Lots of folks have recently claimed that the boom in capital spending enjoyed by the United States is mainly due …Continue reading →
Lots of folks have recently claimed that the boom in capital spending enjoyed by the United States is mainly due to the massive buildout of artificial intelligence (AI) infrastructure. (See, e.g., here and here.) That’s pretty important, since this surge of “capex” (capital expenditures) has been generating more than its historical share of overall U.S. economic growth (at least compared with the last few decades), and since some have even argued that such AI-led capex is masking major weaknesses elsewhere in the economy. (See, e.g., here and here.)
Without weighing in definitively on these controversies about relative contributions, no one should overlook the following: At least according to the official U.S. data, there’s been an impressive amount of American business spending under Trump 2.0 that has little or nothing to do with data center building.
These statistics, which come from the Census Bureau’s monthly reports on U.S.-based manufacturers’ orders for Made in America manufactured goods, certainly make clear that this kind of investment overall has been growing ever faster.
For example, “core” capital spending – which strips out such investments in defense-related products and aircraft (which supposedly is too volatile to reveal much about the fundamentals) – has expanded by 11.10 percent during the first 15 data months of the second Trump administration (from the February, 2025 – the first full month of his current term – through this past May). During the last 15 months of the Biden administration, it grew by just 2.44 percent.
But “core capex” includes AI-related spending, and the Census releases report that the latter has been up even faster. In a broad category called “information technology industries,” it’s risen by 12.51 percent.
At the same time, that performance actually isn’t much better than that of overall core capex. Much the same is true for “computer and electronics products” (up 14.11 percent); and “turbines, generators, and other power transmission equipment” (up 12.20 percent).
Oddly, moreover, spending in two other seemingly AI-related sectors –”miscellaneous electronic components” (11.02 percent) and “electromedical, measuring & control instruments” 7.97 percent, spending has actually increased more slowly than overall core capex.
Here it should be noted we run into a problem with such measurements. Although measuring and control instruments are surely strongly AI-related, electromedical gear surely isn’t. Unfortunately, the Census data are no more granular.
Another type of measurement problem: New orders for lots of goods with significant AI uses also have many other uses. For instance, lots of HVAC spending is related to AI (in particular because data centers need lots of cooling), and it’s been up a strong 19.86 percent during Trump 2.0. But heating, ventilation, and air conditioning systems are used in structures throughout the entire economy.
Ditto for “industrial machinery” (up 58.66 percent since February, 2025). And in “construction equipment” (28.36 percent, although the nation’s housing sector is in the doldrums): and other infotech categories. Of special interest – semiconductors. For some reason, Census doesn’t track new orders for computer chips, but as everyone must know, they’re used in practically every product in the U.S. economy with any electronics content.
Nonetheless, in several manufacturing groupings closely (though again, not exclusively) linked with AI, new orders have outstripped overall core capex. These include “computers and related products” (21.82 percent) and “nondefense communications equipment” (34.60 percent).
At the same time, check out the new orders during the second Trump administration for industries that seem almost completely unrelated to AI. There’s “mining, oil, and gas field machinery” (up 58.05 percent) and “consumer goods” (up 9.06 percent).
New orders, moreover, have been especially strong in transportation. That doesn’t seem to have much to do with building data centers, either. In the overall “transportation equipment” grouping, new orders have risen by 17.60 percent during Trump 2.0. And here are key results within this category:
>”motor vehicles and parts”: +11.60 percent
>”motor vehicles, bodies, trailers and parts: +9.46 percent
>”non-defense aircraft and parts”: +23.41 percent
>”ships and boats”: +95.72 percent
It’s also entirely possible that if Census tracked them, we’d find strong new orders figures for other manufacturing industries with little or no relevance to AI. These include “farm machinery and equipment,” “petroleum and coal products,” “pesticide, fertilizer, and other agricultural chemical manufacturing,” the wide variety of other chemical products not used in data centers, along with primary metals used elsewhere (like transportation equipment).
So no reasonable person can doubt that the artificial intelligence buildout has significantly affect America’s economic performance. But the idea that it’s become some kind of crutch singlehandedly holding up an otherwise doddering system? That should be open to real doubt.
