There is no end to the articles suggesting that automation is about to radically change both the way the service sector works and the types of space it demands.
U.S. manufacturing, however, has been experiencing automation’s changes for more than two decades—since the advent of robotics. We have just completed a study of how industrial automation and robotics affected local industrial space markets over 1990-2007.
As firms transition to robotic manufacturing, local industrial markets are affected in many ways. Most obviously, the amount of factory floor space needed per unit of production can change. Meanwhile, market-wide demand for industrial space will react to changes in the supply chain, inventories, and facilities needed for the repair, installation or modification of the robots—which is why we chose to study how whole industrial markets (warehouses as well as factory space) have reacted to differences in local robotic growth.
Using International Federation of Robotics (IFR) data, we estimated annual figures for industrial robot installations across the 45 largest U.S. metro areas. From this, we determined local growth in robots over the period and then divided by total local employment, obtaining a measure of growth in robot density per worker for each MSA.
Using data from an MIT colleague, we assessed industrial space market performance by MSA, comparing markets whose industries automated over the period to those whose industries didn’t—or did much less. CBRE data let us track the growth in industrial occupied space and in industrial space rents (adjusted for inflation) over the period.
Our initial statistical results were pronounced. Between 1990 and 2007, robotic density increased by 2 robots per worker on average; by market, this varied from virtually zero to 8.4. Within this distribution, an increase of 3 robots per worker represents 1 standard deviation. An MSA with growth of 3 additional robots per worker would see 25% less growth in industrial space demand. To put this into context, average occupied space growth over the period was 34%, so a market with an increase of 3 more robots/worker would have seen occupied space growth of only 9%. The same 3 additional robots per worker reduced industrial rent growth by 20% in our results. From average rent growth of -4%, such a market would see its (constant dollar) rent growth drop to -24%. These results were very statistically significant, though the analysis so far leaves two questions unanswered.
The first is an argument that the correlation between industrial space demand and automation might be spurious. Areas hard hit by import growth, for example, might see lower employment and occupied space, and an infusion of robots as domestic manufacturers try to remain competitive. Under this scenario, automation might not be causing industrial space demand to shrink; rather, adverse shocks in the local economy are jointly causing layoffs in space usage and employment and these are generating the substitution of robots. In our research, we ran additional estimations that controlled for a range of adverse economic shocks (imports, NAFTA, the general shift to services from goods production), but none of these diminished the statistical importance of increases in robots/worker.
The second question concerns the type of space that robot adoption demands. Empirically, most of the local growth in occupied industrial space comes from the construction of new factories and warehouses—and not from changes in vacancy rates, or from the closing and abandoning of existing buildings. If automation required new factories, we should observe a positive correlation between its growth and that of occupied space. The negative correlation we see in the data suggests that automation is perfectly able to fit into existing structures—and at lower space utilization rates than prior industries required.
Industrial automation has enabled many American industries to remain competitive, preventing further declines in U.S. industrial production and increasing output in many cases. That said, automation has not increased industrial employment (rather, the opposite) and we now have some evidence that it may reduce net demand for industrial space.
 McKinsey (2016), Frey and Osborne (2013).
 We excluded the years since 2007 due to the strong influence of manufacturing cyclicality from the Financial Crisis.
 Industrial robots are manufactured in relatively few countries but are installed worldwide, and their installation is time-consuming and expensive. Each robot made since 1990 has a VIN number; the IFR records these along with purchase and installation information.