Goldman Sachs argues that market narratives around AI have been overstated in both directions: the boost to growth looks smaller than the headline spending figures suggest, while fears that AI is draining large amounts of activity from elsewhere in the economy also appear exaggerated.
In a US economic research report dated Aug. 10, Goldman said US AI investment is expected to reach $600 billion in 2026. That would be close to 2% of GDP and more than 10% of business fixed investment. The report focused on a question the bank said is important but rarely asked in the market: what exactly is AI investment crowding out?
Goldman’s answer was that the effect is limited. It identified three channels of crowding out — replacement of other technology spending, pressure on other construction activity, and higher interest rates — with a combined impact of about $50 billion.
Goldman breaks the crowding-out effect into three channels
The first channel is substitution away from other tech investment. Goldman said hyperscalers had funded AI investment in recent years by reducing buybacks. Now that capital expenditure has moved above cash flow, they have started borrowing, and higher rates have not stopped them. Corporate IT spending surveys show that, for most companies, AI costs still account for only 1% to 5% of IT budgets. About one-third of that comes from incremental spending, while the other two-thirds comes from cuts to other spending categories. Goldman estimated this part of the crowding-out effect at about $30 billion.
The second channel is pressure on other construction. Data center construction has risen to 9% of private nonresidential construction spending, according to the report. Goldman said that increase has been offset by a decline in manufacturing facility construction tied to subsidies under the Inflation Reduction Act and the CHIPS Act. Its estimate is that data center building has displaced about $10 billion of other construction activity. The bank said crowding out is more visible in some local markets, with data center construction accounting for more than half of nonresidential building spending in some states. At the national level, though, the data does not show a significant effect. States with heavier data center construction also have not seen construction wage growth that is meaningfully above the national average.
The third channel is higher interest rates. Goldman said AI-related bond issuance now accounts for nearly one-quarter of total investment-grade bond issuance. It estimated that this supply has raised borrowing costs for non-AI companies by about 5 basis points, translating into roughly $10 billion less investment.
Direct GDP support is much smaller than the spending headline
Goldman said the 2% of GDP figure refers to gross spending, not GDP contribution. A large share of the $600 billion goes to imported equipment, and some AI investment is recorded in national accounts as intermediate input rather than final spending.
On Goldman’s estimates, AI’s direct contribution under official statistical treatment is about 0.5% of GDP, well below the 2% gross-spending figure. Converted into annual growth terms, AI investment would add about 0.1 percentage point to US GDP growth in 2026.
If statistical treatment were adjusted — by reclassifying semiconductor investment used to train AI models as final spending and including chip design service exports — the true GDP boost would be around 0.3 percentage point. After accounting for wealth effects from rising AI company share prices, the drag from higher electricity prices on real income, and the roughly $50 billion crowding-out effect, the net impact would be about 0.2 percentage point.
Goldman says both popular market narratives miss the mark
The report’s main contribution, in Goldman’s view, is to offer a framework for calibration. One market view holds that AI is driving a major acceleration in growth. Another says AI is significantly displacing activity elsewhere. Goldman’s data suggests both claims go too far.
The bank said $600 billion is a large number, but much of it is spent on imported equipment. Data center construction has created building bottlenecks in some places, yet those pressures are offset at the national level by weaker manufacturing facility construction. AI bond issuance has lifted rates, but only by 5 basis points.
Goldman’s conclusion is that AI investment is neither a cure-all nor an economic drain. Its real boost to the economy is smaller than many broad market narratives imply, and its crowding-out effect is milder than some critics fear.
For investors, the report suggests AI trading may need to shift away from broad macro storytelling and toward company-level verification, with more attention on capital returns rather than blanket assumptions that AI is changing everything.
The original article said the piece was compiled and interpreted by Chaoxiang Research based on a third-party brokerage research report from Goldman Sachs dated Aug. 10, 2026, together with public market information. It also said that quoted ratings, target prices, earnings forecasts, and related judgments reflect the views of the brokerage analysts and their institution, not Chaoxiang Research, and do not constitute investment advice.
The article also included a risk reminder, stating that market decisions should be made independently and that the piece should not be used as a basis for buying or selling any security.