The Recession That Never Recovers

The Setup

There is a version of this thesis that sounds alarmist. I want to be clear that this is not that version. This is a thesis built on publicly available economic data, labor market research from JPMorgan and Stanford, historical precedent from the last three recessions, and the observable behavior of companies that are already replacing workers with AI during a period of economic growth. What happens when those same companies face an actual contraction is not speculation. It is the logical extension of what they are already doing.

The argument has three parts. First: the recession was forming before the Iran conflict began, and the war is accelerating the timeline, not creating it. Second: when the layoffs come, AI gives companies a tool to permanently replace positions that would have been refilled in every previous recovery. Third: the combination of a cyclical downturn with a structural technology shift targeting 45% of American employment will produce the most prolonged jobless recovery in modern history, with consequences for consumer spending, housing, tax revenue, and social stability that extend years beyond the recession itself.

The Economy Was Already Breaking

The narrative that the U.S. economy was healthy before the Iran conflict is not supported by the labor market data.

Average monthly nonfarm payroll gains collapsed to14,000 over the six months through January 2026, down from 122,000 in 2024. That is not a slowdown. That is a labor market that stopped growing. Fed Chair Jerome Powell stated that he believes payroll data has been overstated and that revised numbers will show the U.S. was actually losing jobs since April 2025.

Unemployment rose from 4.1% to 4.6% through the end of 2025, the highest in more than four years. The Sahm Rule recession indicator, which signals the start of a recession when the three-month average unemployment rate rises 0.50 percentage points above its 12-month minimum, had been flirting with its trigger threshold. Job openings fell. Wage growth decelerated. Consumer sentiment weakened. Hiring was described by Deloitte, Morgan Stanley, and the Conference Board as having entered a "low-hire, low-fire equilibrium," a state where companies are not actively cutting staff but have stopped bringing anyone new on board.

Stanford's SIEPR published its 2026 outlook noting that the labor market had settled into a pattern where layoffs remained low but hiring had dried up. The Conference Board revised its GDP growth forecast to 1.6%, well below trend. Morgan Stanley put the probability of a mild recession at 15% even before the war began.

This was an economy running on fumes. The headline GDP number looked fine because AI-related business investment accounted for roughly two-thirds of GDP growth in the first half of 2025. Strip out the AI capex boom and you were looking at an economy that was barely growing.

The War Is the Accelerant, Not the Cause

The Iran conflict and the closure of the Strait of Hormuz did not create these conditions. It took an economy that was already weakening and hit it with simultaneous supply shocks across energy, fertilizer, and shipping.

Oil went from $67 to $103 in seven weeks. Gasoline hit $4 nationally, $5.89 in California. Diesel surged 45% to $5.45, which flows directly into the cost of everything that gets transported. The March CPI came in at 3.3%, up from 2.4% in February, the biggest monthly jump in nearly four years. Economists are projecting April could top 4%.

But the deeper crisis is fertilizer. Urea prices have surged to $858 per ton, blowing through every optimistic scenario. 70% of American farmers cannot afford full application. That means lower crop yields in September, tighter grain supplies through winter, and elevated grocery prices well into 2027. Food inflation operates on a biological clock, not a market clock. Even if the strait reopens tomorrow, the damage to this planting season is already done.

The combination is toxic: energy costs squeezing consumers, food inflation building with a delayed fuse, headline inflation accelerating toward 4%, and an economy that was already failing to create jobs before any of this happened. This is the setup for a recession. The war is just the match.

The Fed Is Trapped

The Federal Reserve faces a dilemma it has not confronted since 1990: inflation rising while the economy weakens. The textbook response to rising inflation is to raise interest rates. The textbook response to a weakening economy is to cut them. You cannot do both.

With CPI heading toward 4% and the labor market barely producing jobs, the Fed has no good moves. If it cuts rates to support employment, it risks fueling inflation that is already being driven by supply shocks (energy and food costs) that rate cuts cannot fix. If it holds rates or raises them to fight inflation, it accelerates the economic slowdown and pushes the labor market from stagnation into contraction.

The last time the Fed faced this trap was 1990, when Iraq's invasion of Kuwait spiked oil prices into an already weakening economy. The result was a recession. The parallels to today are uncomfortably precise: a Middle Eastern conflict, an oil shock, rising inflation, a softening labor market, and a central bank with no clean options.

But in 1990, there was no AI. When companies laid off workers during that recession, the jobs eventually came back because there was no alternative to human labor for most white-collar functions. That is no longer true.

The Pattern Nobody Learns From

Every recession since 1991 has followed the same structural pattern, and understanding it is essential to grasping why the next one will be different.

In a recession, companies cut costs. The most visible cost is labor. But firing employees is unpleasant, and during normal times, executives delay the decision. A recession gives them permission. It provides cover. "We had to cut because of the economy" is an acceptable explanation in a way that "we replaced you with software" is not.

During the 2001 recession, companies that had been slowly adopting internet-based tools used the downturn to complete the transition. Administrative assistants, bookkeepers, filing clerks, and travel agents were replaced by software that had existed for years but had not been fully deployed. When the economy recovered, those positions did not come back. The software was already doing the work. GDP recovered by 2002. Employment did not recover until 2005.

During the 2008 recession, the same mechanism operated at a larger scale. 8.7 million jobs were eliminated. The economy officially recovered in June 2009. Employment did not return to pre-recession levels until May 2014, more than five years later. The jobs that came back were disproportionately in lower-wage service sectors. The middle-skill, middle-wage positions (bank tellers, manufacturing supervisors, data entry clerks) disappeared permanently.

Academic research from the CEPR confirmed the mechanism: in each recession, the permanently eliminated jobs were the routine ones. The share of American workers in routine occupations dropped from 55% to 40%across these three recessions. Each downturn ratcheted the number down, and it never recovered between cycles. The recession did not create the automation. It accelerated it.

Why This Time Is Worse

In August 2025, JPMorgan senior U.S. economist Murat Tasci published research that laid out the thesis in institutional terms. His analysis was direct: previous jobless recoveries eliminated routine occupations. The next recession will eliminate non-routine cognitive occupations. Knowledge workers. White-collar professionals.

Non-routine cognitive work now accounts for approximately45% of total U.S. employment, up from 30% in the early 1980s. These are the jobs that previous recessions did not touch: analysts, marketers, content strategists, junior attorneys, software developers, HR managers, financial advisors, project managers. The people who sat in offices, worked on computers, and assumed their judgment and communication skills made them irreplaceable.

AI changes that equation. Large language models can draft legal briefs, write marketing copy, analyze financial statements, generate code, summarize research, create presentations, manage customer service interactions, process insurance claims, and perform dozens of other tasks that constitute the daily work of knowledge workers. Not perfectly. Not without human oversight. But well enough, and cheaply enough, that a company under pressure to cut costs can replace three people with one person and an AI subscription.

Tasci's conclusion deserves quoting in full: "A much larger unemployment risk and anemic recovery prospects for these workers might cause the next labor market downturn to look pretty dismal."

"Pretty dismal" is JPMorgan being diplomatic. If the pattern from the last three recessions repeats, but the target shifts from routine workers (40% of employment) to knowledge workers (45% of employment), the scale of permanent displacement will exceed anything in the postwar era.

The Automation Arms Race

Companies are not waiting for a recession. They are already replacing workers with AI during a period of positive GDP growth and record profits.

Block (formerly Square) cut 4,000 employees in February 2026, roughly 40% of its workforce, with CEO Jack Dorsey explicitly stating that AI had made the roles unnecessary and predicting that "within the next year, the majority of companies will reach the same conclusion." Microsoft laid off 6,000 workers in May 2025, mostly programmers, after CEO Satya Nadella confirmed that 30% of the company's code was now written by AI. IBM cut 8,000 positions as AI agents took over its HR department. Amazon eliminated 14,000 corporate roles. Klarna replaced 700 customer service workers with AI and publicly celebrated it.

In total, 55,000 job cuts were directly attributed to AI in 2025, according to Challenger, Gray & Christmas, out of 1.17 million total layoffs (the highest level since the 2020 pandemic). Another 30,000 have been attributed to AI in the first months of 2026.

A March 2026 academic paper from the University of Pennsylvania titled "The AI Layoff Trap" formalized what is happening. The researchers showed that even when companies know that mass AI displacement will erode the consumer demand they depend on, competitive pressure forces them to automate anyway. No individual firm can stop because its competitors will not stop. The researchers concluded that only external policy intervention (specifically, a Pigouvian automation tax) can break the cycle. Capital income taxes, universal basic income, worker equity participation, and upskilling programs were all modeled and found insufficient.

This is happening during growth. Imagine what happens during a contraction, when cutting costs is not optional but existential. A recession will not slow the automation arms race. It will accelerate it beyond anything we have seen, because every company will be looking for ways to do more with fewer people, and AI will be right there, ready, proven, and cheaper than it was last quarter.

The Demand Death Spiral

This is the part of the thesis that makes me think this recession could be the worst in modern history.

Consumer spending accounts for approximately 70% of U.S. GDP. Consumer spending requires consumers with income. Income requires employment. If AI permanently eliminates millions of knowledge-worker jobs during a recession, and those workers do not get rehired during the recovery because their functions have been automated, then consumer spending does not recover either. GDP bounces back because corporate productivity improves (fewer workers producing the same output), but the demand side of the economy contracts permanently.

The math is straightforward. If 45% of U.S. employment is in non-routine cognitive work, and a recession triggers even a 10% reduction in those positions that is not recovered (the approximate magnitude of routine-job losses in each of the last three recessions), that is roughly 7 million permanent job losses. Seven million workers who are not spending on housing, cars, restaurants, retail, healthcare, education, or anything else that drives the consumer economy.

The labor force participation rate is already projected to fall from 62.6% to 61% by 2030 and to 55% by 2050. Those projections were made before incorporating a recession scenario. Importantly, the forecasts show the unemployment rate remaining relatively stable, suggesting that displaced workers will not show up in unemployment statistics at all. They will simply leave the labor force entirely. They will stop looking. They will not be counted.

Fewer workers means less tax revenue. Less tax revenue means reduced government services and infrastructure spending. Reduced government spending means fewer jobs in the public sector. This is the demand death spiral: each round of job losses reduces the spending that supports the next round of jobs.

The Counter-Argument

A fair analysis requires steelmanning the opposing view.

Technology has always created more jobs than it destroys. This is historically true. The agricultural revolution, the industrial revolution, and the information revolution all displaced massive numbers of workers and ultimately created more employment than they eliminated. The World Economic Forum's 2025 Future of Jobs Report projects that AI will displace 92 million jobs by 2030 but create 170 million new ones, a net gain of 78 million. If this pattern holds, the permanent displacement thesis is wrong.

The timeline may be slower than feared. Stanford's SIEPR noted that historically, firms take many years to integrate new technologies. Electric motors took 30 years to reach 80% adoption after 1900. AI adoption may follow a similar curve, meaning the displacement happens gradually enough for the labor market to adjust.

New categories of work will emerge. Just as the internet created jobs that did not exist in 1990 (social media managers, UX designers, cloud architects, data scientists), AI may create categories of employment that we cannot currently imagine. The "centaur" model of human-AI collaboration could create an entirely new class of knowledge work where the human provides judgment and the AI provides knowledge.

These counter-arguments deserve serious consideration. But they share a common weakness: they assume the transition will be orderly. The thesis of this article is that a recession makes the transition disorderly. It compresses what might take a decade into two years. It forces companies to make displacement decisions under financial pressure rather than strategic planning. And it hits all at once, across every industry, rather than rolling through sectors gradually. The historical precedent is not "technology always creates jobs eventually." The historical precedent is "technology destroys jobs during recessions and creates them during expansions." If the destruction phase is larger than any previous cycle, the creation phase will need to be proportionally larger to compensate. That is a bet, not a certainty.

What This Means for Portfolios

If this thesis plays out, the investment implications are significant.

Companies that benefit from AI-driven productivity. The same companies replacing workers with AI are the ones whose margins expand during the transition. The hyperscalers (Alphabet, Amazon, Microsoft) that sell the AI infrastructure, and the AI companies themselves (Anthropic, OpenAI), are positioned on the winning side of the displacement. Their revenue grows as every other company pays for AI tools to replace labor. This is a deeply uncomfortable investment thesis, but the data supports it.

Companies exposed to consumer spending decline. If millions of knowledge workers are permanently displaced, discretionary consumer spending contracts. Restaurants, retail, travel, luxury goods, and entertainment are all exposed. The consumer staples with pricing power (food, household necessities) are relatively defensive because people still need to eat.

Housing. A prolonged jobless recovery concentrated in white-collar workers, the demographic that drives mortgage demand, suburban housing purchases, and rental demand in expensive metro areas, would pressure housing prices in the same markets that have been most inflated.

Dividend and cash-flow stability. In a prolonged downturn with an uncertain recovery timeline, companies that generate reliable free cash flow and pay consistent dividends become disproportionately valuable. The boring utilities thesis and the defense stocks thesis both hold up in this scenario: regulated revenue, government contracts, and essential services do not depend on consumer discretionary spending.

The Bottom Line

The recession was already forming before the first shot was fired. Payroll growth had collapsed. Unemployment was rising. Hiring had flatlined. The war poured accelerant on a fire that was already smoldering: $100 oil, $858 urea, 3.3% CPI heading toward 4%, and a Fed with no clean options.

But the recession itself is not the thesis. Recessions end. GDP recovers. Markets rebound. The thesis is about what does not recover: employment. Every recession since 1991 has produced a jobless recovery where eliminated positions never came back, because technology had advanced enough during the downturn to make the workers unnecessary. In 1991 it was factory workers. In 2001 it was administrative staff. In 2008 it was middle management. Each time, the target was routine work.

This time, the target is knowledge work. The 45% of American employment that consists of the analysts, marketers, programmers, paralegals, project managers, and financial professionals who assumed their judgment made them irreplaceable. AI does not need to replace them entirely. It just needs to make one person with AI as productive as three people without it. When a recession forces the decision, companies will keep one and let two go. And when the economy recovers, the AI will still be there. The two will not be rehired.

That is why this recession, when it comes, could be the worst in modern history. Not because the GDP contraction will be the deepest. But because the recovery will be the emptiest. The jobs will not come back. Not because the economy cannot support them. Because the technology has made them permanently unnecessary.

At Wealth Engine Pro, the philosophy is data over narrative. The narrative says AI creates opportunity. The data says it also creates displacement, and the displacement accelerates during downturns, and the jobs lost during downturns to technology have never come back in any previous cycle. That pattern, applied to the largest category of American employment, at a moment when the economy is already weakening, is not a story anyone wants to hear. But the numbers do not care what we want.