The general principle is that investment is equivalent to spending by businesses on assets, whether physical or intangible, that \u201care used repeatedly or continuously in the production process\u2014that is, in the production of other goods (including other fixed assets) or of services\u2014for more than 1 year\u201d. The BEA measure of investment includes spending on machinery, software, and buildings, as well as spending on research and development that companies normally count as operating expenses on their income statements. It does not include spending on intermediate inputs that are used once and then disappear, such as electricity, advertising services, or materials and components for manufacturing.<\/p>\n
AI-related investment now involves everything from laying more fiber optic cables to building new power plants, but for the sake of simplicity I am focusing on a few components within the NIPAs that seem to map most cleanly to the capital expenditures reported in companies\u2019 cash flow statements: data centers (part of structures), computers and peripheral equipment, and communications equipment. A broader measure could also include \u201cspecial industry machinery not elsewhere classified\u201d (includes chipmaking equipment) and \u201celectrical transmission, distribution, and industrial apparatus\u201d because those also seem to have responded to the surge in AI-related spending, but the macro results do not look much different. Research and development spending is also being affected by the boom in AI but not nearly as much.<\/p>\n
Most of the BEA\u2019s estimates of nonresidential structures investment ultimately come from the Census Bureau. The Census has monthly estimates of construction spending on data centers<\/a> going back to 2014, although the numbers are only available in the depths of downloadable spreadsheets. Data center construction is bundled inside the broader category of \u201coffice\u201d, presumably because office buildings and data centers are both climate-controlled buildings with lots of computers inside. The growth in this category has been extraordinary over the past few years, but it started from a very low base. The numbers reported by the Census make clear that it has only been sufficient to offset the slowdown in warehouse construction, which had surged during the pandemic-induced e-commerce boom only to normalize in the years since. Spending on new chip fabs dwarfs both.<\/p>\n
![\"\"](\"https:\/\/www.newsbeep.com\/uk\/wp-content\/uploads\/2025\/08\/https:\/\/substack-post-media.s3.amazonaws.com\/public\/images\/ce215354-9dcb-4068-8015-76c5dd1169a1_3200.jpeg\")
<\/a><\/p>\n