The entire US economy is downstream of the AI buildout dam, and that dam is almost certain to fail

 

 - by New Deal democrat

I’ve been ambivalent about whether the mania for AI is a Boom or a Bubble. That’s because whether its use is transformative or, like Steve Jobs’ claims that the Segway would revolutionize transportation, not so much. 

But what has become increasingly clear is that virtually the entire economy has been downstream of its growth. Here’s just how exponential the growth of spending on AI data centers has been:

The stock market has Boomed:

But the stock market Boom has been based almost exclusively on that AI spending:

And the wealth effect of that stock market Boom has led to accelerating spending by the affluent, as shown in Redbook’s weekly retail sales data:

And, as consumption leads employment, that spending has spread out to an increase in not just goods-producing jobs (orange, right scale), but broad service providing jobs (blue, left scale) as well:

Meanwhile, the spike in inflation has caused consumers, most of whom do not have much if any stock holdings, to dig deeper into their savings to a near record low:

In summary, if it weren’t for the Boom in AI related spending, it’s likely that consumer spending would be flat or even negative (remember that real average hourly wages have gone negative YoY, and real aggregate nonsupervisory payrolls have grown just barely in the past year:

In other words, the US economy would likely be in a recession right now.

That’s an awful lot of weight that is being borne by one small sector of the economy.

But I have come to conclude that, even if AI’s usefulness were to live up to its hype, there is one aspect that strikes me as clearly a bubble.

That’s because there are 10 or 20 companies all rushing to build out full-blown all-encompassing data centers. But almost certainly when it all shakes out, even under the best of circumstances there are likely to be only 2 or 3 left. All of the others - and their huge construction and usage footprint - are likely to vanish.

In other words, whether or not the AI Boom is like the dotcom bubble of 1999-2000, it is almost certainly like the browser wars of the 1990s, when there are a variety of providers like Alta Vista and Ask Jeeves, that all got eclipsed by Google and all but vanished from the scene.

And think of the auto industry. Early on there were dozens of manufacturers. But by the end of the second World War, there was one dominant company (GM), one secondary company (Ford) and one also-ran (Chrysler).  In computer chips, there has been one dominant company (Intel) and one (until recently) also-ran (Micron). As Ron Insana pointed out, “In 1895, there w[ere] … 1,000 companies [that] made bikes as the new model of transport. By 1905, they were going out of business.” 

 

Fortunately, I don’t have to write a more extensive piece, because it turns out someone else named Dan Wertman of Noetica/Thomson Reuters got there a short time ago, so I will quote them at length:

“Most people liken the AI boom to the dot com bubble. But the right comparison is the lesser-known portal wars. …

“Let’s go back to 1998. The internet had just gone mainstream, and a new kind of company was taking over the web: AltaVista, Excite, Lycos and Yahoo were each racing to become your home base online–the ‘interface’ to the internet. They competed on adding vertical workflows: features like news, email, weather and shopping. Venture capital poured in and they grew fast. For a moment, it looked like any one of them, or all of them, could win, each differentiating themselves in domains in which they were marginally better from the other. 

Then a pair of Stanford graduate students created a new model, a search algorithm called PageRank, which used the web’s own link structure as a proprietary data signal. … [W]ithin a few years, PageRank became what we know as Google and every other portal had been rendered irrelevant.

“We are watching the same movie with AI startups today. Thousands of companies are building products on top of the same AI foundation models — OpenAI’s GPT, Google’s Gemini, Anthropic’s Claude – with no added proprietary content or data, only workflows, each aiming to be the “portal” for their vertical. They have different names, different user experiences, different pitches to investors. What most of them share is that the intelligence powering their product is available to every competitor, every established company, and increasingly to ordinary consumers at low or no cost because they’ve added nothing proprietary to enhance their offerings.”

It’s as if people build 20 Hoover Dams, when only 1 was needed. And the rest will ultimately sit idle - and probably fail.

So, dear reader, let me conclude. This aforementioned AI buildout, and that buildout  (vs. whatever software value AI has) is all but certain to crash. When, and over what period of time, we do not know.

Hopefully not while the rest of the economy, as it is now, is downstream.

Producer prices suggest 5%+ YoY CPI, and very soon

 

 - by New Deal democrat

I just wanted to drop a brief note about this morning’s PPI report. In May headline producer prices rose 1.1% for the month (the second such increase in a row), and on a YoY basis increased 6.5%. Final demand goods prices are up 9.1% YoY, and final demand services up 4.9%:

So at the producer level, there is really no hiding place from price increases.

Meanwhile raw commodity prices 3.5% for the month, and are up 13.1% YoY. The last few monthly increases have been every bit as much as the worst of the post-pandemic inflationary surge:

This is an extreme reading that has been equalled or beaten only 5x in the past 80 years:

Note that in *all* of those cases, consumer inflation peaked at over 5% YoY, and in 4 cases close to or even higher than 10%, vs. 4.2% in the latest consumer inflation reading.

One important difference with the 2022 experience is that this time, shelter is not participating in the consumer inflation advance. But that is of little comfort since, as the below graph of CPI less shelter shows, that metric was higher by over 10% YoY at the 2022 peak:

Finally, I also want to spotlight that sometimes producer prices lead consumer prices, but sometimes they move simultaneously. In fact over the past 80 years, producer prices for goods (the longest extant main PPI statistic) and consumer price spikes YoY have typically peaked within two months of one another, and frequently simultaneously. Here is the historical look:

Faced with a spike in price for inputs, producers can either absorb the increases, pass them on to consumes, or some of each. The regional Fed indexes for the past few months have indicated rampant input price hikes, with much - but not nearly complete - pass-throughs to consumers. That seems to be what we are seeing in the comparison of producer and consumer price spikes so far now. But as pointed out above, it would be unusual if the YoY peak advance in consumer prices stayed below 5% - and that’s without any renewed Iran war escalation.

Of course, while that is “somewhat” good news for consumers, it does mean a hit to the bottom line for producers. And after lower profits tend to come things like abeyances in hiring, and maybe an increase in firing. But we’re not there yet.

More evidence for the return of post-pandemic seasonality in jobless claims

 

 - by New Deal democrat

Last week I wrote that “we’re coming up on the one year anniversary of that change of regime, so it will be interesting to see if the negative YoY comparisons continue, or if they fade away. This week’s numbers are noteworthy in that regard, because they suggest that - maybe - the pattern of increased claims into midyear is reasserting itself. We’ll find out over the course of the summer.”

This week saw both further evidence for the return of post-pandemic seasonality, as well as continued good YoY comparisons.

Let’s look at the numbers. Initial claims rose 4,000 to 229,000, the highest weekly number in 4 months. The four week moving average rose 4,250 to 219,000, the highest since the end of February. Continuing claims, with the typical one week delay, rose 24,000 to 1.795 million, still very low:

So it very much looks like the post-pandemic seasonality of high claims at midyear may have made a return. Exactly this week one year ago is when initial claims were at their peak for all of 2025, with the exception of one week in September. Over the course of the next 6 weeks they declined nearly by -30,000. If seasonality is back, this pattern will not be repeated.

On the YoY basis more important for forecasting, initial claims were down -6.9%, the four week average down -7.9%, and continuing claims down -7.8%, on par with most of the YoY comparisons over the past few months:

This continues to be a very positive signal for the economy over the next few months.

If the change of regime was a one-time thing, driven mainly by immigrant worker issues, then these good YoY comparisons will fade between now and the end of July. We’ll see. 

May CPI is “less bad,” but bad enough to edge consumers closer to recessionary income levels

 

 - by New Deal democrat

Just as forecast by the Cleveland Fed, the CPI in May increased 0.5% - less than the increases in March and April, but still too high. Core CPI, which excludes food and energy, increased only 0.2%. The YoY% gains both also increased, to 4.2% and 2.8% respectively. to increase to 3.8%. 

As per the last several months, in addition to my usual practice of focusing on shelter and any other “problem children” with outsize numbers, this month even more than last month it is important to note the impact on real wages and incomes. 

Let’s start with the YoY numbers for headline inflation (blue), core inflation (red), and inflation ex shelter (gold), which is up 4.7% YoY, the highest in over three years:

Last month shelter costs unexpectedly increased 0.6%. That was not revised down this month. Meanwhile the monthly gain was 0.3%, on the higher end of readings for the past year. This caused YoY shelter to increase another 0.1% to 3.4% (blue in the graph below). Rent rose 0.4% in the month and 2.9% YoY (gold), while Owner’s Equivalent Rent (red) rose 0.3% monthly and 3.3% YoY:

This is disconcerting, because all the other leading indicators for shelter has been telegraphing further declines. This could well go back to ramifications of the shelter inflation “kludge” used to estimate this sector during the government shutdown.

Now let me look at a few other present or former “problem children.” 

There was good news on several fronts, which I’ll note without accompanying graphs. First, both new and used vehicle prices continue to be asleep at the wheel, with new car prices *declining* -0.3%, and used car prices only up 0.1%. On a YoY basis, new car prices have only increased 0.2%, and used car prices have actually *declined* -2.0%. In fact, both new and used car prices have been virtually unchanged for the past 3 years. Since the onset of the pandemic, new car prices are up about 25% and used car prices up 20%, while hourly wages are up about 35%.

Next, food at home prices are only up 2.7% YoY, and food away from home only up 3.5%. Both of these had been rising over 4% YoY until recently, but now both are increasing less than wage growth.

But several other sectors continued show inflationary problems.

Transportation services (mainly vehicle repairs and insurance) followed vehicle prices higher, but had calmed for a brief period before the last several months. This has almost entirely been driven by motor vehicle maintenance and repair prices (likely becuase consumers are holding on to older cars for a longer period of time). In May, while insurance premiums declined -1.7%, and YoY were down -2.0%, maintenance and repair prices increased 0.8% for the month and 6.1% YoY.  In the entire sector prices declined -0.6% for the month, but were up 4.1% YoY:

Secondly, and more importantly, electricity prices jumped another 0.6% in May after a 2.1% in April, and have risen 5.9% YoY:

This is almost certainly due to the impact of the building of AI data centers.

Finally, let’s look at what this means for incomes. In May, nominally the average hourly earnings for nonsupervisory workers increased 0.2%, meaning that in real terms they declined -0.3%. Further, aggregate nonsupervisory payrolls increased 0.4%, meaning that in real terms they declined as well, by -0.1%. Here’s what real wages and payrolls look like in absolute terms normed to their recent peaks:

Real wages are down -1.1% from their peak and real aggregate payrolls are down -0.6%.

Neither of these mean that we are in a recession now. But the YoY comparisons are further cause for great concern. On that basis, real hourly wages are down -1.1%, while real aggregate payrolls did improve YoY by 0.1% to +0.8%:

The former is frequently associated with a near in time recession, and the latter usually crosses the “0” line to the downside within a month or two before or after the onset of a recession. In fact, real aggregate nonsupervisory payrolls have only been higher YoY by 0.8% or less without a recession occurring shortly thereafter for one month apiece in the 1960s and 1990s, and the 2002-03 period. And note that real aggregate nonsupervisory payrolls are currently only 0.4% above what they were last July.

Last month I wrote that “both measures of real wages and payrolls are sending a ‘yellow flag’ recession caution.” The same is true this month, although the yellow flag is a shade more orange, because at -0.6%, payrolls are only 0.1% above the -0.7% decline from peak in real aggregate payrolls have been about the median decline at the onset of past recessions.

In conclusion, this was another poor report, although “less bad” than the reports in March and April. Although gas prices have actually abated during the past few weeks, both the ISM manufacturing and services indexes indicated that there were widespread upstream price increases in May, that presumably have not yet filtered down to the consumer level. In other words, price pressures on consumers are likely to continue for at least several more months, even under optimistic scenarios where the Strait of Hormuz opens back up in the next few weeks. 

This has had a real, negative impact on ordinary consumer finances. In fact, I would go so far as to say we would already be in the opening month of a recession if consumers had not coped with this - so far - by very large increases in credit card balances.

Existing home sales report shows a sub-optimal equilibrium with rangebound sales, prices, and inventory

 

 - by New Deal democrat

First of all, my usual caveat: although they constitute about 90% of all housing sales, I don’t pay too much attention to existing sales because they are not nearly so important as new home sales, since the latter involve much more economic activity in the building process, plus more landscaping and furnishings.

But more than sales, since the pandemic the dynamics that have been more important have been prices and inventory. Because during the pandemic prices skyrocketed, and inventory cratered. It has been a long, slow arduous process of rebalancing since then. 

For the record, let me start with sales. These have been rangebound between 3.85 million annualized to 4.35 million for the past three years. And although they increased from April by 13,000, and were 6% higher than one year ago, they remained rangebound with sales of 4.17 million annualized in May. This is also only 3.2% higher compared with one year ago:

As you can easily see, the current range is well below the pre-COVID average of roughly 5.5 million annualized sales.

So now let’s turn to the metrics that most need to be normalized: prices and inventories. Note that these are not seasonally adjusted, so the only good way to look at them is YoY.

The median price for an existing home in May was $429,300, up a mere 1.3% from one year ago:

This is consistent with the near record low YoY increases (outside of the Housing Bust) in the Case-Shiller and FHFA repeat home sales indexes, and the slight YoY decline in new home prices as developers downsize to meet the market.

But the inventory of existing homes for sale remain well below their pre-pandemic levels. In May these were 1.55 million units annualized, up only 10,000 from one year ago, a 0.6% YoY increase. This is nowhere near what is necessary, as shown in the 10 year graph below:

The current level is still only about 80% of what it would take to return to pre-COVID normalcy.

Finally, last week the NAR also updated its information on new (red, right scale) and total (blue) listing counts:

As you can see, these are also very seasonal. In general, the number of listings has been improving. But, as the YoY% graph of the same data shows below, the improvement has all but come to a halt in the last few months, with active listings only up 2.2%, and new listings only up 2.1%:

This year the housing market has appearred to reach a sub-optimal post-COVID equilibrium, with sideways sales and prices, and at best slowly increasing inventory. Needless to say, the increase in mortgage rates since the onset of the Iran war has not been helpful. The US simply needs much more housing to be built to return to some kind of affordability.

Scenes of strength and weakness from the May jobs report

 

 - by New Deal democrat

It’s the Monday after the jobs report, and as usual there is no new data today. So let’s take a look at a few of the salient trends from Friday’s report.

First of all, this was the fourth good report in a row. Furthermore, it isn’t just goods producing jobs (red in the graph below, *2 for scale) that have rebounded (as has been signaled by the regional Fed and ISM reports since late last year), but service jobs (blue) as well:

To some extent, this may have been signaled by the uptick in monthly real retail sales (light blue in the graph below), which typically lead employment (red) by several months:

In the above graph, I also included real personal spending (dark blue), which have been steadily increasing YoY, and don’t seem to provide much explanatory power.

But in support of the idea that increase retail spending has led to an increase in service producing jobs, here is the last year of the weekly Redbook retail sales report:

The increasing trend in YoY sales is apparent. In fact, since the first of the year, there were only two weeks (in January) that saw gains of less than 6% YoY. And every week in the past six have been prints of over 7% YoY.

I trace this all back to the AI Boom (or bubble), which has led to a sharp YoY increase in stock prices, which in turn has likely led to a pronounced “wealth effect.”

Next, the unemployment rate remained steady at 4.3% for the third month in a row. So has my forecast for a decline towards 4% been busted? I don’t think so, as shown in the below graph of initial+continuing jobless claims (blue, right scale), the unemployment rate (orange, left scale) and the raw data on which it is based; namely, the number of unemployed vs. the number of people in the entire labor force (red):

The trend in the raw data is indeed a decline, especially if we look on a three month moving average basis. The steadiness in the unemployment rate has been a function or rounding. The bottom line is that I still expect the unemployment rate to decline towards 4% in the next few months, based on 60 years of history.

Finally, let’s look forward to how this Wednesday’s CPI report might impact important employment data. 

Friday’s report showed a nearly .25% increase in average nonsupervisory wages, and a nearly .45% increase in aggregate nonsupervisory payrolls, that rounded to 0.2% and 0.4% respectively:

Two important forecasting tools I use are real nonsupervisory hourly wages (blue in the graphs below) and real aggregate nonsupervisory payrolls (red). The first graph shows their absolute values, normed to 100 as of their recent peaks:

Real wages are down -0.9% from their February peak, and real aggregate payrolls down -0.7% from January.

As of today, the Cleveland Fed is forecasting that May inflation will be reported up 0.5% on Wednesday, which would increase those declines to -1.2% and -0.8% respectively. These would be significant declines frequently - but not always! - consistent with the onset of a recession.

Here’s the same data YoY for the past three years:

Real wages are already down -0.1% YoY, while real payrolls are up 0.7%.

Here is the historical pre-pandemic look at both:

A YoY decline in real wages has been a feature of every recession except for the shallow, producer-led 2001 recession; but from the 1980s through at least 2015 they were also negative for extended periods without there being a recession.

On the other hand, a YoY decline in real aggregate nonsupervisory payrolls has been a perfect indicator with the exception of the extended decline during 2002-03 (the one month each of nearly negative readings in the 1960s and 1990s were strike related).

Last May real average wages increased 0.4%, while real aggregate payrolls were unchanged. If inflation is as per forecast by the Cleveland Fed, YoY real average wages will be down -0.4% YoY, and real aggregate payrolls will be up 0.6%. If that happens, those will be yellow “caution” signals, but not “red flag” recession indicators.