By Ernie Tedeschi

Sometimes there are financial narratives that aren’t worth debunking—they have a grain of truth—but they require an appropriate level of skepticism. Three such narratives are widespread today.

The first is that AI is driving US GDP growth. The second that artificial intelligence weakens the labor market. And the third is that the recent increase in consumer spending comes mainly from high incomes. None of these narratives are false, but they all contain several uncertainties, so we must be careful how we interpret them and precise about what we know and what we don’t.

In terms of GDP, there is no doubt that investment in AI has grown significantly. At the end of 2021, businesses invested 1 trillion. dollars, annually, in software, IT equipment and data centers. By the second quarter of 2025, this amount had reached 1.4 trillion. dollars, an increase of 40% in 4 years.

But there is one critical detail: a significant portion of this AI investment is imported, which means that the contribution of AI investment to GDP growth is not as large as it seems (because GDP is a measure of value produced within US borders).

The final contribution of real-time imports cannot be recorded with absolute certainty, but the circumstantial evidence is strong. From the beginning of 2024, total imports of IT equipment and software increased by $189 billion annually, while domestic business investment in these categories increased by $172 billion and consumer spending by just $29 billion.

This demonstrates that much of the AI ​​investment “boom” was fueled from abroad.

Of course, it’s not necessarily a bad thing for companies to import equipment that allows them to expand here in the US – quite the opposite – but it affects the calculation of GDP.

Excluding the impact of imports, AI-related products – software, IT equipment and data centers – accounted for 1.3 percentage points of annual real GDP growth compared to 1.6% in the first half of 2025 in consumer spending, business investment and exports, an impressive figure. However, if the increase in AI-related imports is removed, this contribution drops to 0.5 percentage points.

The percentage remains impressive, but not as much as it seems at first glance. To put that 0.5% in perspective, it’s about the size of the negative impact on growth in 2025 that the Yale Budget Lab expects from tariffs.

The second narrative argues that AI leads to rising unemployment rates, especially among new college graduates. Recent research has found significant effects of AI on new hires, with early-career workers in occupations with high AI “exposure” actually seeing a decline in employment, while more experienced workers in the same fields have maintained or increased their employment rate.

However, the data picture is not uniform. My Budget Lab colleague Martha Gimbel and her colleagues recently found, for example, that changes in the composition of occupations from 2022 are commensurate with earlier technological transitions, such as the penetration of personal computers or the Internet.

They also found that neither the proportion of workers in occupations with high AI exposure nor their unemployment duration has changed significantly since AI began to gain a prominent role.

Furthermore, it is clear that even if AI is affecting the labor market, it is not the only cause. The 12-month moving average of the unemployment rate hit its lowest point in June 2023 at 3.5% – since then, it has risen by 0.6 percentage points.

The largest increases in unemployment rates are recorded in both the occupations with the most and least exposure to automation by artificial intelligence. This suggests that, even if AI is indeed responsible for the recent labor market slowdown, it is not the only cause. It’s possible that new hires are being disproportionately affected right now, simply because they’re usually the first to take the strain whenever cracks appear in the labor market.

There is also concern that American growth is “K-shaped,” where higher-income consumers primarily fuel economic growth while lower-income households struggle. This concern is not unfounded. If the labor market does slow, low-income workers are the first to be hit.

However, I would recommend extreme caution regarding estimates of the distribution of consumer spending, especially when they come from individual private sources. While the government publishes reliable aggregate consumer spending figures every month, reliable spending figures, with the necessary breakdown by income group and validation of private data, are often years behind schedule.

The data we have so far complicates this narrative. Bureau of Economic Analysis data through 2023 shows that the poorest households accounted for 9% of consumer spending – the highest share since the pandemic era.

The Federal Reserve Bank of New York’s Survey of Consumer Expectations shows that households with incomes under $50,000, those with incomes between $50,000 and $100,000 and those with incomes above $100,000 reported similar increases in spending – about 4% – during the 12 months to in August 2025.

Moreover, although consumer spending does not represent wages, it is highly correlated. While inflation-adjusted weekly earnings at the 10th percentile lag somewhat behind higher-paid workers, they have still risen by about 2% over the past two years. Moreover, the bottom fifth’s share of total weekly earnings has actually increased over the past two years, while the top fifth’s share has fallen.

Again, none of these narratives are wrong. Each of them contains nuggets of truth but also asterisks, and nuances that don’t fit newspaper headlines. The American economy is in a phase of change and uncertainty, and no simplistic explanation will suffice.