经济周刊：产能增长——这次不一样

William Blair The scale of investment spending today is astounding. Justfour companies—Alphabet, Amazon, Meta, and Micro-soft—spent a combined $376 billion on capex in 2025,equal to 28% of total S&P 500 capex. Comparisons with thelate-1990s tech boom are inevitable, particularly in regard an attempt to capture the concept ofsustainablemaximum output,the greatest level of outputa plant can maintain within the framework of arealistic work schedule after factoring in normal to how much productive capacity is being built.In thisEco- nomics Weekly, we look at aggregate investment spend- ing and the expansion in industrial capacity; we alsocompare the current cycle with the 1990s and discusswhat those differences might mean for the Fed—and As shown in exhibit 1, there was a massive increase inindustrial capacity during the late 1990s—the largest onrecord and against which today’s buildout looks extreme- Digging into the industries behind this change, exhibit 2shows that the increase in maximum sustainable capac-ity during the late-1990s boom was concentrated almostentirely in computers and electronic products. At its peak(1996–1998), capacity in that category rose by nearly In exhibit 1, we plotted the Fed’s measure of industrialcapacity data from 1968 to 2026. Given that we are sup-posed to be in the midst of the biggest capital investmentboom at least since the late 1990s, the result is somewhat The Fed’s G.17 Industrial Production and CapacityUtilization release provides data on industrial capacity,compiled from a total of 89 detailed industries—71 inmanufacturing, 16 in mining, and 2 in utilities. The Fedalso uses data from the Census Bureau’s Quarterly Surveyof Plant Capacity Utilization (QPC), which itself is a mixof qualitative assessments of sustainable operations. The As shown in exhibit 3, capacity growth in the current cy-cle is again being driven by computers. In March, capacityin that category was increasing at its fastest pace of thiscycle (7.2%); however, that remains far below the sus-tained 30%-40% multiyear growth rates from the prior William Blair This data suggest that in 2022 and 2023, much of thegrowth in structures was related to the combination of acatch-up from COVID weakness and an expansion in theenergy sector in response to higher prices (an area wherewe are likely to see further expansion in the coming At face value, the data suggest that today’s boom is mod-est compared with the one in the late 1990s, but is morewidely dispersed across the economy. Unfortunately, wedo not think it is quite that simple. Data definitions andthe nature of the current expansion mean that much of Back in the late 1990s, much of the expansion in capacitywas focused on fiber optics, servers, and factories. Theassumption was that internet demand would continue toexpand rapidly and that miles of unlit fiber optic cables Over the last year growth in structures has begun to cool,as many of those shells have been built. The next phase isoutfitting these structures with equipment. As shown inexhibits 5-8, this spending is taking place from a num-ber of sectors, though predominantly in computers and A large share of this expansion is occurring in less tangibleareas that are not well captured in the industrial sectorsthat the Fed uses to measure production and capacity, Importantly, the Fed’s concept of capacity is tied to whatit views as being sustainable and how much capacity acompany reliably has. What is considered sustainable andreliable is not always clear, and the estimate is influencedby factors such as the availability of labor, energy-costvolatility, and regulatory limits. In the late 1990s, forexample, there was a deeper pool of available labor, fewerregulatory constraints, and greater energy availability, To look at business investment spending, the national ac-counts data break down actual dollars spent on areas suchas structures, information processing equipment, indus- William Blair The less tangible nature of the current expansion couldhave important implications for the Fed and monetary If the capacity being installed is shorter-lived and moreexpensive to purchase, then depreciation rates will bemuch faster and that capital stock could be more difficultto replace. Similarly, if today’s capex becomes obsoletemuch sooner and the investment cycle is much shorter,then the productivity benefits could also turn out to be For the Fed, this shift in structural capex spending meansthat its capacity utilization data are a much less reliablegauge of both economic slack and inflation pressure thanwas the case in the past. High utilization readings could From an inflation perspective, the implications are alsoless straightforward. If the cycle is increasingly dominat-ed by shorter-lived capital, and firms need to refresh thatstock more frequently, this would 1) accelerate spend-ing, adding to inflationary pressures, and 2) potentiallyreduce the durability of the positive productivity gain