How much is AI contributing to US economic growth?

The arrival of ChatGPT and subsequent release of other generative artificial intelligence tools has prompted a technology investment boom. US technology firms are vying for the lead, battling foreign competitors, primarily from China. Meanwhile, the US government sees this as a critical economic and national security issue for coming decades and has been willing to incentivise and support initiatives.

The result has been a substantial increase in spending on construction of data centres and electric power generation. In fact, these are the only sectors where we are currently seeing construction spending growth. Residential construction is being hampered by elevated mortgage rates and a general lack of affordability following the post-pandemic surge in residential property prices. Technology, excluding data centres, has fallen back following the conclusion of President Biden’s 2022 CHIPS and Science Act that provided $52.7bn to boost domestic semiconductor manufacturing and research. Meanwhile, all other non-residential construction is flat lining.

The surge in activity is even more apparent when we look at the GDP measure of non-residential business capital expenditure. If we group computing and peripheral, together with software investment, we see a cumulative 62% increase in real spending since the first quarter of 2022. The frenzied spending on tech appears to have effectively cannibalised other non-residential business investment, which fell in year-on-year terms for six consecutive quarters between 4Q 2024 and 1Q 2026.

This has led to plenty of debate over calculations of how much tech investment tied to AI is contributing to US growth rates. Jason Furman at Harvard University suggested that investment in technology and software contributed 92% of US GDP growth in the first half of 2025, at a time when the non-tech sector was struggling. Yet others have suggested it is far less, pointing to the scale of imported technology that needs to be subtracted from the total.

The starting point is looking at investment and its composition. Some analysts have used what we feel is too broad a characterisation of 'technology' investment spending. Information Processing is the headline category that is sometimes cited since it includes computers & peripheral equipment, but it also includes lots of other categories that are not tied to specifically AI-related areas, such as medical instruments and photocopiers and printers. We favour using purely the computer and peripheral equipment components while also including software investment.

However, even computer & peripheral equipment is too broad a category. While it does include the servers and computer equipment that are used in data centres, it also includes regular computer equipment used in offices. Unfortunately, we cannot break this down, but rationalise that the overwhelming proportion of higher spending in recent quarters is tied to AI. Similarly, we are cognisant that not all software spend is AI related, yet assume that the bulk of the increases in spending are the result of it.

We are on more solid ground with data centre construction spending. However, given it is the computer equipment inside that is the major cost, rather than the physical structure, the contribution from this component will be relatively small.

Not all tech investment spending is on items that are made in the United States. Imports subtract from GDP, and we should take account of that when calculating the 'true' contribution of tech/AI investments. Imports of technology-related items have tripled in value over the past two years to $60bn per month. US exports have also increased, but on a far more modest scale, from $9bn per month to $17bn per month. The result is that the trade deficit in computers, peripherals and semiconductors has risen from $20bn per month to more than $40bn per month.

One mitigation is that prices of chips and semiconductors have surged in recent quarters due to strong demand and limited supply, so we need to deflate the nominal dollar values in order to calculate the 'real' growth in imports. We chose to use the PPI measure of 'electronic components and accessories'.

Regarding international trade in services tied to AI and technology, the applicable category is 'computer services', which captures foreign AI subscriptions and token purchases, but it will also include the likes of computer system design and other priced digital data services. In nominal dollar terms, exports are around $22bn annualised with imports at $19bn with volume growth up 15% YoY for exports and 7.5% YoY for imports in second quarter 2026. For this inclusion, we are again taking probably too aggressive a view that AI-related activity is responsible for all the growth. That said, it is only a tiny proportion of overall economic activity.

We’ve put together a series of charts showing different ways of calculating the contribution to YoY% real GDP growth. The first shows the broadest measure of 'tech' investment – total information processing, data centre and software investment – without subtracting imports. This categorisation has headline investment accounting for 50.2% of YoY GDP growth in second quarter 2026, with 46.6% being the average contribution to YoY GDP growth over the past four quarters.

If we were to take a stricter definition and include only computing, peripheral and software investment, then it follows a very similar trend to the chart below, but the result is a 44% of GDP growth contribution for second quarter 2026 with a 43.3% average over the past four quarters.

Then if we take that 'core' measure and subtract net imports of computers, peripherals and semiconductors, deflated by 'PPI electronic components & accessories', then we get a more modest 36%, which is actually slightly lower than the 37% average contribution over the past four quarters. We view this as the fairest measure of tech investment we can currently calculate.

Given the issue surrounding uncertainty over what is truly AI/data centre investment and what is, what we might term, legacy tech investment, we suspect there is some over-estimation in these results. However, given the acceleration in tech capex growth since the release of ChatGPT we would only revise down the contribution marginally. As such, we believe that the overall tech investment rollout accounts for around a third of the current YoY rate of US GDP growth.

AI enthusiasm is also contributing to GDP growth via consumer spending. Directly through AI token purchases and subscriptions and indirectly via positive wealth effects thanks to the surge in technology company stock prices.

Right now, AI subscriptions account for a tiny proportion of overall consumer spending. Proprietary client spending data from Bank of America and PNC Bank suggest only around 2-3% of American households are spending money on these tools, with a typical monthly spend of $20-30. That is dwarfed by internet, TV and cell phone subscriptions right now, but over coming years that is likely to change, and it will start to make a meaningful contribution.

Instead, it is the wealth effect that is having the largest impact on consumer spending today. Chat GPT was released on 30 November 2022 and, since then, the NASDAQ stock index has risen 130% while the S&P 500 is up 90%. This has contributed to household holdings of financial assets rising from $109tn to $142tn over that three-and-a-half-year time frame. Ownership is heavily skewed towards the top 20% of households by income, who, according to Federal Reserve data, hold 72% of the wealth of America. The bottom 60% of households by income hold merely 15% of its wealth. In an environment of weak consumer confidence and flat-lining real household disposable incomes, this AI-driven wealth surge is likely to be a key factor maintaining the K-shaped consumer narrative whereby high-income households are the driving force behind consumer spending today.

Calculating a number for what extra consumer spending is due to AI wealth gains is difficult. A 2025 Federal Reserve paper suggested that because of the increased concentration of wealth towards higher income groups, the propensity to consume out of wealth has dropped markedly. For the top 20% of households by income, they estimate 0.8 cents of every dollar increase in wealth is spent. For the other 80% of households, who have seen far more modest increases in wealth, that number is closer to 7.5 cents per dollar increase in wealth. That paper estimates a weighted average of 2.6 cents in the dollar until 2020 versus 3.5% in 1990-2005. But with further concentration of wealth in the top 20% of households by income over the subsequent six years, we strongly suspect the number is below 2 cents today and assume 1.5 cents based on that shift in wealth dynamics. Therefore, $33tn of extra household equity wealth translates into roughly $500bn of cumulative extra spending since fourth quarter 2022. Simplistically, that is around an extra $36bn of consumer spending per quarter ($144bn annualised), equivalent to around 0.65pp of 2Q 2026 consumer spending and 0.44pp of GDP.