In Your Backyard: Jobs, Growth, and the Race We're Actually In

Geoffrey Hinton, the Nobel laureate known as the godfather of AI, has publicly told people to "train to be a plumber." A LinkedIn report found that 55-65% of Gen Z workers across the U.S., U.K., Germany, and France now say skilled trades offer more meaning than a traditional office job. This is a real and rational response to real uncertainty, and pretending otherwise would be its own kind of misinformation.

That being said, the Pew Research poll shows that the pessimism towards AI taking jobs is real. 71% of Americans feel that there will be fewer jobs because of AI in the future, up from 64% just a year earlier.

What the Data Actually Shows

But belief and outcome are different things, and right now the hard labor-market data is not cooperating with the pessimism.

The Yale Budget Lab, using a rigorous differences-in-differences approach that compares AI-exposed occupations to similar unexposed ones, found that as of its most recent analysis, employment shares in AI-exposed jobs have moved by an amount "close to zero" and are not statistically distinguishable from no change at all. Wages showed no significant divergence either. Perhaps most tellingly, the study found no unusual increase in workers switching occupations, which is exactly what you'd expect to see first if AI were meaningfully displacing people at scale. Their conclusion: "AI is probably not (yet) the reason for labor market weakening." Note the "yet." This is a live question, not a closed one, and we will all be watching the data closely.

A companion analysis from MIT Technology Review, drawing on Bureau of Labor Statistics data, found something that runs directly counter to the popular narrative: the unemployment rate for workers in the most AI-exposed occupations is actually lower than for workers in less-exposed ones. Only about one in five companies currently use AI in any business function at all, according to Census data: a useful reminder of how far "AI is everywhere" rhetoric has outrun actual corporate adoption. And a chart making the rounds from Apollo's chief economist, Torsten Slok, makes the point vividly on a single industry regulars have been predicting AI would gut for years: U.S. payroll employment in travel agencies today sits almost exactly where it did before the pandemic. There is, in Slok's words, "still no sign of AI lowering employment in travel agencies."

There are exceptions we need to pay attention to, because a one-sided case isn't worth much. Stanford researchers found a real, 16% decline in entry-level hiring within AI-exposed fields like software development for workers aged 22 to 25, an effect that has been growing through 2025. Recent college graduate unemployment is running around 5.6%, the highest since the pandemic and, before that, the 2008 recession. That is a genuine cost being paid by a specific slice of the workforce: new graduates trying to break into a small number of white-collar fields where AI is good at the "codified knowledge" tasks a junior employee used to cut their teeth on. It is not evidence of the broad jobs apocalypse the headlines promise, but it is real, and it deserves a real policy response. By that I mean apprenticeships, different hiring pipelines, something; rather than a blanket "don't worry about it."

This is actually more concerning than it might be at first glance. It’s those entry-level jobs, where new graduates and young workers learn how to function in a business and work. They are the feedstock for future management positions and expansion. Just like an economy needs more workers every year in order to grow, businesses need young people who can be trained into more responsible positions. Yes, you can hire from outside, but then you are cannibalizing another company’s growth. Then again, free markets work.

Most of us learned the basic fundamentals of working as a young person: showing up on time, being productive, responding to instructions, learning more and contributing to a team, etc. That became the springboard for us becoming more productive. That is one reason that employers want to know about your past employment: what have you learned and how did you perform? If 16% of the workers in AI-related fields are not there anymore, that is less on-the-job training. Then again, if those jobs are not going to exist in the future, then there is no point in being trained for them. It’s a quandary that we won’t know the answer for a very long time.

There's a second, quieter data point worth your attention: Gallup finds that among workers whose employers have actually rolled out AI, engagement is running eight points higher for weekly AI users than non-users (39% versus 31%), and when management pairs AI adoption with a clear plan and active support, engagement jumps to 53%. Properly managed AI adoption looks a lot more like a tool that makes people more engaged in their work than a guillotine hanging over it.

One final and major point. All new significant technologies have ended up creating vastly more jobs than those destroyed. Artificial intelligence will be no different. It will end up creating jobs that we don’t even realize possible today. That is a long-term optimist view.

There is also a more optimistic and perhaps more realistic view. In the past, major technologies were adapted over time. We went from the farm through the Second Industrial Revolution over 8-10 generations. We had time to adapt. My concern is that we are trying to adapt to one of the most potentially disruptive technologies in less than half a generation. That gap can create a great deal of pushback and would only serve to make the negative views of data centers and artificial intelligence increase.

The Jobs Data Centers Actually Create

Set aside the general AI-and-jobs debate for a moment and look specifically at what happens when a data center gets built in a county, because this is the part directly relevant to the "not in my backyard" fight.

A rigorous new study covering roughly 1,500 U.S. data center facilities and 52 canceled projects which compared counties that got a facility to similar counties where a planned facility was canceled, which is about as close to a controlled experiment as economics gets, found that data-processing sector employment rises 56% over the first decade in counties that receive their first large data center, with telecommunications employment up 43% in counties that land a hyperscale facility from the likes of Amazon, Google, Microsoft, or Meta. Home prices in these counties rose a modest 2–5%. There is a caveat: wages in the broader county didn't move much, and the researchers note that industry-funded estimates of data center job impacts tend to overstate the benefit by ignoring growth trends the county was already on. I know it will be a shock, but sometimes rosy numbers from self-interested industries don’t survive actual scrutiny.

The construction side of the ledger is not in dispute. Data center construction spending hit a $59.3 billion annualized rate in May 2026, up 23% year-over-year, and now accounts for roughly 8% of all private nonresidential construction in the country, a figure "The Situation" newsletter puts even higher, noting data centers account for over 3% of all U.S. construction of every kind, and that private data center construction spending has now surpassed total public spending on all transportation combined: airports, transit, marine terminals, everything! In June, the Commerce Department reported data center outlays up almost 50% year-over-year to $68.3 billion, even as spending on every other category of private construction (housing, hospitals, schools) fell by $101.6 billion over the same period. Construction wages are rising faster than the broader private sector too: 5.0% year-over-year versus 3.4% economy-wide, with construction workers now earning over 20% more than the average private-sector worker.

At the state level, the numbers get concrete fast. A 2025 Ohio study found a single mid-sized data center project supports nearly 9,700 construction jobs during the build phase, $2.4 billion in total economic output, a $1 billion contribution to state GDP, and roughly $84 million in annual peak state and local tax revenue.

Virginia, the most data-center-dense state in the country, sees an estimated 74,000 jobs annually tied to the industry, a $9.1 billion contribution to state GDP, and up to 30% of total local tax revenue in some jurisdictions. McKinsey estimates the broader data center build-out represents a $7 trillion opportunity over the coming years for the industrial companies that supply the power, cooling, and equipment behind it. That is a supply chain far larger than the hyper-scalers themselves.

Without the massive investment in data centers, US GDP growth would be flat at best. That brings into question the real strength of the US economy, both in very positive and very problematic ways. That is a letter all on its own.

The Bigger Picture: Growth, and the Race We're Actually In

Zoom out further and the case gets stronger still. A comparison of major U.S. capital-spending waves, using data from Columbia Business School, shows the current AI buildout running at roughly 3.5–4% of GDP from 2025 to 2032. That is larger, as a share of the economy, than the canal boom of the 1830s, the railroad boom of the 1870s–1890s, rural electrification, the interstate highway system, or the telecom and fiber buildout of the late 1990s. This is, by the best available measure, the largest capex-driven boost to GDP in American history.

Mark Mills, whose primary-source analysis for the National Center for Energy Analytics I trust as much as anyone writing on this topic, makes a related point: if AI does nothing more than nudge U.S. productivity growth back to its postwar average of 2.2% a year, from the 1.4% we've been running, that alone is worth roughly $10 trillion in cumulative additional GDP over the next decade, and the energy consumed by that broader wealth creation will dwarf the energy the AI infrastructure itself consumes directly.

There's also a case here that goes beyond economics to geopolitics. China's leadership has been explicit about embracing open-weight AI models. Not entirely, I'd argue, out of generosity, but substantially because China is behind the U.S. in computing power and AI capital expenditure, and open weights let it recruit the whole world's engineering talent to close the gap for free. China's answer to a slowing, aging workforce has been to install one of every two industrial robots deployed worldwide last year. That effort is succeeding, where its attempts to boost domestic consumption have not. The American answer to the same demographic problem - U.S. population growth slowed to just 0.5% in 2025, with net migration near historic lows - has to come from somewhere, and data centers and the automation they enable are a large part of the "build robots" option, alongside more babies and more immigration, that keeps the American economy from following a stagnating Japan and, increasingly, China itself into a demographic wall. A country that decides not to build the infrastructure for this technology doesn't stop the technology from being built. It just decides someone else builds it first.

And given the critical nature of this technology, it is imperative that the US maintains its lead. This gets into geopolitics and basic defense. We are watching in real time what AI does on the battlefield.

Finally, I'd point you back to a concern raised in Part 1 that applies here too: the fear that data centers are financially fragile, debt-fueled bubbles waiting to pop, stranding capital and jobs when they burst. The primary-source view from Mark Mills, who has spent his career studying industrial infrastructure cycles, is that this fear misunderstands what a data center actually is. It is not a single-purpose asset like a railroad to a ghost town; it is closer to a warehouse or a factory floor, built with extra power and cooling capacity precisely so new chip generations can be slotted into it as they arrive, while the older racks don't wear out the way mechanical equipment does and keep generating income producing cheaper tokens for years afterward. That is a very different risk profile than the popular “AI bubble” narrative assumes, and it's one more reason the jobs and the capital tied up in this build-out are more durable than the skeptics believe.

Now, do I believe every dollar invested in a data center is going to pay its developers back? Absolutely not. Some of these companies are only going to get a fraction of what they invested back. But we don’t know which company at this time. We have watched the leadership change through various companies, and there are literally hundreds of companies working in the ecosystem, any of which could rise up and become the next giant killer.

But for the big companies, getting AI right is existential. It is an arms race. Right now, every data center built and turned on is immediately put to use. The market is demanding more data centers - more computer power - not less. I don’t doubt that some hyper-scalers are overly optimistic about the future and their return on investment potential. It’s just that nobody knows which one.

In Summary:

Across both letters, the pattern is consistent. The loudest objections to data centers - water, power, jobs - have a kernel of legitimate concern buried inside a much larger amount of exaggeration, and in every case the trend line is moving in the industry's favor: cooling technology cutting water use, grid rules adapting to speed up connections, nuclear power finding a buyer of last resort, and labor markets that, so far, aren't showing the damage the headlines promise.

The main answer to the physical side of the data centers is siting. You shouldn’t build a data center next to a housing development. The water issues must be resolved before a piece of dirt is turned. Hyper-scalers must present a viable energy plan that ensures that consumers are not stuck with a higher electric bill because of them. All of those are rational and doable.

What could be a much bigger problem is the transition for younger workers. People are working on it, but I have not seen a grand plan that really addresses the issues.

But "not in my backyard" is, on the evidence, the wrong instinct. The country that builds this infrastructure fastest is the country that captures the growth, the jobs, and the geopolitical position that comes with it. I'd rather the US be that country.

Austin, New York, DC, and Cleveland

First, let me say thanks to all of those who expressed concern about my daughter Abbi. Mike Roizen was able to get us into the head of neurosurgery at the Cleveland Clinic. Her brain tumor is in the middle of her brain, making it very difficult to get to. Thankfully it is 90% likely that it is benign but it must be addressed. She is scheduled for brain surgery November 5^th. The technology they are using is nothing short of amazing: intense MRI during the surgery, AI and robotics guiding the surgeon. It literally says take out this cell but don’t touch that cell. There are only a handful of places in the world that can do this.

Ironically, it was only because she was misdiagnosed when she had a seizure that they discovered the brain tumor. If they had correctly diagnosed it as a heart issue the tumor would still be growing and when there were finally symptoms it would’ve been extraordinarily problematic.

I will be in Austin Sunday through Tuesday. I am there primarily to attend the funeral ceremonies for Patrick Watson’s wife, Grace. Sunday morning, I will have a late brunch with Lacy Hunt, George and Meredith Friedman, Brad Rotter and a few friends. Should be a great time. That afternoon I will go to Joe Lonsdale’s where he is hosting a barbecue party.

I will be in New York the second week of September, DC the second week of November and other trips are being scheduled. I was actually only out of Puerto Rico for 31 days last year (not counting flying days as the IRS doesn’t count flying days as not in Puerto Rico days, which is partly how they figure your taxes.)

Briefly, let me give you an update on Lifespan Edge, our longevity clinics. We have officially opened clinics in West Palm Beach and Columbia, Maryland (essentially DC). Dallas and Dorado Beach are open as well. We are booking patients in all of these locations. We are in significant discussions with well over a few dozen doctors and locations throughout the country to expand our services.

The reason for the interest? More and more, longevity experts are telling their clients the first part of their journey should begin with Therapeutic Plasma Exchange. TPE has profound effects on inflammation, muscle loss as we get older, Alzheimer’s and dementia and more.

My partner in this endeavor is Dr. Mike Roizen, one of the truly leading longevity experts in the world. You can learn more at Lifespan-Edge.com. I will recommend reading our main research paper.

The research and other information (plus our fabulous documentary on the website) can make a real difference in your life. You can set up a discovery call to talk with our doctors about the procedure, as well as look at a lot more research.

And with that, I will hit the send button. You have a great week and make sure to enjoy friends and family. All the best…

Your dealing with family issues analyst,

John Mauldin

A message from Advisor Perspectives and VettaFi: Discover something new! Click here to register for our upcoming webcasts.

© Mauldin Economics

Read more commentaries by Mauldin Economics