US turns to AI productivity as worker income share hits record low

Treasury Secretary Scott Bessent and Federal Reserve Chair Kevin Warsh are betting that AI-driven productivity will make America richer and deflationary enough to ease worries about the $40 trillion national debt.

However, the statistics backing up the optimism are far more skewed: the labour share of income in the US is already down to 52.8%, the lowest seen by the government since it began measuring in 1947.

Business profit margins, meanwhile, reached an all-time high of 14.9%. As Gregory Daco, chief economist at EY-Parthenon, puts it, most of the productivity gains driving the split occurred before the rise of AI.

"Productivity growth protects margins, not income," he wrote in a note, pointing out that second-quarter output grew 1.7% on just 0.3% more hours worked, while compensation rose 2.6%, a gain that inflation from this spring and summer's oil prices erased almost entirely.

Large, vertically integrated firms captured early gains during the 19th-century railroad boom and the 1990s dot-com era too, while smaller companies absorbed higher costs and policy uncertainty.

What made the 90s different, he says, is that productivity gains from cheaper software eventually spread through the broader economy, pulling wages up with them. AI, Daco warns, isn't guaranteed to follow that same timetable.

One of the reasons why this happens can be attributed to the way AI infrastructure capital expenditures get accounted for in the GDP. 

The net imports of the big computer servers utilised in AI data centres were recorded at an annualised rate of $450 billion, which is a significant jump from $50 billion per year from 2023, based on the Census data by economist Joseph Politano.

Since the importation of a server increases investment and reduces imports within the same calculation, the net contribution of an imported server to the GDP will effectively amount to zero, while such an item contributes to a booming industry despite being in a recession.