From AI productivity to lower debt?

Sometimes, you come across a brief note that basically says all there is to say about a topic. I have nothing to add besides salute the authors for doing the best job anywhere. And ironically, these kinds of crystal-clear insights come from unexpected sources. Not the big investment banks or consulting firms, but some think tank or research house ostensibly focused on something completely different. Enter Germany’s development bank KfW and its analysis of the impact of AI productivity gains on US debt.

For years, people have been arguing about the potential productivity gains from AI. There are a range of estimates, from the rather pessimistic numbers of Daron Acemoglu to the boosterism of Goldman Sachs.

Forecast productivity gains from AI

Personally, I am more in the camp of Daron Acemoglu and the CBO and expect productivity gains in the order of 0.1% per year. But besides this trend effect, there is also the timing question. As Stephan Bales, the author of the KfW study shows, the transformative technologies of the past (notably the steam engine, electricity, and the computer) had years if not decades of hype which coincided with zero or very low productivity growth. Only after a while, the ‘killer application’ of a new technology could be found and productivity growth took off.

AI might be similar to that in the sense that we are living through the hype with little to show for it, but in five, ten, or twenty years, someone will come up with the breakthrough that boosts productivity gains from AI by orders of magnitude.

Transformative technologies and their impact on economic growth

But let’s assume that AI will increase productivity immediately and for a long time. Some articles have already been written about the benefit that might bestow to government finances. The mechanism is simply. If the economy experiences a productivity boom, it boosts GDP growth. Higher GDP growth leads to higher tax revenues for the government and lower spending on social welfare. Together, these two forces reduce the budget deficit, which means that the debt-to-GDP-ratio grows slower or, in the best case, may even start to drop.

In the case of the US, this may look like this. If AI boosts productivity by 0.2% per year, the projected primary deficit of the US (deficit before interest expense) in 2056 may shrink from 2.8% to 1.5%. If AI boosts productivity by 0.4% per year, it might even eliminate the primary deficit altogether in the next 30 years.

Impact of AI productivity on US primary deficit

But if the primary deficit shrinks, the US accumulates less debt and – importantly – spends less money on interest (it will likely experience lower bond yields to add to the lower interest expense). The result would be a dramatic reduction in debt-to-GDP. Currently, US debt-to-GDP is on track to reach 185% in 2056. With an 0.2% boost to productivity growth from AI, it would reduce the debt burden to 143% and with a major AI boost of 0.4% it is projected to be as low as 113% in thirty years.

As you can see, a lot rides on the impact of AI on the US economy and I would go as far as to say that if you are buying 30-year US Treasuries right now, you are implicitly making a large bet that AI will significantly increase productivity and GDP growth. Because if it doesn’t Treasury yields may drift much higher to reflect the ever-increasing debt burden.

Impact of AI productivity on US debt-to-GDP