The mystery of productivity

Productivity takes into account the myriad ways that efficiency spreads through organizations and societies. Some developments are rapid, massive and obvious: like the improvements to agriculture that vastly reduced the share of people who live on farms. Others are harder to grasp and show up indirectly, befitting the remark of economist Moses Abramovitz that productivity is “a measure of our ignorance about the causes of economic growth.”

We can, of course, try. The simplest method might be through gross domestic product (GDP) growth as defined by the change in the labor force plus the change in productivity. Since the labor force can be fairly easily measured, changes in productivity can be inferred from changes in GDP.

In 2025, US economic growth exceeded expectations even as labor market growth was limited by stringent immigration enforcement. The combination of a rise in GDP without a corresponding increase in labor implies that productivity growth was high. The Trump administration expects that ongoing growth will stimulate job formation and the labor market will improve. The claim is, essentially, that this initial burst of productivity is bringing few new jobs for now, but more jobs will be created in the long run. At least, that’s the hope.

The twist

To understand how this scenario might develop, another GDP equation is helpful. In this academic version, GDP is defined as the sum of an economy’s consumption, investment, government spending and net exports: the familiar C + I + G + NX equation.  Assuming consumption remains steady, government spending moderates, and imports have been reduced by the tariffs, investment becomes the key source of growth. But the profile of “investment” today is different than in the past and in the textbooks.

Today, the largest and most conspicuous US capital expenditure (capex) is investment in AI. But the economic implications of hyperscaler spending are less clear. Traditional expenditures on machinery, equipment, factories, offices and even residences have well-understood economic multipliers. In contrast, AI data centers create few permanent jobs. Once built, their impact mainly comes from the productivity they enable.

Outlook

The best outcome? We could see the boost to US growth from investment morph over time into higher-wage jobs as AI enhances worker productivity, warranting greater compensation. This would drive consumption and, critically, improvement in the lower leg of the “K-shaped” economy as cyclical sectors like housing and autos reawaken. Along the way, growth could exceed government spending, enabling progress towards the administration’s 3% deficit target. This would truly be a virtuous cycle.

What could go wrong? For one thing, AI could eat our jobs. That is a sharply debated matter with a great many unknowns, but vast, permanent disemployment seems unlikely. Some disruption is inevitable, but we are not doomed. Just as our ancestors adapted once they left the farm, so will we in the age of AI.

For another, the arms race of spending could fail to translate to longer-term economic output. As with the internet bubble in 2000-2001, malinvestment in nonproductive endeavors could lead (at least for a time) to bad results. This uncertainty is reflected in the recent market gyrations of the technology sector.

While there will undoubtedly be winners and losers, projecting a broadly disappointing economic outcome means betting in aggregate against the largest, and arguably most successful, companies in modern history.

If all goes well, we can hope to see a virtuous cycle from investment to productivity, jobs and growth. The result would bring significant returns both to capital and also to labor. But it’s still early days. The question of whether we get the virtuous cycle or jobless growth or a flat-out bust promises to be the key macroeconomic issue of the next few years.