人工智能投资：高估其力——新希望已现，但为时尚早

Overestimating the force: A newhope, too soon Our calculations using BEA estimates suggest that AI-sensitive categories of BFI contributed about 1.0pp to GDPgrowth in Q1 and Q2 this year. Despite the shock and aweabout the immense scale of hyperscaler capex, industryprojections imply that this year's growth impetus may well bethe peak. Jonathan Millar+1 212 526 4876jonathan.millar@barclays.comBCI, US Marc Giannoni+1 212 526 9373marc.giannoni@barclays.comBCI, US Mark Cus Babic+1 212 526 1870mark.cusbabic@barclays.comBCI, US •In this note, we estimate theeffectof AI capital expenditures on the US GDP growth trajectoryin the medium term, and assess the prospects for supply-sideeffectsto permanently boostproductivity growth. Pooja Sriram+1 212 526 0713pooja.sriram@barclays.comBCI, US •Our estimates suggest that expenditures of AI-sensitive investment categories(computer equipment,softwareand data centers) likely boosted US GDP growth byabout 1.0pp saar during H1 2025. This estimate is about 0.2pp smaller when we use amethodology that attempts to subtract the imported content of such expenditures.Thisis large for such a small group of categories, but by no means unusual for BFI as a whole. Colin Johanson+1 212 526 8536colin.johanson@barclays.comBCI, US •In our view, popular discussions about the massive scale of hyperscaler capexconsiderably overstate the momentum this would create for aggregate GDP growth.Although the hundreds of billions being plowed into capex by individual hyperscalers isimpressive, it pales in comparison to aggregate US capex, which typically exceeds $4trn peryear — especially given lackluster outlays elsewhere. Although industry projections show thatthe largest five hyperscalers will likely boost capex by about 30% to roughly $500bn/year by2027, this represents deceleration from 2024 and 2025. If this is representative of the broadertrend,most of the impetus to aggregate demand growth from this spending is already behindus. •There is significant optimism that hyperscaler expenditures will significantlyliftstructuralproductivity growth. If so, almost all of this would have to come through the alchemy ofmultifactor productivity, which isdifficultto predict. Simple math shows that capitaldeepening, by itself, is unlikely to move the needle significantly: permanently boostingproductivity growth by 1pp would require the entire future trajectory of business fixedinvestment to beliftedby about 20%. The last time we saw anything comparable was in the1990s, when the US saw seven years of double-digit inflation-adjusted capex growth. Macroeffectof AI capex: Focus should be ongrowth,notlevels Newsflow abounds with reports about the immense scale of capital expenditures by AIhyperscalers.1For instance, outlays by the largest five hyperscalers, which were alreadyestimated to have increased by nearly 65% in 2025 to ~$375bn, are projected to increaseanother 30% by 2027, to nearly $510bn (Figure 1). On the surface, this seems quite impressive. It certainly is impressive from the perspective thatjust a handful of companies are leaving such a significant imprint on the aggregates. But to adequately assess the macro implications, some perspective is in order. During H1 2025,business fixed investment (BFI) rose 8.4% saar,afteradjusting for inflation. This stands outsomewhat relative to recent experience, but it is small in comparison to many historicalbenchmarks (Figure 2). One clear point of comparison is the dotcom boom, when BFI more thandoubled during the seven-year period from Q1 1992 to Q1 1999, rising at an average annual rateexceeding 12% saar. But there have been a number of other episodes of persistent double-digitgrowth, such as during the early 1960s and the mid-2000s. By comparison, this boom so far has been very short-lived, and smaller by proportion. But will it persist? Perhaps the point is the AI spending boom will become a persistent growth impetus. In ourview, discussions around the AI spending exaggerate the extent to which the direct growthimpetus from this spending is likely to endure. Again, we believe some perspective is needed. Although hundreds of billions of dollars in capitaloutlays sounds big, one should keep in mind that aggregate US BFI is truly immense, at about$4.2trn in Q2 2025. Because of this, it takes a lot to move the needle in terms ofgrowth.Sustaining strong growth requires not just largelevelsof expenditure, but largeincreasesinthese levels through time. For instance, the 8.4% annualized increase in inflation-adjusted BFIduring H1 2025 corresponded to a cumulative increase in expenditures during Q1 and Q2 ofabout $145bn,afteradjusting for inflation, which is about $160bn in current-dollar terms.Sustaining the same pace during H2 2025 would require afurther increaseof about $160bn,placing the cumulative increase from Q4 2024 onward at about $320bn. Sustaining the samepace during 2026 would require nominal BFI to rise byanother$320bn, which amounts to a