- by New Deal democrat
Let’s wrap up this week of limited economic data by looking at perhaps the most prevalent theme about the state of the US economy right now: that it consists of basically 2 components, (1) AI and (2) everything else. This is a round up of a bunch of graphs I have seen in the past few weeks that have all circulated around this same point.
First of all, construction spending. While unfortunately FRED doesn’t separate out data center spending, some other writers have. Here’s a good graph from Steve Rattner showing how, when we take out AI data center construction spending, the entire remainder of construction spending has been declining since the middle of 2024:
Here’s a fuller breakdown from Prof. Jason Furman:
And Prof. Furman also points out that total CapEx investment is more concentrated now than at any point in the last 50 years:
This is showing up in stock market valuations:
The valuations of AI related companies are Booming. Everything else is moribund.
This is apparent in the advance-decline line. If share prices in more companies in, e.g., the S&P 500, increase on any given day vs. decline, the A-D line is positive. If the declining number of stocks outnumber advancing ones, the A-D line is negative:
Here is a longer term historical view of a similar, basically self-explanatory, metric:
The last time the number of outperforming stocks was this narrow was during the dotcom bubble at the end of the 1990’s.
Meanwhile borrowing costs are rising for lower rated corporate debt:
Here is a similar FRED graph that, in addition to Treasurys (purple), includes high rated AAA corporate debt (blue), lower rated but still investment grade BBB corporate debt (orange), vs. risky high yield CCC or lower corporate debt (red), all normed to “0” as of the lows in the bond market last October:
In general, the lower rated the debt, the more it blows out to the upside in the yield investors demand to buy that debt. This includes before and during recessions, but also during and after other scares, such as the Silicon Valley Bank collapse in March 2023.
Basically the US economy has put all its eggs in one basket, called AI. Earlier this week Paul Krugman wrote on his substack that "Massive spending on data centers is crowding out investment in everything else in the economy,” adding that "this isn't a hypothetical risk." And as I have written in the past several months, while the *software* aspect of AI may not be a bubble, i.e., the end result of AI may be a boon to human productivity, the *hardware* aspect, i.e., 10 or 20 different companies all making massive construction investments in plant and equipment, where only 1 or 2 companies are likely to survive as winners (much as in the 1990s all of the internet search companies withered except for the big winner, Google), almost certainly *is* a bubble.