THE GOLF COURSE OFFSET

Data centers use a lot of water. Don’t hide it or minimize it. But put it in context instead.

In 2023, U.S. data centers directly consumed roughly 17 billion gallons of water for cooling. Lawrence Berkeley National Laboratory projects that could double, maybe quadruple, by 2028.

Now compare that with golf. America’s roughly 14,000 golf facilities applied 531 billion gallons of water in 2024. That’s thirty times the direct data center figure.

I’ll do the math before someone does it for me. That 17 billion gallons only counts direct consumption at the data center. Generating the electricity those same data centers used carried an additional water footprint of roughly 211 billion gallons, according to the same LBNL analysis. Put the two together and you’re closer to 228 billion gallons. Golf is still higher, but now we’re talking about roughly twice the water, not thirty times.

There is another distinction. Golf reports water applied while LBNL estimates water consumed. Those aren’t identical measures. I’m using them here to establish scale, not pretend they’re interchangeable.

That’s the orientation I think has been missing from this debate.

A DEBATE WORTH HAVING, AND ONE WORTH WATCHING

Communities have legitimate questions about data centers: water, power, noise, land use and who pays for the new substation. Developers should have to answer them.

But some of the noise may be disingenuous. OpenAI disclosed in June 2026 that it disrupted a likely China-linked operation using ChatGPT to generate anti-data-center content, prompted in Simplified Chinese and posted by accounts posing as Americans. X identified a separate network two months later. A Clemson researcher who examined some of the accounts found almost no engagement before they were shut down.

China didn’t invent American skepticism about data centers. There is plenty of homegrown skepticism without any help from Beijing. But foreign actors including China are trying to exploit it. We ought to know when that’s happening.

The skepticism I run into more often isn’t foreign and doesn’t come from a bot farm. I still get invited into university classrooms now and then, and I hear it straight from the room. Smart kids, well-intentioned, absolutely certain AI is either going to take every job or boil the planet, and often unfamiliar with what’s actually inside the buildings they’re protesting.

I don’t think they’re stupid. I think they’re doing what people have done around disruptive technology for a very long time. The actual Luddites smashed the actual looms. New technology changes people’s livelihoods, and people have every right to ask hard questions about it.

But fear isn’t a substitute for understanding scale, tradeoffs and what we get in return.

WHO ACTUALLY BENEFITS?

One criticism I hear is that data centers consume community resources to benefit wealthy technology companies and relatively few people.

I’m not neutral on golf. As a kid, I caddied on the PGA Tour a couple times at the Jackie Gleason Pro-Am. I played golf in high school. I still get out when I can, and I love the game. Nobody can reasonably accuse me of having it in for golf courses.

Even with that history, I use and depend on data and AI far more than I use a golf course. I’d bet that’s true for most people reading this.

The National Golf Foundation counted 29.1 million Americans playing on a course in 2025, fewer than one in ten. Add ranges, simulators and Topgolf and the number reaches 48.1 million. The industry has made real progress opening the game to more people, but golf still requires a course, equipment, time and usually money.

Compare that with the infrastructure behind a phone or laptop. Students use it. Mechanics run diagnostics on it. Nurses pull medical records through it. Farmers run precision agriculture with it. Small businesses run payroll, inventory and logistics on it. Forty percent of U.S. employees reported using AI at work in 2025, twice the share from two years earlier.

A data center is infrastructure sitting underneath millions of products, businesses and jobs.

THE JOBS ARE CHANGING

There is another claim I hear from college students: AI is going to eliminate work.

Maybe some work. Certainly some tasks. But that’s different from eliminating jobs across the economy.

Economists Pascal Michaillat and Emmanuel Saez studied U.S. unemployment and job vacancies going back to 1930. Their estimate for the average American full-employment unemployment rate is 4.1 percent. Full employment has never meant zero unemployment. People quit, move, retrain, enter the workforce, change careers and spend time between jobs.

The latest unemployment rate is 4.1 percent. So effectively Americans are fully employed.

I’m not hanging an argument about the future of work on one month’s employment report. I’m making a simpler observation. We are watching the division of labor between people and machines change while the country remains at roughly what economists have calculated as full employment.

Software is a good example because we’re already watching it happen.

A company may no longer need a room full of programmers writing every line of code. One or two very good software engineers working with AI can increasingly accomplish what required a much larger team only a few years ago.

The human doesn’t become less important. Different human skills start commanding the premium.

The valuable engineer increasingly needs to understand the problem, architecture, customer, schedule and mission. He or she needs to know what to ask the machine to do, recognize when it gets something wrong, integrate the pieces and deliver a working product. Project management, judgment, strategy and the ability to work with machines become more valuable.

We’re already seeing that shift in hiring. Software development job postings have risen since early 2025 even as overall postings declined, with most of that growth coming from senior positions and a substantial share from jobs explicitly involving AI. BLS still projects roughly 268,000 additional software developer jobs over the coming decade, along with strong growth in data science and cybersecurity.

Some occupations will shrink. Some jobs will disappear. Others we haven’t named yet will emerge.

BUT BE HONEST ABOUT WHO’S PAYING FOR IT

National averages hide things. This one hides something I don’t want to slide past, especially since I’ve now twice pointed at college students.

The hiring rebound in software is disproportionately benefitting people who already have careers. That’s the same statistic I just cited. Most of the growth came from senior positions. Research using payroll records rather than job postings also shows younger developers have taken a much harder hit than their older counterparts. Recent computer science graduates have faced unemployment above the overall national rate.

So AI isn’t eliminating work. But in software, it may be collapsing the traditional entry point.

A twenty-three-year-old holding a fresh degree and no offers has every reason to be angry about that. We told them to go get that degree.

We also know how to respond.

Universities can’t keep producing graduates optimized for a job that’s disappearing: writing code line by line from a specification somebody else wrote. Nobody would defend a university still granting horse-and-buggy degrees while Detroit was tooling up. We’d expect the curriculum to change, and quickly.

Teach students to direct AI tools and defend the output. Make them own architecture decisions and live with the consequences. Teach them to read code critically because reviewing what a machine produced is increasingly part of the job. And put far more of their education inside actual workplaces. More experiential, less theoretical.

Then deal with the graduates already caught in the transition.

Nobody leaves Air Force pilot training combat ready. You earn a set of wings, then go to a formal training unit for a specific airframe, then mission qualification, then upgrades after that. Every stage is hands-on and ends with somebody certifying that you can do a specific job.

We never expected the initial credential to be the finished product.

We shouldn’t expect that from a computer science degree either.

Build paid six-to-twelve-month apprenticeships with employers and universities. Put young graduates with good fundamentals on real work under experienced engineers. Let them leave with demonstrated skills against jobs that actually exist, not another academic certificate.

I paid for engineering school welding rototiller frames on third shift at Troy-Bilt. I’ve never believed the classroom is where competence gets built.

The classroom builds the foundation. The shop floor builds the engineer.

The Labor Department has been developing apprenticeship frameworks for AI-related work, and companies are experimenting with earn-while-you-learn programs. We don’t need to invent the idea. We need more of it.

WE’VE DONE THIS BEFORE

A hundred years ago, widespread electrification was reorganizing American industry. Electric motors didn’t merely replace steam power. They changed where machines could sit, how factories were laid out, how production flowed and what a worker could accomplish in an hour. Entire industries grew around applications nobody could have predicted when the first power stations went online.

I’m glad our answer wasn’t to stop building the electrical grid until somebody could guarantee that every existing job would survive.

AI is forcing another division of labor, this time between humans and machines. The winning skill is knowing how to use the machine to do something useful, not trying to outrun it at whatever it does best.

Some of those workers will have PhDs. Others will be electricians, technicians, machinists and operators. What they will increasingly have in common is access to enormous amounts of computation.

And that computation has to live somewhere.

THE PART ABOUT NATIONAL SECURITY

I spent my last stretch in the Pentagon running production acceleration for munitions and standing up what became the Wartime Production Unit. Before that I spent years working with NATO’s defense production and procurement and in Arctic security forums. I watched strategic competition play out in ice and airspace. Now we’re watching another part of it play out in silicon.

I came away from that work convinced that an industrial base is a national security asset whether it makes artillery shells or trains models.

Compute is production capacity in this era much as steel mills and shipyards were production capacity in the last one.

China isn’t debating whether to build this capacity. It’s building it aggressively, with state support and tens of billions in private investment from Chinese technology companies on top of it.

And you don’t see anti-data protestors in China or college professors teaching kids to resist technological progress.

The question isn’t whether data centers will be built. It’s how much of that capacity America builds here, under our rules, employing our workforce, versus how much accumulates somewhere we don’t control.

Golf courses aren’t strategic infrastructure.

Compute is.

I say that as someone who loves golf enough to have caddied just to be near the game.

But I don’t want the next generation of airmen I used to command wondering why we lost time in that competition because we couldn’t get our own water math straight.

NOT EVERY FIGHT IS ABOUT WATER

In plenty of places, water isn’t the main objection. It may be the transmission line, the substation, the possibility of higher electric bills or a community that wasn’t consulted before the bulldozers arrived. Pricing water won’t fix those problems.

I’m talking specifically about places where water is actually the constraint, particularly the desert Southwest and river basins already under stress. In those places, solve the water problem as a water problem.

SO PRICE THE WATER

Working around airfields years ago, we deliberately drained standing water near runways because it attracted birds and created a bird strike hazard. We weren’t against water. We just didn’t want it next to a runway.

Apply the same thinking here.

If a proposed data center’s direct water consumption creates a problem in a particular watershed, require the developer to offset that consumption there. Make it gallon for gallon, verified and adjusted for season and source.

The electricity-related water footprint is different. Generation may occur hundreds of miles away in another watershed. Disclose it, measure it and address the impact where the electricity is generated rather than pretending every gallon comes from the community hosting the data center.

The basic idea isn’t new and might give us a model to use. Federal policy under Clean Water Act Section 404 has long required compensatory mitigation in certain cases when development damages wetlands and other aquatic resources. Mitigation banks and in-lieu fee programs provide ways to offset those impacts.

We can apply similar thinking to scarce water.

A developer could finance reclaimed-water infrastructure, pay for conservation improvements, convert other high-consumption uses or buy water rights from willing sellers.

And yes, it could buy a golf course.

Don’t mandate which use loses. Establish the water requirement and let people figure out the least expensive way to meet it.

THE REAL CHOICE

Scrutinize these projects. Make developers pay their actual costs. Measure water where water is scarce. Don’t quietly push private infrastructure costs onto existing ratepayers. Fix the pipeline that’s supposed to put young Americans into these new jobs.

But make the decision using the whole picture. Don’t block progress out of fear of change or the unknown.

A country that wants AI but doesn’t want data centers is making roughly the same choice as a country that wanted the Industrial Revolution without factories. The infrastructure comes with the technology. In this case, it also comes with a strategic competition against a country that has already decided to build.

Where water really is the constraint, we have options. Conserve it, reclaim it, price it or offset it.

If the numbers work, buy a golf course.

I love golf. But if we’re going to choose between uses of a scarce gallon of water, let’s at least know what we’re choosing.