How an AI investment supercycle props up U.S. growth as traditional engines sputter — October 2025

The American economy in late 2025 feels like a magic trick you can’t quite explain. Growth looks resilient again, markets are buoyant, and factories are rising from Arizona to upstate New York. But pull back the velvet curtain and a single force dominates the stage: artificial intelligence. Without the extraordinary capital pouring into AI chips, data centers, and model training, the United States might be flirting with stall speed rather than cruising at a seemingly comfortable clip.

Consider the recent macro collage. After a shallow contraction in the first quarter, output rebounded strongly in the spring. Productivity has perked up, unit labor costs cooled, and the Federal Reserve has at last started to glide rates lower after two bruising years of tightening. Yet consumer momentum has softened at the margins, credit is tighter, and the labor market has lost some of its once‑feverish heat. Thread those strands together and you get an economy that—outside of AI—looks fragile, if not fatigued.

That fragility is masked by a once‑in‑a‑generation capital‑spending boom centered on AI infrastructure. The giants of cloud computing and social platforms are plowing tens of billions apiece into server farms, networks, and custom silicon to train and deploy ever‑larger models. Their suppliers—chip designers, contract manufacturers, power‑equipment makers, and real‑estate developers—have become the new growth complex. In a narrow sense, AI is an “investment story.” In a wider sense, it has become the economy’s mood music, lifting risk appetite, wealth effects, and corporate animal spirits.

The AI Supercycle, in Numbers

For now, the numbers favor the optimists. Real GDP snapped back after a wobbly start to the year, and measured labor productivity posted one of its better quarterly showings of the expansion, helped by output growth that outpaced hours worked. Meanwhile, the Fed cut rates in September for the first time this year, nudging the policy rate into a lower range and easing financial conditions at the margin. The headline unemployment rate, while up from 2023 lows, remains historically moderate.

Behind those aggregates, however, the composition of growth has shifted. Business equipment and intellectual‑property products tied to AI are doing outsized work. Capital‑expenditure plans from the “Big Three” cloud providers now read like public‑works budgets: vast sums earmarked for GPUs, custom accelerators, fiber, and megawatt‑hungry campuses. Suppliers from chip designers to server integrators are reporting record backlogs. Even power markets—long an afterthought for Silicon Valley—have become a strategic chokepoint, with utilities racing to accommodate data‑center demand and developers scouring for generation and transmission capacity.

If that sounds like a narrow base, it is. The new industrial policy of the 2020s—subsidies for semiconductors, batteries, and clean tech—has married itself to the AI gold rush. The result is regionally concentrated booms in construction and manufacturing, even as other interest‑sensitive sectors—commercial real estate, parts of consumer discretionary, small‑business capex—struggle to keep pace. One can welcome the investment while still asking whether it can carry the whole economy indefinitely.

What’s Holding Back the “Old” Economy

Start with consumers. Excess savings built up during the pandemic have largely dwindled outside the top income cohorts. Real wage gains have improved as inflation cooled, but the reopening sugar high is gone. Households face higher carrying costs on revolving credit, student‑loan repayments have resumed for many borrowers, and the appetite for big‑ticket durables has normalized. Retailers continue to guide cautiously, noting bifurcation between higher‑ and lower‑income shoppers. Travel demand is still healthy, but leisure‑spending growth has decelerated.

Housing is a similar study in cross‑currents. The Fed’s gradual pivot has eased mortgage rates from last year’s peak, but affordability remains stretched and inventories tight. New construction has been more resilient than resales, supported by builder incentives and the spillover from factory‑town growth in a handful of states. Yet housing’s multiplier effects look more muted than in past cycles. In credit markets, banks have tightened standards and the cost of capital remains elevated for smaller firms, even as mega‑caps tap bond markets at fine spreads.

The labor market, finally, has cooled to a sustainable simmer. Job openings have fallen from extreme highs, wage growth has normalized, and quit rates have come back to earth. That’s good for inflation management, but it also means less fuel for consumer demand. Productivity gains can offset that—indeed, they may be the best part of the 2025 story—but it’s too early to declare a new era until those gains broaden beyond a handful of large firms and sectors.

AI: Productivity Engine or Mirage?

The pivotal question is whether AI can deliver the productivity uplift implied by the capex boom. History warns that general‑purpose technologies diffuse slowly; the payoff arrives after complementary investments in skills, processes, and organizational change. The last decade’s cloud migration offered a preview: substantial gains for firms that re‑engineered workflows, modest ones for laggards who merely lifted‑and‑shifted. Generative AI poses the same test at far greater scale. Early pilots show promise in code generation, customer support, and document synthesis. But widespread gains depend on data quality, human‑in‑the‑loop design, and a long tail of boring automation tasks—not splashy demos.

In the meantime, AI behaves like any other capital deepening: it boosts demand for upstream goods (chips, servers, power) and for specialized labor (ML engineers, electricians, construction crews). That’s stimulative. It also concentrates profits in firms that can amortize enormous fixed costs across global platforms. That’s great for equity indices and for the households who own them, but it risks widening disparities. In other words, AI can raise GDP while leaving the median business or worker unsure what changed.

The Stall‑Out Scenario

What happens if the AI boom underdelivers—or simply pauses? Imagine GPU supply catches up just as corporate budget fatigue sets in. Utilization rates dip, project ROIs take longer to pencil, and the hyperscalers slow orders to digest capacity. Construction of new data‑center campuses hits grid bottlenecks. If, simultaneously, consumer spending remains tepid and exports face a weaker global backdrop, the growth cushion thins quickly. A few quarters of middling job gains and soft core demand would feel very different without the loud drumbeat of AI capex.

A separate risk is policy and politics. Data‑center siting is already running into local‑power constraints and permitting fights. If electricity prices rise or reliability suffers, the social license for hyperscale growth could erode. On the fiscal side, a divided Congress and repeated funding dramas—up to and including temporary shutdowns—inject volatility into public‑sector demand and jam the flow of timely economic data that businesses use to plan. Even well‑intentioned industrial policy can turn lumpy, as subsidies bunch and supply chains struggle to sequence equipment, labor, and grid connections.

What Would a Balanced Expansion Look Like?

The antidote to an AI‑dependent economy is not less AI but more breadth. Three things would help. First, diffusion: push practical, smaller‑scale AI tools into the middle of the economy—healthcare practices, municipal offices, small manufacturers, logistics firms—paired with training and process redesign. Second, infrastructure: accelerate power‑grid upgrades and permitting reform so that data‑center demand doesn’t crowd out other electrification goals. Third, inclusion: help households on the wrong side of rate hikes and price shocks through targeted relief and pro‑work policies that raise labor‑force participation and mobility.

Monetary policy can help by keeping the disinflation trend intact while not leaning too hard on a still‑healing supply side. Rate relief should feed through to housing and small‑business credit over time. On trade and immigration, stable rules and improved legal pathways would ease bottlenecks in skilled and unskilled labor alike, supporting both construction and services. These are not headline‑grabbing moves, but they are how you turn a narrow boom into a broad expansion.

The Bottom Line

The excitement around AI isn’t pure smoke; it’s a real investment cycle with real output behind it. But it is also a crutch—one that lets the economy limp past weak spots without healing them. If the crutch holds, we may stride into a more productive, less inflation‑prone era. If it cracks, the limp will be obvious. Either way, it’s time to stop marveling at the trick and do the harder work of strengthening the rest of the act.