An Overly AI-Dependant Economy?

Invictus here. During Friday’s sell-off, I wanted to review the major economic factors to see whether the market is foreshadowing a slowdown or worse. The TL:DR: The United States is still growing, but at a slower pace. Compared with the 1990s, growth has become increasingly dependent on a narrower range of economic activity, which may leave us more vulnerable than headline GDP figures suggest. Let’s dive in: The American economy has a breadth problem.

At first glance, the numbers seem reassuring. Real gross domestic product continues to expand (3.4% SAAR), consumers keep spending, and businesses keep investing. But beneath those headline figures lies a potentially troubling development: an unusually large share of recent economic growth comes from information-processing equipment and software.

That raises an important question: Is today’s economy as fundamentally robust as the American economy of the 1990s was, or is it running on fewer cylinders — and therefore more fragile and susceptible to shock? Let’s take a look at the various components of GDP and compare what we’re seeing today against what we saw in the ’90s.

All data comes from BEA table 1.5.2, Contributions to Percent Change in Real Gross Domestic Product, Expanded Detail, and BEA table 1.5.5, Gross Domestic Product, Expanded Detail. I uploaded those tables (Q1 1990-Q3 2026) to ChatGPT for its analytical capabilities and, frankly, to save me countless hours in Excel.

As the chart below shows, an unprecedented amount of economic activity is coming from a single sector:

Two Expansions, Two Very Different Economies

Consider two periods: the five years from 1995 through 1999, and the eight quarters ending in September 2026.

BEA data show annualized real GDP growth averaged about 4.17% during the late-1990s expansion. Over the latest eight quarters, it has averaged 2.34%. The U.S. economy is obviously growing more slowly today. Let’s look at the components of that growth and assess how robust it is. Are we seeing a multi-pronged expansion, or are we vulnerable to a sudden slowdown in one area? If that area slows, how disruptive might it be to the economy as a whole?

Will weakness in the economy’s strongest component produce a meaningful slowdown or recession?

In the late 1990s, investment in information-processing equipment and software contributed an average of 0.68 percentage points to quarterly annualized GDP growth. During the most recent eight quarters, that contribution increased to 0.81 percentage points. In other words, technology-related investment is contributing more to GDP growth today, even as overall growth has slowed considerably.

Measured relative to total growth, the contrast is striking: 1995–1999: Information-processing equipment and software accounted for the equivalent of approximately 16% of average GDP growth. 2024 Q4–2026 Q3: Their contribution was 35%.1

The table below, comparing the mid-90s to today, illustrates how the composition of GDP has shifted:

What Happens When We Subtract Technology?

Let’s perform a simple thought experiment to highlight the issue: Subtract the contribution of information-processing equipment and software from overall GDP growth. 2

From 1995–1999, the rest of the economy contributed about 3.49 percentage points to annualized growth. During the latest eight quarters, that figure was just 1.53 percentage points. That’s a decline of approximately 56%.

This is not to say the “ex-IT” economy is contracting; it is to say its contribution to overall growth is substantially smaller than it was in the ’90s. Other BEA figures reinforce the concern. Residential investment contributed positively during the late 1990s but negatively during the latest eight quarters.3

Consumer spending remains an important source of growth, but it contributes far less. These are all among the important legs supporting the chair that is the US economy. A couple appears wobbly.

The late-1990s economy had multiple powerful engines. Today’s economy appears to lean more heavily on fewer of them. The AI Question

The AI question matters because the current technology investment boom is not necessarily self-sustaining — certainly not indefinitely. Companies are spending enormous sums on compute capacity, data centers, networking equipment, and software in anticipation of future demand and productivity gains.

Those expenditures contribute to GDP today. Whether they eventually generate adequate financial returns is a separate question. If expected returns disappoint, businesses may scale back investment, and we have already seen signs of this. Reduced spending would directly weaken GDP growth and could indirectly affect employment, construction, suppliers, and consumer spending, with economic consequences that could extend well beyond the technology sector.

This is not a prediction that the AI boom will collapse; I have no idea how this story unfolds (though I do have thoughts). Nor is it proof that the investment is irrational. Transformative technologies often require substantial capital expenditures before their full economic benefits emerge. But an economy unusually dependent on a particular investment cycle is exposed to that cycle’s fortunes — and I believe we are seeing fraying begin. An Important Historical Qualification

The late 1990s were hardly free of speculative excess. The dot-com bubble eventually burst, and a recession followed in 2001. Nor were all the 1990s equally strong. The early decade included a recession and a sluggish recovery. Our comparison deliberately focuses on the powerful expansion of 1995–1999.

An important distinction also exists between concentrated growth and systemic fragility. Today’s households, banks, and corporations may, at least for now, possess financial strengths that help cushion an investment reversal. The US consumer continues to defy expectations and has been remarkably resilient. GDP contribution figures alone cannot settle the concentrated growth v systemic fragility question.

Still, I believe the historical comparison is instructive. Even during an extraordinary technology boom, the late-1990s economy saw stronger contributions from the rest of its economic activity than today’s does. The Bottom Line

The question isn’t whether the American economy is growing; it is. My question is whether that growth rests on a sufficiently broad foundation to continue organic expansion for the foreseeable future. BEA data suggest today’s expansion is slower and more concentrated in information-processing equipment and software than the late-1990s expansion. That does not establish that a recession is imminent. It does suggest that a reversal in technology investment could be especially consequential.

A healthy economy requires more than a single powerful engine. The data today suggest we have become overly dependent on that single driver of economic expansion…

FOOTNOTES:

1. This is not to say that artificial intelligence accounts for 35% of economic growth; it does not. The BEA categories encompass considerably more than AI. But they capture important components of the extraordinary investment boom surrounding computing infrastructure and software.

2. Lots of AI Capex has shifted from other sectors of investment into technology because it’s the new hotness, and managers tend to chase performance. If the AI CaPex boom wasn’t happening, much of this pending capital would flow to other sectors, making this an imperfect counterfactual…

3. Despite mortgage rates comparable to those of the late 1990s, today’s housing market is burdened by home prices that have risen far faster than household incomes, making homeownership substantially less affordable. Compounding the problem, millions of existing homeowners are reluctant to sell and surrender the exceptionally low mortgage rates they locked in years ago, which is suppressing both housing inventory and transaction volume. Nonresidential structures show a similar reversal.

Data: U.S. Bureau of Economic Analysis, National Income and Product Accounts, Table 1.5.2, revised October 29, 2026. Comparisons use averages of quarterly annualized GDP growth contributions. Information-processing equipment is line 31; software is line 38. Percentages comparing contributions with total GDP growth are descriptive ratios, not exact cumulative growth shares.

The post An Overly AI-Dependant Economy? appeared first on The Big Picture.