A close observer of the economy might be tempted to ask, “What is going on?”

The stock market continues to do well (subject to day-to-day presidential musings) and GDP growth is strong, while nearly every other measure of economic activity has stagnated or is in deep decline. Why? I have a three-part answer.

First, tariffs continue to see-saw the U.S. economy.

Since President Donald Trump’s Liberation Day, manufacturing has lost 42,000 jobs nationwide. At the same time, manufacturing GDP rebounded from its first-quarter decline. Most of that growth appears to be due to accelerated production in advance of tariff price increases.

The strongest evidence for the “beat the tariff” economy comes from the enormous spike in imports by large manufacturing states. Indiana, the most manufacturing-intensive state, saw imports double from the previous record in January and February, then jump to almost four times the previous record in March. As of July, they were back to mid-2023 levels.

This import spike meant manufacturers could stockpile as much as eight months’ worth of production before the tariffs came due. This would explain why the sector is now shedding jobs at what appears through murky data as an accelerating rate. It also explains why tariff-related price increases appear most heavily in wholesale prices on products not yet at stores.

So, the economic drag of tariffs is just starting and will take many more months to be felt in full.

Second, the stock market continues to grow.

Most Americans, or at least the 62% who own stocks, have been pleasantly surprised over the last year at the sustained growth of their portfolios. But, if you look under the hood, the growth offers some room for worry. Since the public release of the ChatGPT artificial intelligence platform in late 2022, 7 out of 10 dollars of stock growth have come from a handful of AI-related firms – maybe 10% of the S&P 500.

At the same time, nearly all the business investment this year has been concentrated in AI infrastructure (primarily data centers). This investment might be warranted for a number of reasons. There may be a huge advantage to a single business that builds a breakout AI technology that crushes the others. If so, investors may be smart to make broad investments.

Likewise, most of us who have worked with AI believe that it will have enormous long-term productivity benefits to the global economy. At the same time, most of us have seen some real current limitations that temper the most optimistic claims of productivity growth.

Still, the most immediate way to think about AI is not the bullish stocks of AI producers, but whether AI-consuming firms are seeing growth in their valuation. That is not happening. Neither are job losses occurring in occupations in AI-using firms, or among workers with a college degree, who are expected to be most at risk of AI job losses.

Since January, the unemployment rate of high school graduates spiked from 5.1% to 6.7%, while college grads saw their unemployment rate rise from 2.3% to 2.7%. That is not the AI revolution coming for white-collar jobs. It is tariffs.

The third piece of the puzzle is putting this all together in the GDP data.

The AI investment has fueled stock growth and, at the same time, is propping up GDP growth. Household consumption is up, but half of all the growth is fueled by the richest 10% of Americans. This is almost certainly the wealth effect of higher stock holdings allowing more affluent families to buy before tariffs. At the same time, more than 90% of business fixed investment is in AI-related infrastructure.

So, the strong GDP growth is partially from the pre-tariff purchase effect, but most of it comes from the AI boom – through either business investment or folks spending more due to stock market gains.

I have seen much speculation that the AI boom is a bubble. It may be, but I don’t know and neither does anyone else.

There could be vast over-investment in AI infrastructure, which would lead to numerous bankruptcies and business failures. Or, the exuberance surrounding the short-term effect of AI could be wholly rational, leading to vast economic gains as our economy grows quickly.

There’s just no analytically effective way to make that prediction with confidence, though that hasn’t stopped lots of folks from trying. As for me, I would be unsurprised if the AI infrastructure boom, and stock boom, continued at a high pace for many months or longer. I would be equally unsurprised if we saw investment in new data centers slow to a crawl and AI stock valuations drop by one-third in the coming weeks.

There is an analytically effective way to evaluate the effect of tariffs on the U.S. economy. We’ve had lots of tariff shenanigans in the past, though none carry the deep uncertainty of this one. Those experiences have ranged from mildly bad to very bad. We are in the very bad range right now.

It is altogether possible that AI investment will continue to keep the U.S. economy from the recession that tariffs promise, at least for a few more months. It is equally likely that investors will flee from AI stocks and lenders will become much more wary of data centers and their associated energy and broadband infrastructure.

What is clear is that without the AI boom, the U.S. economy would be in recession right now, and it may be anyway. We don’t feel it yet. The jobs numbers weren’t released in October and price increases aren’t yet broadly apparent in stores.

In many ways, this moment feels much like late 2007 and early 2008. The economy appeared to be stable, even as we entered recession. Labor markets slowed less than today, but consumption continued robustly for months into a recession. Some major stock indices continued to rise more than six months into the Great Recession.

I’m not predicting a Great Recession redux, or even a stock bubble. But, it is important to make clear that nearly all the obvious economic risks are on the downside of economic performance. This would be an ideal time to end tariffs, carefully trim regulations and hope for a healthy 2026. That seems vanishingly unlikely.