AI data center and electricity supply production as drivers of industrial production and capital goods spending

 

 - by New Deal democrat

There is more and more accumulating evidence that manufacturing, at least in the aggregate, is something close to Booming. That message was apparent in yesterday’s durable goods orders report for December. While the headline number (blue in the graph below) declined -1.4%, the three month average for this very volatile series made a new post-pandemic high. The much less noisy capital goods new orders number (red, right scale) increased 0.6% to a new all time high. The three month average for that metric made a new all time high as well:

These numbers have been rising for the past 18 months, and even accelerating, as the YoY% gains have also been increasing:

The same dynamic was apparent in yesterday’s industrial production release, in which headline production rose 0.7% for the month, and the manufacturing component 0.6%. The latter made a 3+ year high, and the former a post-pandemic high. But as I pointed out, the real driver was electric utility production. The below graph drives this home by norming all three to 100 as of just before the pandemic. Headline production is up 1.3% since then, but manufacturing production, despite the increases of the past 18 months, is still down -0.4%. Utility production, however, is up 13.9%!:

As I indicated then, I suspect this is almost all driven by AI data center and attendant power plant construction. Below I again show headline industrial production (blue) compared with employment in manufacturing (red), and in non-residential construction (gold), again all normed to 100 as of just before the pandemic:

Employment in manufacturing has been declining for several years, and is now -1.2% below its pre-pandemic levels. But employment in construction ex-residential housing is up 9.4%, and as of the latest report was still rising!

We know that some non-residential construction employment was increasing due to the Inflation Reduction Act, which led to a surge in construction of manufacturing plants, but that peaked in the middle of 2024 and has been declining since:

While this isn’t definitive, and I haven’t yet found a way to do a more granular analysis, it all points to the big increase in both electric utility production and non-residential construction employment being concentrated on the building of AI data centers and the gargantuan energy demands to power them. On this (I submit) rather slender reed is the current growth of the economy reliant.

Presidents’ Day week jobless claims pose a quandary

 

 - by New Deal democrat

Later this morning I’ll discuss yesterday’s positive durable goods orders release, and in that context I’ll also have more to say about the likely reason why industrial production also improved so much. Tomorrow we’ll get personal income and spending, and new home sales, both from December, as well as the first pass at Q4 GDP — all of which are one month late - even now three months after the end of the government shutdown!

In the meantime let’s take our usual weekly look at jobless claims. I do this because they are a good short leading indicator, particularly of the labor market.

Initial claims declined sharply last week, down -23,000 to 206,000, the best reading since mid-January. The four week moving average declined -1,000 to 219,000. And with the typical one week delay, continuing claims rose 17,000 to 1.869 million:

As per my usual practice recently, the above graph includes the last three years to show that there has been a pattern of unresolved seasonality whereby claims rise in the first six months of the year, and then decline in the last six months. 

Now let’s look at the YoY% changes which are more important for forecasting pursposes:

Initial claims were down -8.0%, but the four week moving average was still higher by 0.6%. Continuing claims were also higher by 0.4%.

So was the big decline this week a return to the positive comparisons YoY that we saw beginning last July? Or was it simply a function of Presidents’ Day being one week earlier this year than last year? Notably, in 2023 and 2024 there was a big spike downward in claims during the equivalent week in February, and while there was no such spike last year, there was a big upward spike the following week.

All of which means we’ll have to wait one more week to see if this week’s initial claims number was an outlier or not. Another week of lower claims YoY suggests that the trend since last July of comparatively very low claims has not abated, while a spike upward would suggest that unresolved post-pandemic seasonality has resumed being the bigger driver of the trends.

In either event, for forecasting purposes the numbers remain either mildly or strongly positive.

Industrial production surged in January, further evidence of an AI data center led rebound

 

 - by New Deal democrat

Industrial production is much less central to the US economic picture than it was before the “China shock,” but it remains an important if diminished economic indicator, particularly since the month it has peaked in the past has typically been the month the NBER has chosen as the economic cycle peak.

In January, headline industrial production (blue in the graph below) rose 0.7%, establishing another new post-pandemic high (although it remains below its 2018 all-time high).  Manufacturing production (red) increased 0.6%, and was the highest since October 2022: 

Recently the issue of utility production, which has been supercharged both by the needs of crypto mining and AI data center construction, has become important to the overall numbers. This increased another 2.1% in January to another all-time record:

This was a 0.7% increase YoY. 

There has been all kinds of evidence from the regional Fed reports, and to a lesser extent the ISM manufacturing index, that production has been rebouding since the fall. This morning’s report was powerful further evidence of that rebound. A big caveat, particularly with the impact of tariffs, is how much of this broader rebound, like utility production, is contingent on the AI data center Boom continuing.

Housing short term indicators say “recession;” longer term indicators suggest “recovery” is close

 

 - by New Deal democrat

When I updated this information in September, I wrote that “a puzzling relationship this year has been that the housing data has been classically recessionary for a number of months, and yet the economy has not rolled over.” I concluded that update by noting that while the report was “very much recessionary, [ ] in some YoY comparisons, I would expect further damage before the actual onset of one. But that could easily occur within the next four to six months.” 

Indeed, some of that did happen in the October report released last month. On the other hand, I noted that the fact that permits did rebound for two months and units under construction did not decline further argues for the possibility of a bottom in the housing market and the proverbial “green shoots.”

So the short term outlook and the longer term outlook may have begun to diverge.

Let me start by reiterating the basics: mortgage rates lead sales, which in turn lead prices, which in turn lead inventory. Sales, in the form of permits and starts, are long leading indicators. Inventory, in the form of units under construction, is a shorter leading indicator. Further, as has become increasingly important in the last year, employment in residential construction, and the inventory of houses for sale, are the last to turn down before a recession. Here’s the historical pre-pandemic look at that last relationship:

Per my opening remarks above, the housing market has been in recessionary territory for many months. So let me take the data out of my usual order, and focus on the short leading indicators first, before discussing the longer leading ones.

To wit, housing units under construction declined -1.5% to 1.277 million annualized, the lowest number since 2021, and -25.5% below their late 2022 peak:

There have been only two times in the past 50 years when such a decline did not yet give rise to a recession: in 2007, when the decline was just -0.1% lower, at -25.6%, and 1991, when the decline was 28.2%. As is apparent from this, just a further -2.8% decline from here would give us an all-time decline without a recession.

Now let’s look at the post-pandemic record of residential construction employment and new housing inventory for sale in comparison with units under construction:

Both of the above “final shoes to drop” have in fact dropped. With the recent revisions, employment in residential construction peaked in March 2024, although it is only down -1.5% from there. New housing for sale peaked in March and May of last year, and is currently down -3.2%. Historically this decline in inventory under these circumstances meant a recession was already underway, although the decline in employment in the past had typically been on the order of -5% to -10% before a recession began.

But if the above discussion suggests that housing remains recessionary - in fact, by most measures a recession should already be underway, the more leading indicators suggest that if it does not happen in the next few months, it is not likely to happen at all.

First of all, although I won’t bother with a graph, mortgage rates have fluctuated in a range between just over 6% to 7.6% in the past 3+ years. In the past several months, interest rates have been near the bottom of their range, and permits have responded.

In more detail, in December total permits (red in the graph below) increased 60,000 to 1.448 million, . Single family permits, which convey the clearer signal, increased 18,000 to 876,000. Meanwhile the much noisier and slightly lagging housing starts rose 60,000 to 1.448 million units, over 100,000 higher than their recent October low:

Even with these gains, permits and starts remain in territory below their peaks sufficient to be consistent with a recession. On the other hand, all three measures are down less than -10% YoY, where in the past it has taken a more severe decline of greater than -10% to be consistent with with a recession:

Further, in the past once permits and starts have bottomed, it is a signal that a recovery from a recession will begin within the next 6 - 8 months.

Similarly, the YoY% decline in housing units under construction, which I discussed further above, has also improved significantly from its worst levels of 2025. In the past, this has almost always meant that a recession has already ended:

The only episode which is consistent with the present was the late 1980s, where YoY declilnes stabilized for several years before the 1991 recession began.

In short, most of the housing metrics are telling us that we should already be in recession, while the more leading ones are telling us that if one does not happen in the next few months, it is unlikely to happen at all.

I continue to believe that the decisive factor which has prevented the economy from rolling over so far (unless the government shutdown last autumn proves to have been the precipitant) is the AI Boom, which has led to stock market gains, and a “wealth effect” increase in spending by the uppermost income consumers.

Short and medium term inflation, interest rates, and the overstretched consumer

 

 - by New Deal democrat

The deluge of data resumes tomorrow with housing permits and starts, industrial production, and durable goods orders. In the meantime, let me make a few “big picture” observations of the economy, and in particular, the short and longer term trends in inflation and interest rates.

1. Short term inflation

A very big outlier in observations of inflation in the past several months has been Truflation, which updates is US CPI calculation daily, presumably based on masses of prices posted online from commercial sites. In the past two months, its measure of YoY inflation has declined precipitously, from 2.66% in mid-December to as low as 0.68% last week:

This has met with open skepticism in some quarters, fueled in part by Truflation’s unfortunate hire of T—-p crony E.J. Antoni just as the big decline started to occur. So this is very much an acid test of the site’s credibility.

According to Truflation, the median time lag between their observations and when the changes show up in the official CPI has historically been roughly two months:

Here is their post-pandemic view:

If their information is accurate, there should be a big decline in YoY CPI no later than in the report for March. And the only way that happens is if there are widespread actual price reductions.

It’s at least within the range of possibility that could happen.

In the first place, official CPI less shelter has paced the Truflation numbers in 2025:

Note that, similarly to Truflation, YoY CPI excluding shelter decelerated sharply from 2.2% to 1.4% between January and April 2025, then gradually increased through the remainder of the year through September 2025, peaking at 2.7% before declining to 2.0% in January. A similar decline through March would take it down to about 1.5%.

And FWIW, there are anecdotal reports of price cuts, particularly for staple food items. In my own neck of the woods, I have seen the price of store brand sodas, which doubled from $0.67 to $1.33 during the pandemic, cut back in the past few weeks to $1.00.

We will see. I’ll keep track of this over the next several months.

2. Longer term inflation

As I have noted many times since the pandemic, house price indexes have a multi-decade record of leading the official CPI measure for shelter costs by roughly 12-18 months. Here is the most recent comparison:

Over the past two years, the FHFA Index has declined from a YoY high of 7% to a low of 1.7% in October. Similarly, the Case Shiller national index has declined YoY from 6.6% to 1.4%. And the house price indexes typically are more volatile than the official CPI shelter index, suggesting it could decline to under 1% over the next 18 months. Beyond that, the median price for existing homes increased only 0.3% YoY through January. The median prices for new single family homes, meanwhile, have actually declined by an average of roughly -2.5% YoY for the past 3 years.

While the relationship isn’t perfect (note, for example, that while YoY price changes in the repeat sales indexes measured roughly 1.5% during mors ot hte 1990’s, CPI for shelter remained in the 3%-3.5% range), the likelihood is that shelter prices in the CPI will not accelerate YoY for the next 12 -18 months, and may very well decline under 2%, making the Fed’s official “target” more attainable (depending on other costs, such as the volatile price of gas, of course).

If, in the face of tariff increases being passed on to consumers, there are signs that inflation might moderate, that is almost certainly due to weakness in consumer spending, at least by the lower 80% of the income distribution. In other words, inflation might be moderating for the same reason it does during recessions: consumers simply cannot afford price increases.

3. Interest rates

A stagflationary scenario is likely playing out in interest rates as well.

The below graphs all compare Treasury rates for the 10 and 30 year durations (orange and gold) with the Fed funds rate (black) and mortgage rates (blue, minus 2% for easier visual comparison).

During the 1980s through 2019, when the Fed lowered interest rates, US Treasury interest rates and mortgage rates followed, albeit not with the same intensity:

But the post-pandemic comparison is more problematic:

Not only have the 10 year bond and mortgage rates not gone down in lockstep with Fed rate cuts, but they haven’t followed the last several rate cuts at all. And the 30 year bond has not followed at all. Interest rates on 30 year Treasurys are just as high now as they were when the Fed funds rate was at its peak 1.75% higher than it is now.

This is reminiscent of the stagflationary 1970s, shown below:

In the 1970s, both the 10 year Treasury and mortgage rates barely responded to Fed rate cuts. This was because bond traders well understood that the underlying inflation dynamics over the medium term were poor.

It is hard to escape the implication that bond traders have similarly responded to the US fiscal situation, as typified by the Big Bad Budget Bust-out Bill last year, as portending the necessity of higher interest rates in order to persuade bondholders to purchase US Treasurys. And that will bleed into longer term consumer rates as well.

The overall picture is that of an overstretched US consumer, unable to absorb price increases, and driving recessionary-type price concessions from sellers, with little prospect of longer term relief as interest rates are unforgiving. 

Real aggregate nonsupervisory payrolls remain relentlessly positive

 

 - by New Deal democrat

Today is Presidents’ Day, so there are no official economic data releases; and there will be no significant releases tomorrow either, before a torrent of both timely and delayed data from Wednesday through Friday, including GDP for Q4.

In the meantime, because of the January updates for employment and inflation last week, one of my important series for forecasting purposes can be updated as well: real aggregate nonsupervisory payrolls.

To recap, this series represents the total amount of paychecks in the economy for all workers except bosses, adjusted for inflation. It is noteworthy not just because the data goes back 60 years, but because it make real world sense: if ordinary working families have less money to spend in real terms, they are likely to cut back spending, and that retrenching brings about a recession.

First, here is the entire historical series. Note it is presented in log scale, so that later data does not obliterate earlier data:

With the exception of the COVID lockdown recession, the series has almost always turned down shortly before a recession has begun; or at very least turned flat.

Here is the same data presented in a YoY fashion:

With the exception of the 2002 “double-dip,” real aggregate nonsupervisory payrolls turning negative YoY has been a perfect indicator of recession, and remaining positive has been a perfect indicator of expansion; and further has typically turned negative within 2 months of the data of onset.

Now here is the post-pandemic look at the absolute series:

The trend has been almost relentlessly higher.

And here is the YoY look:

Again we see that it has never been negative, indeed has never shown less than 1% growth during this entire period. 

The one caveat is that because of the poor “shelter” kludge during the government shutdown, which suggested that rent and owners’ equivalent rent had only grown by 0.1% during those two months (vs. their typical 2025 average of 0.3%), the YoY total change should probably be between 0.2% and 0.4% lower, and that problem will persist through next October.

But even so, real aggregate nonsupervisory payrolls are sending an important signal that, despite virtually nonexistent job growth, wage growth has been strong enough to continue to power consumer spending, which in turn is negativing the onset of any recession in the next few months.

Weekly Indicators for February 9 - 13 at Seeking Alpha

 

 - by New Deal democrat

My “Weekly Indicators” post is up at Seeking Alpha.

There were no significant changes in the past week. In particular, despite the massive downward revisions to employment showing almost no new jobs were added to the economy last year, the best real-time measures of consumer spending, including such discretionary things as dining out at restaurants, continue not just to be positive, but are becoming even *more* positive in the past few months.

As usual, clicking over and reading will not just bring you up to the virtual moment as to the state of the economy, but reward me a little bit for my efforts in collecting and organizing the data for you.