Why the smartest businesses are hitting the gas and the brakes at the same time
— Opinion column by manufacturing expert Buckley Brinkman
Reality is catching up with AI’s true believers.
The excitement about AI’s market potential, its promised transformations, and the riches it is supposed to deliver is colliding with the reality of business returns, competitive pressure, and the yawning gap between addressable markets and real ones. The top AI players know exactly what they’ve invested, what returns their investors expect, and what happens if they miss those stretch goals.
Microsoft, Amazon, Alphabet, and Meta plan to spend $725 billion on capital expenditures in 2026, up from $410 billion in 2025, with roughly 75 percent of that money supporting AI. Beyond the hyperscalers’ own balance sheets, a tangled web of side deals with Oracle, Nvidia, AMD, and CoreWeave put OpenAI on the hook for more than $800 billion in additional commitments. Nvidia added to the pile in late July, floating up to $500 billion in fresh backing for a single OpenAI data-center campus. And much of this build-out runs on borrowed money: hyperscalers alone will add roughly $400 billion in new debt in 2026. AI’s growth must justify an enormous bet.
Analysts disagree on how much revenue would make that bet pay off, but none of them are
optimistic. AI’s cumulative revenue today sits around $150 billion. NYU marketing professor Scott Galloway puts the incremental revenue the industry needs at more than $2.5 trillion — 15 times current performance. Sequoia Capital’s David Cahn pegged the “Growth Gap” at $600 billion back in 2024. Today, he puts that gap at $3 trillion. Both point to the same conclusion: the AI industry has a serious demand problem.
That demand problem is about to get worse. Cheaper Chinese models are eating into the market as businesses realize they don’t need a Mercedes model to run a Kia-level agent. Companies are already experimenting with ways to cut their AI spending without sacrificing performance. Chinese models’ share of global AI token usage jumped from roughly 1 percent in 2024 to more than 60 percent today, and that shift puts real revenue pressure on the leading U.S. model providers.
History has never been kind to early investors in transformational technology. Most of them watched overexpansion wipe out their equity. Railway investment reached 6 to 10 percent of GDP in the 1800s, and redundant track construction — 26 competing lines were built to connect Chicago and St. Louis alone — permanently depressed returns and wiped-out investors. The telecom and fiber-optic boom of the 1990s buried more than 80 million miles of cable and nearly $1 trillion in capital; more than 85 percent of that fiber sat dark, and telecom stocks lost more than $2 trillion in value when the bubble popped. The dot-com crash that followed erased more than $5 trillion in market value, because revenue growth never caught up to the spending it was supposed to justify.
The AI buildout is already bigger than the telecom boom and closing in on railway mania. Data center capital spending is on pace to exceed 3 percent of GDP in 2026, and by some analyst estimates, AI investment now accounts for more than 70 percent of U.S. GDP growth. That’s a lot of eggs in one basket.
There’s one more difference, and it cuts against AI. The infrastructure from prior booms — rail lines, fiber cable, internet capacity — kept paying off in the economy for decades after the busts. Microchips don’t get that grace period. They carry a much shorter useful life and may be obsolete before demand ever catches up to the supply being built today.
That combination calls for a business to hit the accelerator and the brake at the same time. If you haven’t put AI to work in your business yet, you’re already behind. Pick one process this quarter — customer service, operations, marketing — and put AI against it. Measure what it does to your bottom line, then expand from there. Waiting for certainty means waiting too long. Your competitors are already making money on their AI advances.
If your revenue depends on supplying the AI buildout itself, start tapping the brakes now. Stress-test your backlog against a slowdown in hyperscaler capex. Tighten your credit terms and watch your trade receivables closely — a customer’s growth story is not a substitute for its ability to pay. Run scenario plans for your strategic and operating plans so a market jolt doesn’t catch you flat-footed.
The AI boom will eventually sort winners from losers – the way every infrastructure bubble before it has. The businesses that prepare now — leaning into AI’s real productivity gains while guarding against its financing risks — will be the ones still standing when the dust settles. Don’t wait for the correction to start asking these questions.
