1. Goldman Sachs AI investment outlook: the growth boost is near its peak
Goldman Sachs economists expect the contribution of artificial intelligence investment to US economic growth to peak between late 2026 and early 2027, with the yearly rise in hyperscaler investment as a share of GDP topping out at about 1.5% in the fourth quarter of 2026. The bank also estimates that AI spending has lifted US GDP growth by only about 0.1 percentage points.
The Goldman Sachs AI investment view does not predict a collapse in spending. The bank expects the largest technology companies to keep pouring money into data centers, chips and related equipment. Its point is narrower and more technical: once spending stops accelerating, the extra growth it delivers to the economy starts to fade, even if the absolute amount of money involved stays very large.
That distinction matters for anyone trying to judge how much of the recent strength in the US economy can be credited to the AI buildout, and how much support that buildout can be expected to provide in the quarters ahead.
2. Hyperscalers carry the weight of the buildout
At the center of Goldman's analysis are the hyperscalers, the cloud computing giants that build and operate enormous fleets of servers and rent that computing power to businesses, developers and AI labs. The bank singles out Amazon, Microsoft and Alphabet, the parent company of Google.
Goldman projects that the combined capital expenditure of those three companies could reach between $725 billion and $800 billion by 2026. Capital expenditure, or capex, refers to money spent on long-lived physical assets rather than day-to-day operating costs. For the hyperscalers, that means land, buildings, power infrastructure, cooling systems, networking gear and, above all, the specialized processors that train and run AI models.
Spending on that scale ripples through the economy. Construction firms build the facilities, utilities expand capacity to power them, and equipment makers ship servers and networking hardware. In the national accounts, all of this registers as business investment, one of the main components of gross domestic product.
Yet the size of the outlay is not what concerns Goldman. The bank's argument rests on how quickly that outlay is growing, and on where the money ultimately ends up.
3. Why does AI's boost to US GDP growth fade?
AI's boost to growth fades because GDP growth measures change, not levels. A company that spends the same enormous sum every year keeps adding to the size of the economy, but it adds little or nothing to the rate at which the economy is expanding. Only rising spending pushes the growth rate higher.
An everyday comparison helps. A household that moves from spending $50,000 a year to $60,000 a year has increased its consumption by 20%. If it then spends $60,000 again the following year, its spending is still high, but the increase is zero. The same arithmetic applies to the hyperscalers. Their investment can remain at record levels while its contribution to growth shrinks toward zero.
Goldman captures this by tracking the year-over-year change in hyperscaler investment as a share of US GDP. The bank projects that measure will peak at roughly 1.5% in the fourth quarter of 2026 and then decline. It places the peak in AI's overall growth impact in a window running from late 2026 into early 2027.
After that point, investment may remain elevated, but its incremental contribution to GDP growth is expected to diminish. In other words, the AI buildout could continue at full speed in dollar terms while doing less and less to lift the headline growth figure each quarter.
4. Jan Hatzius and the import problem
Goldman Sachs chief economist Jan Hatzius has been skeptical about how much AI capital spending actually adds to US output. His reasoning centers on the destination of the money rather than the amount.
A large share of AI investment goes toward chips and hardware that are manufactured abroad and imported into the United States. In the way GDP is calculated, investment adds to the total, but imports are subtracted from it, because GDP is designed to measure what is produced domestically. When a US company buys a server built overseas, the purchase shows up as investment, and the matching import largely cancels it out.
The result is that the net contribution of AI spending to US GDP is considerably smaller than the headline investment figures suggest. The domestic portion, such as construction labor, local electrical work and US-made components, still counts, but a significant portion of each dollar effectively flows to foreign suppliers.
This is why Goldman's estimate of AI's effect is modest. The bank calculates that AI spending raised US GDP growth by only about 0.1 percentage points, and that in some periods the effect was close to nothing at all.
5. How GDP accounting treats AI spending
To see why imports carry so much weight in Goldman's reasoning, it helps to recall how GDP is assembled. Economists add together four broad categories of spending: household consumption, business investment, government purchases, and net exports, which is the value of exports minus the value of imports.
Business investment is where the AI buildout appears. Every new data center, every rack of servers and every upgrade to a power substation lands in this bucket. On its own, that would suggest a sizable lift to output whenever the hyperscalers increase their budgets.
The net exports line works in the opposite direction. Because GDP aims to measure what the domestic economy produces, any goods bought from overseas are subtracted so that foreign production is not counted as American output. When much of the equipment in a data center is built abroad, a large part of the investment gain is offset by a rise in imports.
The arithmetic leaves behind only the domestic value added: the work of US builders, engineers and electricians, the American-made components, and the services provided by local firms. That residual is real, but on Goldman's numbers it adds up to far less than the headline capex totals imply.
6. A sharp contrast with bolder estimates
Goldman's figure stands well apart from some earlier assessments that credited AI investment with a much larger share of recent US economic growth. Those higher estimates helped popularize the idea that the data center boom was single-handedly keeping the American economy afloat.
The gap between the two views comes largely down to the treatment of imports. Analyses that focus on gross investment flows naturally produce bigger numbers, because they count the full value of the hardware purchased. Goldman's approach nets out the share of that spending that leaves the country, which shrinks the measured effect substantially.
Neither approach denies that AI spending is enormous. The disagreement is about how much of it translates into domestic production and income, which is what GDP is meant to capture. For policymakers and investors who track growth data closely, the methodology can change the conclusion about how dependent the economy has become on the AI trade.
7. A long-term shift rather than a short burst
Goldman does not frame its findings as a verdict against AI. The bank describes the buildout as a long-term structural shift in the economy rather than a temporary spike in activity.
In Goldman's view, the larger economic payoff from AI is still ahead. It expects major productivity gains to emerge as businesses gradually adopt the technology and fold it into their operations over the coming years. Productivity, meaning output per hour worked, is the main driver of sustained increases in living standards, and it tends to respond to new technologies with a lag.
That pattern has precedent. Past general-purpose technologies, from electrification to personal computers, often took years to show up clearly in productivity statistics, because companies needed time to reorganize workflows, retrain staff and build complementary systems around them.
Seen this way, the fading of the investment boost and the arrival of productivity gains are two different phases of the same story. The first phase is about building the infrastructure. The second, if Goldman's expectations hold, is about putting that infrastructure to work.
8. What it means for the growth picture
If AI's contribution to growth peaks in the fourth quarter of 2026, as Goldman projects, the US economy will need other sources of momentum to maintain its pace. That raises a central question for economists: what replaces AI investment as a driver of expansion once its incremental effect begins to shrink?
The answer matters beyond the technology sector. Business investment has been one of the more closely watched supports for the economy, and a slowdown in its growth rate, even without an outright decline in spending, would shift attention to consumer demand, housing, government outlays and trade.
Goldman's analysis also adds nuance to a widely shared narrative. The scale of hyperscaler spending is real and historically large, but its effect on measured GDP growth appears to be smaller and more temporary than the raw figures imply.
9. What to watch next
Two sets of data will show whether Goldman's projections are on track. The first is capital expenditure guidance from Amazon, Microsoft and Alphabet. Their spending plans will reveal whether combined outlays are heading toward the $725 billion to $800 billion range the bank projects for 2026.
The second is the official GDP data itself. As quarterly figures arrive, economists will be looking at whether AI's share of growth begins to shrink as the projected fourth-quarter 2026 peak approaches and passes.
If both trends unfold as Goldman expects, the conversation is likely to move from how much the AI buildout is lifting growth today to how and when the technology begins to raise productivity across the wider economy.
