There’s an AI Shock Story Lurking Under the Hood of the April Producer Price Report
Producer prices came in far hotter than expected in April, with the headline Producer Price Index rising 1.4 percent for the month—nearly triple Wall Street’s forecast—and six percent from a year earlier. Services prices rose 1.2 percent, and goods prices climbed two percent.
Energy prices have understandably captured most of the attention. Final demand energy prices jumped 7.8 percent in April. Gasoline alone surged 15.6 percent. Diesel fuel, jet fuel, and residual fuels also rose. Not surprisingly, energy-adjacent services—especially trucking services—also rose sharply. This was, first and foremost, an energy-driven supply shock.
But that is not the whole story. Hidden beneath the soaring energy prices, the April PPI revealed that we’re seeing a totally different kind of inflation in another part of the economy. It is demand-driven rather than a supply shock. And it is hitting businesses rather than consumers. We’re talking, of course, about the artificial intelligence investment boom.
To get the obvious concern out of the way, this isn’t something that’s going to drag down consumer sentiment. It won’t get much attention from the Fed because our monetary policymakers define price stability in terms of consumer spending. And prices for data processing and internet services actually fell 0.4 percent in April and were down 0.4 percent from a year earlier. Wireless telephone prices are down almost 3.7 percent for the year and were flat in April. In other words, the PPI is not saying that “the cloud” is getting more expensive as a service.
It is saying that the physical world underneath the cloud is getting expensive.
Prices for Physical AI Stuff Are Skyrocketing
AI is often discussed as if it were weightless—software, algorithms, models, and digital intelligence floating somewhere above the economy. But the AI boom is also a capital-goods boom. It requires chips, circuit assemblies, storage devices, servers, networking equipment, switchgear, electrical systems, cooling capacity, power infrastructure, warehouses, trucks, and construction crews. AI might live in the digital world, but it is built from the world of atoms.
It’s easy to miss. The broad private capital equipment index does not look especially dramatic. It rose just 0.3 percent in April and four percent from a year ago. But that category is too broad to serve as a clean AI gauge. It includes all sorts of equipment with little or nothing to do with AI: ordinary machinery, industrial tools, medical equipment, vehicles, and other business investment goods.
The narrower categories tell a more revealing story.
The category that includes printed circuit assemblies, loaded boards, and modules posted one of the most dramatic moves in the entire report: up 26.5 percent in April and up an eye-popping 156.5 percent year-over-year. These are core components for accelerator boards, servers, networking hardware, and data-center equipment.
Prices for electronic components and accessories rose 8.1 percent in April and 27.6 percent from a year earlier. Computer storage devices rose two percent for the month and more than 20 percent over the year. These are exactly the kinds of components one would expect to come under pressure during a historic buildout of data centers and high-performance computing capacity.
The electrical side may be even more important. Electrical machinery and equipment rose 2.8 percent in April and 13.3 percent from a year earlier. Switchgear, switchboard, and industrial controls equipment rose 3.8 percent for the month and 12.1 percent over the year (with several switchgear sub-categories up 3.9–4.3 percent month-over-month and 14–17.5 percent year-over-year). These are the unglamorous guts of the AI economy: the systems that help move, manage, and control the enormous flows of electricity required by data centers.
The same pattern appears in infrastructure. Inputs to power and communication structures rose 2.1 percent in April and 7.4 percent from a year earlier. Within that category, goods inputs were up 2.9 percent for the month and 9.8 percent over the year. Transportation and warehousing services for these structures jumped 8.2 percent in April and 15.3 percent year-over-year.
It’s a very good time to be in the business of selling products into the AI investment boom. Margins for businesses selling capital equipment expanded 3.8 percent in April and 12.3 percent for the year. Machinery and equipment and parts and supplies wholesaling margins expanded 3.5 percent for the month and 15.1 percent over the year. Machinery and vehicle wholesaling margins expanded 5.4 percent in April and 18.4 percent year-over-year. Professional and commercial equipment wholesaling expanded 3.6 percent for the month and 10.9 percent over the year. Those aren’t pure AI categories, of course, but AI is likely driving the expanded margins.
Not Quite Hidden in Plain Sight
This also explains why the signal is easy to miss. AI is not a single PPI category. You cannot cleanly pick it out of the report. It is spread across electronic components, circuit assemblies, storage devices, communications equipment, electrical machinery, switchgear, construction inputs, equipment wholesaling, and transportation. To really uncover the influence of AI, you have to go beyond the standard 34-page PPI report and look at the “detailed report” that ran to 329 pages in April.
The Bureau of Labor Statistics (BLS) may also be understating the scale of the AI buildout because of how it assigns the relative weights of the various components used to construct the PPI lag the economy. BLS weight structures are updated only periodically and still largely reflect 2017 economic patterns. That means today’s artificial-intelligence investment boom—data centers, power systems, switchgear, storage, networking equipment, and specialized electronics—may be showing up in price changes but be too small to move the index as much as they probably should.
In the long run, AI will likely have deflationary effects through productivity improvements. But before AI can make anything cheaper, someone has to build the machinery of AI. And that build-out is increasingly expensive.