One-Third of US GDP Growth Is Now AI, ING Estimates — But It's Capex and the Wealth Effect, Not Consumers

How much of America’s economic growth is actually artificial intelligence? It is the macro question of 2026, and ING has now put a careful, defensible number on it: roughly one-third of current US GDP growth is being driven by the tech investment boom — data centers, servers, and software — even after you subtract the imported chips that make it all run.

The analysis, published on ING THINK by James Knightley, the bank’s Chief International Economist in New York, landed in the Wall Street Journal’s coverage this weekend as the reference point for how entangled the AI buildout has become with the headline economy. Its most striking conclusion is not the size of the contribution but its composition: this is growth built on capital expenditure and stock-market wealth effects, not on consumers actually buying AI products.

The methodology ladder: from 92% down to a third

Estimating AI’s contribution to GDP is harder than it looks, and the ING piece is essentially a guided tour of how much the answer moves depending on what you count.

At the top of the ladder sits Harvard economist Jason Furman’s much-cited estimate that technology and software investment contributed 92% of US GDP growth in the first half of 2025 — a figure from a period when the non-tech economy was barely growing at all. Others pushed back, arguing that the sheer scale of imported technology has to be netted out before you can claim domestic production.

ING walks through three definitions:

• The broadest measure — all information processing equipment, software investment, and data center construction, with no import adjustment — accounts for 50.2% of year-over-year GDP growth in Q2 2026, and averaged 46.6% over the past four quarters.
• A stricter cut — only computing equipment, peripherals, and software — lands at 44% for Q2 2026, with a 43.3% four-quarter average.
• The narrowest and, in ING’s view, fairest measure — the core category minus net imports of computers, peripherals, and semiconductors, deflated using the PPI for electronic components — comes to 36% for Q2 2026, against a 37% four-quarter average.

That final figure is where the “around a third” headline comes from. Knightley acknowledges the residual uncertainty: you cannot cleanly separate AI servers from office PCs, and not all software spend is AI-related. But given how sharply tech capex accelerated after ChatGPT’s release, ING would “only revise down the contribution marginally.”

The import problem: a $40 billion monthly deficit in chips

The gap between the 44% and 36% estimates is the story of global supply chains. Tech-related imports have tripled in value over the past two years to $60 billion per month, while US tech exports rose only from $9 billion to $17 billion per month over the same period. The result: the trade deficit in computers, peripherals, and semiconductors has widened from $20 billion to more than $40 billion per month.

Because chip prices surged on strong demand and limited supply, ING deflates the nominal import values before subtracting them — a detail that separates serious GDP accounting from headline math. Services trade tells a smaller, brighter story: “computer services” exports (which capture foreign AI subscriptions and token purchases) run around $22 billion annualized against $19 billion of imports, with export volumes growing 15% year-over-year in Q2 2026.

Capex is eating the rest of the economy

The investment surge is real and enormous. Real spending on computing equipment and software is up a cumulative 62% since Q1 2022. But ING’s read of the data is that this frenzy has “effectively cannibalised” other business investment: non-tech, non-residential private fixed investment fell year-over-year for six consecutive quarters between Q4 2024 and Q1 2026.

The construction data tells the same story from a different angle. Data centers and electric power generation are now the only sectors of the US economy where construction spending is growing. Residential construction is hamstrung by mortgage rates and affordability; non-data-center technology construction fell back after the conclusion of the Biden-era CHIPS and Science Act’s $52.7 billion program; everything else non-residential is flat-lining.

Jobs: hiring delayed, layoffs attributed

For all the GDP contribution, the labor-market signal is oddly muted — and where it exists, it cuts both ways.

The Federal Reserve’s Beige Book reported in April that “most Districts indicated that AI had not yet significantly impacted overall staffing levels,” though some noted that AI-driven productivity gains had “enabled many firms to delay or reduce hiring.” The straws in the wind: LinkedIn data shows entry-level hiring for graduates down 17% since 2019, and the Bureau of Labor Statistics put the July unemployment rate for recent graduates aged 20-24 at 9.7%, versus just 2.7% for all graduates.

On the layoff side, Challenger, Gray and Christmas reports that AI has been the leading stated reason for US job cuts for five consecutive months, cited in 112,713 job-cut announcements — 24% of the total. The firm adds an uncomfortable observation: “naming AI in a layoff announcement can win over investors while pushing current and prospective employees away,” which means the true AI-driven share of layoffs is probably becoming harder, not easier, to track.

The K-shaped consumer

Perhaps the most consequential finding concerns who is actually paying for AI. Proprietary spending data from Bank of America and PNC suggests that only 2-3% of American households spend money on AI tools, with a typical outlay of $20-30 per month — a rounding error next to internet, TV, and phone bills.

What IS driving consumer spending is the market. Since ChatGPT’s release on November 30, 2022, the NASDAQ has risen 130% and the S&P 500 is up 90%, lifting household financial assets from $109 trillion to $142 trillion. That wealth is heavily concentrated: the top 20% of households by income hold 72% of America’s wealth; the bottom 60% hold 15%.

Using a marginal propensity to consume of roughly 1.5 cents per dollar of wealth — adjusted downward from the Fed’s 2025 estimates to reflect that concentration — ING calculates the AI-driven wealth effect has generated about $500 billion in cumulative extra consumer spending since Q4 2022, or roughly $36 billion per quarter. That equates to around 0.65 percentage points of Q2 2026 consumer spending and 0.44 percentage points of GDP — a meaningful tailwind flowing disproportionately to households that were already the wealthiest.

What it means

The ING decomposition sharpens the debate about the quality of AI-era growth. A third of GDP growth resting on capex and wealth effects is not inherently fragile — railroads and electrification also ran on investment booms — but it does mean the boom’s durability depends on continued corporate capex plans and elevated equity valuations rather than on consumer pull. The piece itself frames the stakes plainly: direct AI spending by households is “a tiny proportion” of consumer spending today, and the transition to AI subscriptions becoming a meaningful GDP line item is a story for the coming years.

Meanwhile the St. Louis Fed has been tracking AI’s direct contribution to measured growth for over a year, and estimates like Furman’s 92% for early 2025 show how lopsided the economy became when non-tech sectors stalled. ING’s contribution is to show that even under conservative assumptions — narrow categories, imports subtracted, prices deflated — the AI buildout remains the single largest engine of US growth. Whether that engine converts into the productivity gains that justify it is the question the next four quarters will have to answer.