US Factories Shipped a Record $86 Billion in Core Capital Goods in September

American factories shipped a record $86 billion of core capital goods in September, according to The Kobeissi Letter, and the strongest growth came from the machines powering artificial intelligence. Core capital goods — the machinery, computers and equipment companies buy to expand, excluding defense hardware and aircraft — rose 11.6% from a year earlier. That is the fastest annual growth since July 2021. Over the past three months, shipments climbed at a 16.5% annualized pace, the quickest since 2011 once you set aside the pandemic rebound.

The number that explains the boom sits inside the aggregate. Orders for computers, servers, storage devices and related equipment jumped 27% year over year. That is the strongest gain since January 1997, when the dot-com expansion was picking up speed.

What Is Driving the Investment Boom?

One category of buyer, not a broad industrial recovery. The 27% surge in computing equipment orders dwarfs the growth in the rest of the capital goods data, which means a small group of companies — largely the hyperscalers (Amazon, Microsoft, Google and other operators of massive cloud data centers) — is responsible for a disproportionate share of the boom. Their spending now holds up a large share of U.S. economic growth.

The government's broader statistics show the same concentration. Real gross domestic product grew at a 3.4% annualized rate in the third quarter. Gross private investment rose 19.4%, its biggest jump since early 2025. Private nonresidential fixed investment — the measure tracking business spending on structures, equipment and software — reached a record annual rate of $4.95 trillion, up 12.1% from a year earlier.

The takeaway: AI infrastructure spending has moved from a tech-sector story to the main engine of U.S. investment growth.

Where Does the Money Come From?

Corporate balance sheets can carry the load, at least for now. Corporate profits ran at a $4.7 trillion annual rate in the second quarter, up 20.8% year over year, which gives large companies room to fund capital projects from earnings rather than debt.

Hyperscalers are expected to spend $845 billion on data centers this year and close to $1.4 trillion in 2027, according to estimates discussed on 24/7 Wall St.'s AI Investor Podcast. If those figures hold, spending would nearly double within two years — a pace that invites scrutiny of what buyers expect in return.

The takeaway: the boom is cash-rich today, but its projected growth rate assumes AI returns keep justifying it.

Why Can't Suppliers Keep Up?

Equipment makers cannot build fast enough. Unfilled orders have grown for more than a year, meaning demand is running ahead of production. That backlog can sustain the cycle for electrical, cooling and networking suppliers well outside Silicon Valley — companies that sell power equipment, heat management and connectivity hardware to data center operators.

But not all of the benefit stays domestic. Imports rose 19.7% in the third quarter while exports fell 5.7%. Every imported server rack subtracts from the domestic GDP gain, so the headline investment numbers overstate how much of the boom accrues to U.S. factories and workers.

The takeaway: the supply chain is congested in the buyers' favor, but the trade data shows the boom leaks abroad.

What Are the Warning Signs?

Investors have already priced in doubt. Tech stocks just suffered their biggest selloff since Covid on fears that the data center boom could end early, Fortune reported. Morningstar has warned that hyperscalers' AI investment will outpace near-term earnings — a gap that can only persist if future AI revenue materializes.

Local opposition is the second constraint. In the midterms, voters approved restrictions in nearly every data center ballot measure, with one notable exception in Hardwood, North Dakota. Communities are increasingly unwilling to host the physical footprint of the AI economy.

Power is the third. The Energy Department projects data centers could use up to 12% of U.S. electricity by 2028, and retail electricity prices have risen faster than inflation since 2022. Rising utility bills give voters and regulators a concrete grievance against further expansion.

The structural weak point, though, is simpler. AI companies could cut back because returns disappoint, because borrowing gets more expensive, or because computing capacity exceeds demand. If they do, equipment orders could fall quickly, and the spending that is supporting growth today could become an economic drag. The same concentration that makes the boom powerful makes its reversal fast.

The takeaway: three independent brakes — investor sentiment, local politics and electricity supply — could each slow the cycle on their own.

Which Numbers Will Decide 2027?

Two figures matter most for anyone tracking whether this boom continues.

• Unfilled orders in the durable goods reports. A shrinking backlog would be the earliest sign that buyers are pulling back, arriving well before any weakness shows up in GDP or corporate earnings.
• Capital spending guidance from the largest cloud companies, due in their fourth-quarter earnings reports early next year. Higher budgets would signal the cycle has room to run.

The alternative scenario is concrete. If hyperscalers hold budgets flat while fighting local opposition and power shortages, the record factory shipments of September may turn out to be the high point of the cycle rather than a waypoint — the moment the 15-year investment boom peaked with AI's name on it.

The takeaway: watch the backlog and the guidance; they will reveal the boom's direction before the headline numbers do.