In the fall of 2008 I was sitting in a political science classroom at Virginia Tech while the things we were studying were also happening outside the window. Lehman Brothers went down in September. I read the assigned material and then read the news, and the gap between the two kept narrowing until there wasn’t one.
By the following spring I was applying for jobs. Hundreds of them, through graduation and well past it. The entry-level listings wanted one to three years of experience, which is a contradictory sentence that bothered me massively. But I had no choice besides applying anyway.
I am doing that again now, at 38, with three kids and a lot more to risk than a dorm room and a meal plan.
Here is what changed in between, and it is not the thing people assume.
For nearly thirty years, recent college graduates had a lower unemployment rate than the country as a whole. Every year from 1990 forward. Through the dot-com bust, through the financial crisis, through my own graduation into the wreckage. The degree was worth something measurable even when nothing else was.
That reversed starting in 2018. Bloomberg reported in 2022, working from New York Fed data, that the trend had flipped in 2018 and that recent graduates had exceeded the national unemployment rate every single month since the start of 2021. Today the rate for recent graduates sits at about 5.6 percent, with underemployment — a term that refers to working a job that does not require the degree — at 42 percent. The national unemployment rate is 4.1.
We have to kill both of the easy explanations before we even get started. The flip predates the pandemic by more than a year and generative AI by four. Whatever did this was already a machine before anybody had a story to hang on it.
What made 2008 survivable
For starters, I have to make the counterargument because on its own terms the counterargument is correct.
By any aggregate measure, the 2008 financial crisis was worse than right now. The unemployment rate went from 5.0 percent in December 2007 to a peak of 10.0 percent in October 2009, with more than fifteen million people out of work. Employment fell by 8.8 million between the January 2008 peak and the February 2010 trough. Unemployment among people aged 16 to 24 hit a record 19.5 percent in April 2010. Recent-graduate unemployment kept climbing after I graduated and did not peak until March 2011, at 7.1 percent, well above today’s figure.
Line those numbers up and the period I graduated into wins the misery contest, and it isn’t close. Anyone who tells you the present economy is objectively worse than the 2008 financial crisis is wrong.
Here comes the big but — look at what was happening underneath. In March 2011, the month recent-graduate unemployment hit its all-time high of 7.1 percent, the national rate was 8.8. Graduates were 1.7 percentage points ahead of the country at the worst moment their own series has ever recorded. Recessions gut construction and manufacturing first, and those are the industries that employ people without degrees, so the diploma held its value precisely when work was hardest to find.
Today graduates are 1.5 points behind. But by all measures we are not in the midst of a massive financial crisis.
The crisis I graduated into was legible. Foreclosure signs went up on streets where you knew the families. Bank failures ran on the front page for a year. There was an event with a date, an industry with a name, a hearing schedule, and a policy response you could be for or against. That legibility is why we had TARP, Dodd-Frank, the Tea Party, Occupy, and a decade of argument about what had happened and who did it.
What we have now is a condition with no event. No date. No name. But most of all, there is no shared experience, because a very large number of Americans are not experiencing it at all.
The numbers are true and the numbers are lying
Start with the one everybody quotes.
The unemployment rate in July was 4.1 percent, down from 4.2 in June and 4.3 in May. That reads like an economy quietly improving. In the same report, nonfarm payrolls fell by 23,000. Employment went down. The unemployment rate went down with it, because labor force participation has dropped seven tenths of a point since January and the employment-to-population ratio has dropped half a point.
People stopped being counted.
There were 4.8 million people working part time who wanted full time. There were 5.9 million people outside the labor force who said they wanted a job and were therefore not unemployed. A rate that falls because people quit looking is a different economic condition from a rate that falls because people got hired. The problem is that these are numbers on a screen that no one wants to report about.
Then the part that explains why the entry-level market feels the way it does. The labor market has not collapsed. It has frozen, which is worse if you are trying to get in rather than stay in. The Economic Policy Institute found the hires rate has fallen to roughly where it sat in 2013 and 2014 — years when unemployment ran three full percentage points higher than today. Employers are not firing. They are also not hiring.
If you already have a job that arrangement is invisible. If you need one it is a wall.
The credit data pulls the same trick. The New York Fed’s most recent household debt and credit report went out under a headline announcing that delinquency transition rates held steady. Inside it, the share of mortgage balances newly falling ninety days or more past due rose from 1.22 percent a year earlier to 1.48. Home equity lines went from 0.88 to 1.15. Student loans went from 8.04 to 10.86. Across every category, the flow into serious delinquency rose from 2.45 percent to 2.83.
Steady is one word for that. Not the one I would use.
Then there is the one you feel every month. Every quarter the Federal Reserve publishes two credit card interest rates in the same table. One is the average across all accounts. The other is the average on accounts actually assessed interest, which means accounts belonging to people who carry a balance because they cannot pay it off. In the most recent release, the all-accounts rate fell from 21.00 percent to 20.94. The rate for people carrying a balance rose from 21.52 to 22.15.
The number describing people who can afford to pay went down. The number describing people who can’t went up, by ten times as much as the other one fell. Same table. Same quarter. Same economy, allegedly.
If you have been staring at a statement wondering how the news can keep saying rates are easing while your minimum payment climbs, that is the answer, and it is printed by the Federal Reserve on a fixed schedule for free. You were not confused.
The best argument against me
I do my best to point out the obvious, even if I disagree with it.
There is one piece of evidence that cuts hard against everything above, and it comes from serious people rather than from the comment section. It deserves a real hearing.
Real wages have risen. Adjusted for inflation, the typical American worker earns more per hour than before the pandemic, and the gains were largest at the bottom of the wage distribution. That is a well documented fact, and it is the single strongest piece of evidence that the economy has worked for ordinary people. If you only heard one statistic about the last five years, that would be a defensible one to hear.
There, I said it. So now we have to look at what it has meant lately.
Real average hourly earnings rose 1.8 percent over the year ending March 2025. By February 2026 that had slowed to 1.3 percent. By March, 0.3. By June, 0.1. And in the most recent Real Earnings release, covering July 2025 to July 2026, real average hourly earnings did not slow. They fell, by 0.2 percent. For production and nonsupervisory workers — the BLS category covering roughly 110 million private-sector employees, the cashiers and nurses and drivers and line cooks and clerical staff — they fell 0.1 percent.
Real weekly earnings technically rose 0.1 percent over that year. They rose because the average workweek got 0.3 percent longer. People took home slightly more by working more hours, not by being paid more for the hours they already worked.
USAFacts, working from the same BLS series, puts the number in dollars: nominal weekly wages grew 3.5 percent over the year while inflation ran 3.4, leaving real growth of about a dollar a week.
So it is imperative we grant the counterargument in full. Real wages did rise. The gains were real and they were broad. But then they decelerated for four straight quarters and went negative, and then thing that made your paycheck go up this year was working longer.
But nobody announced any of that because the five-year number still looks fine.
What it actually costs to be alive
Wages are one side of the ledger. How you spend them to live the other.
Nearly half of American renters — 22.7 million households, or 49 percent — now spend more than 30 percent of income on rent and utilities, a record in Harvard’s series. More than twelve million spend over 50 percent. Cost burdens rose in 88 of the 100 largest metropolitan areas between 2019 and 2024, and the burden has climbed the income scale: among renter households earning between $45,000 and $75,000, nearly half are now cost burdened.
That is not a poverty statistic. Census puts median household income at $83,730, which means that entire band sits at or below the middle of the country. This is not the bottom rung anymore. It is the median household, and half of them cannot afford their rent.
Health coverage does the same thing from a different direction. KFF’s 2025 benchmark survey puts the average family premium for employer-sponsored insurance at $26,993, with workers contributing $6,850 of that out of their own paychecks before a single deductible is met. Premiums rose six percent against general inflation of 2.7, and it is the first time in two decades that family coverage has risen six percent or more three years running. Insurers in sixteen states and the District have already asked regulators to approve a median 14 percent increase for 2027.
Buying a house has moved out of reach on the same timeline. The National Association of Home Builders reports home prices up 53 percent since 2019 against median household income up 24. First-time buyers have fallen to 21 percent of the market from 44 percent in 1981. The median age of a first-time buyer is 40, up from 29. More than half of adults between 18 and 24 live with their parents.
None of those numbers are hidden. Every one is published by a federal agency, a university research center, or a foundation, on a schedule, for free. They are simply not the numbers that lead the broadcast.
The people who are fine are not lying
This is where most writing on this subject goes wrong, and getting it wrong is why none of that writing accomplishes anything other than just taking up space.
There is a version of this argument that treats everyone doing well as either oblivious or complicit. That version is just as satisfying as it is useless. The people who are fine are not misinformed. Their 401(k) is up. Their house appraised higher than they expected. Their job is stable and their employer is not laying anyone off. Their mortgage payment is fixed, so the shelter inflation that is eating renters alive does not touch them. The unemployment rate is 4.1 percent and their own experience confirms it.
Nobody has deceived them. They are reading accurate data that accurately describes their life. If you are one of them, you are not doing a damn thing wrong.
Researchers at the New York Fed found that since the first quarter of 2023, real net worth for the top percentile grew by more than 25 percent while the middle 40 percent grew by less than 10, and that the gap came almost entirely from financial assets rather than wages. Consumption in the aggregate holds up because the people doing the consuming are the people holding the stock.
Economists at the Minneapolis Fed reviewed the evidence this year and concluded the divided-economy story is messier than the headlines suggest, with different datasets showing anything from a sharp split to no obvious pattern. Their most useful observation is not that the divide is imaginary. It is that measures built on income may understate it, because what drives spending at the top is accumulated wealth rather than what shows up on a W-2.
Grant them the methodological point in full. The direction survives it.
So the problem is not that half the country is wrong about the economy. Both halves are right about their own, and the machine producing both outcomes is the same machine. That is a much harder thing to explain than a lie, and it is why telling people to wake up has accomplished nothing.
One machine, two directions
Artificial intelligence spending is now large enough to hold up the national growth figures by itself.
Goldman Sachs projects AI capital investment at about 1.8 percent of U.S. GDP this year, climbing toward 2.5 percent next year. That is the broad measure, covering chips, construction, and everything downstream. Narrow it to the physical buildout alone and Epoch AI puts data center construction, compute hardware, and networking at roughly 0.8 percent of GDP in the first quarter of 2026 — a smaller slice of the same phenomenon, measured differently, pointing the same direction. Pantheon Macroeconomics found that with AI spending stripped out, American corporate investment in equipment would be negative, with investment in other equipment falling 17 percent in a single quarter.
Follow that money in both directions.
Going up, it inflates the equity prices concentrated in the top tenth of households, which is what makes those households confident enough to keep spending, which is what keeps aggregate consumption from cracking, which is what keeps the recession everyone predicted from arriving.
Going down, it lands on the electric bill. PJM Interconnection, which runs the grid for Maryland and twelve other states plus the District, has for three years held its capacity prices flat under a temporary ceiling negotiated with Pennsylvania’s governor. In an appendix to each auction report, PJM also publishes what the price would have been without that ceiling. Those figures run $388.57, then $529.80, then $554.72. Flat on the surface, up 43 percent underneath. In its planning documents PJM attributes the load growth driving that to what it calls Large Loads, a category built substantially of data centers.
One buildout. It makes the people who own assets wealthier and it makes the fixed cost of keeping the lights on higher for everyone, and a capacity charge does not care what your income is. The same few dollars a month is a rounding error in Potomac and an argument about the thermostat in Cumberland.
Nobody designed that. It assembled itself out of a hundred separate decisions, each one defensible in the room where it was made, and not a single person had to be corrupt for it to work exactly this way.
Which is worse, honestly. A conspiracy you can indict. This just exists.
What it looks like from here
Maryland is where the two economies are easiest to see at once, because the state ran the experiment.
Between January 2025 and March 2026, roughly 29,700 federal jobs disappeared in Maryland, about fifteen percent of the state’s federal workforce, with 9,700 going in October alone, the steepest single month in decades. The state’s unemployment rate went from 3.6 percent in February 2025 to 4.3 percent a year later.
And then the aggregates healed. In July, Maryland added 11,700 jobs and was the only state in the country to post a payroll gain that month. The Maryland Department of Labor reports the state has added 29,800 nonfarm jobs through the first seven months of 2026, a 1.1 percent growth rate against 0.3 percent nationally, with unemployment back down to 4.2 percent and labor force participation at 63.8, well above the national 61.4.
By every number a governor would put in a press release — and it is a safe bet to say that he used them all — Maryland is fine. Better than fine, actually. Maryland is recovering faster than the rest of the country.
Now hold that against David Harbourt, who lost a federal job in Frederick County and now spends half of every month on the road, moving contract to contract out of motel rooms while his wife and two kids stay home in New Market. He is employed. He counts as employed. The recovery statistics include him.
That is the whole pattern again, at state scale, and this time the recovery is real rather than a measurement artifact. The jobs came back. They came back in health care and food service and professional services, and they came back to a different set of people than the ones who lost them, and there is no line in any BLS release for a household that traded a federal career for a per-diem contract and a highway.
Why nothing happens
So why has none of this produced a political response? I promise this is not a setup for a bad joke or a sizzling hot take.
Because a crisis needs a date. In 2008 there was a Monday when Lehman failed and a Tuesday when your neighbor’s house went up for auction and a Wednesday when Congress held a hearing about it. The harm was legible, simultaneous, and shared, and a legible simultaneous shared harm is the raw material every political response in American history has been built from.
This is diffuse. It arrives as an application that goes unanswered, a lease renewal, a premium increase, a utility bill that creeps, a kid who moves back home. Each of those is a private event. None of them lands on the same week as anyone else’s. There is no hearing, because there is no incident. There is no villain, because the mechanism is distributed across a regulatory docket, an investment cycle, a load forecast, and a hiring freeze, none of which is a person.
What that produces is not organized demand. It is generalized rage with nowhere to go, and every political operation in this country has figured out that harvesting it is cheaper than fixing it. That is the business model now. Find the anger, rent it for a cycle, hand it back unspent. It is happening on both sides of the aisle, from the people who disgust you to the people you adore.
To be blunt, I do not have a complete answer. I have three places to start, and all three are about forcing the invisible onto a schedule where somebody has to answer for it.
Make the capacity market answer for who it is billing. PJM publishes the uncapped price in an appendix and no state regulator is required to do anything about it. Maryland’s Public Service Commission should be required to report annually on what share of capacity cost growth is attributable to large-load interconnection, in public, before the rates take effect rather than after.
Make employers count the jobs they advertise. Maryland requires salary disclosure in postings and requires nothing about whether the position exists. A posting that is never filled is not a job opening, and right now there is no law anywhere in this country that makes anyone count them.
And stop letting a national average stand in for a regional collapse. When a state loses fifteen percent of its federal workforce in fourteen months, that should trigger something automatic — extended benefits, retraining money, a hearing — not a press release about how the private sector added jobs somewhere else.
None of that fixes anything. But all of it puts a date on it. The date is what we need.
The line
I have been on both sides of this.
In 2010 I was 22, the economy was visibly on fire, and my degree was worth more than it had ever been worth. Everybody in my graduating class knew it was bad, but almost all of us had jobs when we walked across the stage. That was the strange mercy of it.
Now I am 38, sending out applications again, in an economy the data describes as fine, holding a credential and professional experience that no longer do what they did.
And I know what you are thinking, because it is the right thing to think. Almost nobody went to jail for 2008. They should have. Tens of millions of people lost jobs and homes and in some cases nearly everything, and the number of American banking executives who saw the inside of a cell for it rounds to zero.
You have every right to be angry about that. I still am.
But that is the point, as twisted as it may be. There was a name to put on a subpoena. Yet accountability failed, and the failure was itself the argument for the next ten years — it built movements, it moved legislation, and it is still being argued about. This time there is nothing to subpoena. The harm has yet to resolved into an event, so we will yet again never produce a defendant regardless of what bank is to blame.
Just more of the same arguments. Likely continuing until the next crisis materializes.
The line crossed in 2018 while people argued about whether the economy was good. It crossed quietly, in a Federal Reserve dataset, before the pandemic and before anybody had heard of a large language model. It is still crossed. And the twenty-two-year-old sending out their hundredth application this month is going to be told the numbers look fine.
They do look fine. That is the problem.
A longer piece on the PJM capacity auctions, and what happens when the price ceiling expires in December, is coming later this year.
Sources
Federal economic data
Bureau of Labor Statistics, The Employment Situation, July 2026. Source for the 4.1 percent unemployment rate, the 23,000 decline in nonfarm payrolls, the participation and employment-population declines since January, and the part-time and marginally attached figures.
Bureau of Labor Statistics, Real Earnings, July 2026. Source for the 0.2 percent decline in real average hourly earnings, the 0.1 percent decline for production and nonsupervisory employees, and the 0.1 percent rise in real weekly earnings driven by a longer workweek.
Bureau of Labor Statistics, The Employment Situation, March 2011. Source for the 8.8 percent national rate in the month of the graduate peak.
Bureau of Labor Statistics, “Great Recession, great recovery? Trends from the Current Population Survey,” Monthly Labor Review (2018). Source for the 10.0 percent October 2009 peak, the fifteen million unemployed, and the 19.5 percent youth rate in April 2010.
Bureau of Labor Statistics, State Employment and Unemployment, July 2026.
U.S. Census Bureau, Income in the United States: 2024 (P60-286). Source for median household income of $83,730. The 2025 figure is scheduled for release in September 2026 and should be substituted if it lands before publication.
Federal Reserve Bank of New York, The Labor Market for Recent College Graduates. Source for the 5.6 percent graduate unemployment rate and the 42 percent underemployment rate, both as of the second quarter of 2026.
Federal Reserve Bank of New York, January 2016 announcement of the college labor market interactive. Source for the 7.1 percent March 2011 graduate unemployment peak.
Federal Reserve Bank of New York, Quarterly Report on Household Debt and Credit, Q1 2026.
Board of Governors of the Federal Reserve System, Consumer Credit G.19, released August 7, 2026. Both credit card rate series appear in the same Terms of Credit table.
Liberty Street Economics, “Explaining the K-Shaped Economy: What’s Behind the Divide?” (May 1, 2026).
Federal Reserve Bank of Minneapolis, “Have U.S. consumers gone K-shaped? A review of the data” (2026). Cited for the methodological objection, not for its conclusion.
Research and analysis
Harvard Joint Center for Housing Studies, America’s Rental Housing 2026.
KFF, 2025 Employer Health Benefits Survey. The 2027 marketplace rate-request figure comes from a Peterson-KFF analysis covering sixteen states and the District of Columbia.
Economic Policy Institute, Class of 2026 labor market analysis.
National Association of Home Builders, February 18, 2026 release on affordability and demographic shifts.
Goldman Sachs Research, global AI investment forecast (2026).
Epoch AI, data center share of U.S. GDP. Measures a narrower category than the Goldman figure and the two should not be added together.
USAFacts, wage and inflation tracker, updated August 21, 2026, derived from BLS series.
Grid and energy
PJM Interconnection, 2028/2029 Base Residual Auction Report (July 14, 2026). Source for the three uncapped simulation prices and for the attribution of forecast load growth to Large Loads.
Press
Bloomberg, “Young US College Graduates Face Tougher Job Market Than Average” (July 30, 2022), reporting New York Fed data showing the reversal in 2018 and monthly persistence since early 2021.
Capital News Service, “A year later, laid-off federal workers are still piecing their lives back together,” carried by Maryland Matters and others (June 2026). Source for the 29,700 figure, the fifteen percent share, the October drop, the February-to-February unemployment change, and David Harbourt’s account.
Maryland Department of Labor, monthly employment releases (July 2026). Source for the July job gain and the seven-month growth rate. Note that the roughly 94,000 private, state, and local jobs figure circulating in coverage of the federal layoffs is cumulative since January 2023 and does not represent a one-year offset to the federal losses.
Fortune, reporting Pantheon Macroeconomics on non-AI equipment investment (February 2026).
