vast AI investment cycle more expensive. To put that in concrete terms, every additional percentage point in financing costs means another $1 billion in annual interest for every $100 billion of new debt. If companies respond by slowing construction, orders for chips, electricity, cooling equipment and building work immediately weaken. So does the portion of GDP growth that has been concealing the fragility of the rest of the economy. Shares fall, the sense of wealth evaporates, and consumers feel the prices coming off the anvil even more acutely.

Technology can win while investors lose

History offers little comfort. Railways genuinely transformed nineteenth-century America, yet railway mania still helped bring down banks and trigger the Panic of 1873. The internet fulfilled almost every grand promise made by its pioneers, but that did not save the companies that built too much fibre-optic cable with borrowed money during the dot-com frenzy. The eventual winners later acquired and used that infrastructure at fire-sale prices. A transformative technology can prevail even as its first financiers lose everything.

We have previously examined all of this in considerably greater detail (see: A bubble running on four badly mismatched clocks: AI can change the world and still prove a catastrophic investment).

The political clock is ticking even faster than the economic one. The congressional elections take place on November 3. Trump’s approval rating has fallen to 32 percent, just 17 percent of respondents approve of his handling of the cost of living, and Democrats lead Republicans by 43 percent to 35.^[8] The cost of his Iran policy has already appeared at the pump, on freight invoices and on supermarket shelves. Voters do not need to understand refinery margins or the free cash flow of technology giants to feel their combined effects.

How many floors must slip before the election changes?

Will the house of cards collapse before the long-awaited congressional elections? Fortunately for Trump, a complete breakdown in such a short period is not the most likely scenario. Yet November is both very close and still a long way off, depending on how one looks at it. The stock market certainly does not need several quarters to change its mind. One weak set of results from an AI giant, one delayed initial public offering, another interest-rate increase or a serious failure in the already overstretched refining system could be enough. The house need not collapse to its foundations to change an election. A few floors beginning to slide may be all it takes.

Until now, America has sustained itself through an unusual equilibrium: it could tolerate an increasingly expensive physical world because the financial world kept growing richer. Diesel is now eroding the first, while the mounting cost of AI is beginning to cast doubt on the second. The anvil is not moving. The hammer is already raised.