News
The AI Boom Is Riskier Than You Think
Most big decisions come with trade-offs.
A friend of mine recently faced one that every parent eventually understands: Do you let your teenager get a driver's license?
On the one hand, congratulations! You no longer have to spend half your life driving somebody to school activities, a part-time job and whatever social engagement apparently became absolutely essential 15 minutes ago.
On the other hand, now you're insuring a teenage driver. Maybe buying another vehicle. And presumably spending at least a few sleepless nights wondering where your child is.
Neither choice is necessarily bad. Neither guarantees disaster. But there's no option where every downside conveniently disappears.
Something similar is happening with artificial intelligence.
I've written before about just how important AI investment has become to the U.S. economy:
(I'm not going to repeat that argument here.)
Instead, I want to look at the next question:
What happens if all these enormous expectations surrounding AI turn out to be right? And what happens if they're wrong?
Because, surprisingly, there are risks on both sides.
AI doesn't have to collapse to create problems
Former Treasury Secretary Robert Rubin raised an interesting concern this week.
Rubin served as Treasury secretary during the technology boom of the late 1990s, so he's seen something like this movie before. He's not arguing that AI is a fraud or predicting that the boom is about to collapse.
In fact, Rubin thinks AI could produce major productivity gains.
His concern is what he calls “circularity risk.” The term sounds complicated, but thebasic idea really isn't.
It works like this:
- Company A promises Company B enormous future business
- So B expands its facilities and hires staff based on that expected business
- B signs a contract with its supplier, Company C
- So C hires workers and invests in new manufacturing lines because it expects B's orders to start flooding in
- In fact, C might subcontract some of the work to Company D…
Everything works beautifully… As long as everybody’s plans work.
That's where circularity becomes dangerous.
If one company can't fulfill a major commitment, the problem doesn’t stay with that company. Someone else may have borrowed money, built a facility, hired employees or ordered equipment based on the assumption that the promised business would arrive.
Multiply that relationship across an entire industry and suddenly a disappointment in one place can create problems that ripple into the far corners of the economy
That's especially important because the AI buildout is getting very, very large.
Reuters recently reported on research from Columbia Business School professor Stijn Van Nieuwerburgh estimating that AI infrastructure could require more than $10 trillion through 2032 – roughly 3.6% of U.S. GDP annually.
And as the buildout has expanded, companies have increasingly turned to outside financing to fund it.
To be clear, Van Nieuwerburgh does not say financial distress is imminent. Strong AI adoption and continued technological improvements could justify much of today's expansion.
But he does identify a troubling combination: enormous scale, uncertain future demand, complicated financing, rapid technological change and significant borrowing.
His conclusion is worth taking seriously: if expectations change, there is “meaningful downside risk.”
That's really the issue.
If expected demand, revenues or productivity gains fall short, the consequences don't necessarily stop with one unsuccessful data center or one failed tech company.
Suppliers ordered equipment. Builders started construction. Utilities expanded generating capacity. Lenders supplied financing. Communities approved projects based on expected jobs and economic activity.
Increasingly, they're all making plans around some version of the same assumption:
AI demand will keep growing fast enough to justify everything we're building for it today.
What happens if that assumption is wrong?
We've seen the danger of shared assumptions before.
There's an important lesson from 2008
I'm not saying AI is another subprime mortgage crisis. The industries, financing and circumstances are obviously different.
But there is a useful lesson from the Great Financial Crisis.
In June 2008, then-Federal Reserve Chairman Ben Bernanke explained how rapidly rising home prices had become a basic premise underlying parts of the housing boom.
Many borrowers and lenders assumed that rising home values would create enough equity for borrowers to refinance.
Then home prices stopped cooperating. When that premise failed, families and companies found they couldn’t afford to pay for what they’d bought. Mortgage defaults rose. Banks saw this and tightened lending standards. Credit became harder to obtain, which further weakened the housing market.
And, as Bernanke put it, losses that began in subprime mortgages helped “trigger the end of the broader credit boom.”
That's the important historical lesson.
Not that some homeowners couldn't make their mortgage payments.
The same crucial assumption about home prices had worked its way into decisions made throughout the wider economy. Whole businesses grew up around it! Media franchises like Flip That House. Furniture manufacturers, roofing companies – homebuilders, too.
Once the assumption failed, the consequences traveled much further than anyone expected.
I don’t mean AI is “housing in 2008.” But the principle is worth remembering:
When enough people build their plans around the same economic expectation, their mistake can become everyone's problem.
So that's one risk.
AI disappoints.
Demand comes in lower than expected. Revenue doesn't justify the infrastructure. Some projects fail. Companies pull back.
But here's where this story gets interesting...
What if AI succeeds?
Suppose the optimists are right.
Suppose AI really does transform productivity across the economy. All those data centers are necessary. Businesses discover ways to automate work faster and more effectively than we currently imagine.
Great!
But that creates another set of problems.
Rubin himself pointed to the possibility of substantial worker displacement, particularly among knowledge workers.And JPMorgan Chase CEO Jamie Dimon has raised essentially the same concern from another direction.
At the World Economic Forum earlier this year, Dimon described a hypothetical future in which self-driving technology replaces large numbers of truck drivers. His example was deliberately extreme: Imagine 2 million workers suddenly jobs paying $150,000 a year – to minimum wage work.
(Those figures were a thought experiment, not a claim about typical truck-driver wages – I checked his numbers, and most truckers make about $60,000 a year although owner-operators make up to $180,000.)
Dimon's larger point wasn’t about the specific dollar amounts, but about the speed of the disruption.
Technological progress can make the economy more productive while simultaneously creating severe hardship for the workers and communities forced to adjust.
That's not an argument against progress. It's a reminder that an economy isn't just an efficiency spreadsheet.
Families have mortgages and grocery bills. Communities form and grow around particular industries. Careers based on experience that took decades to build can’t be replaced with a six-week retraining course.
If AI changes work gradually, most of us will have time to adapt.
If it changes millions of jobs quickly? The adjustment becomes a crisis – not for everyone. Not at the same time. But a crisis nonetheless.
And jobs aren't the only challenge created by an AI success story…
AI's hunger for electricity is already insatiable
Making a request to an AI bot is effortless. Type something into a box, press Enter, get an answer – the way Google used to work.
Behind that simple app on the screen are enormous physical facilities filled with thousands of computers running on megawatts of very real electricity.
The Energy Information Administration now expects U.S. electricity consumption to reach record highs in both 2026 and 2027. Reuters reports that AI-hungry data centers, along with broader electrification, are helping drive that growth.
And the potential growth from here is astonishing.
According to the Electric Power Research Institute, U.S. data centers consumed roughly 177-192 terawatt-hours of electricity in 2024.
By 2030? Somewhere between 383 and 793 terawatt-hours. At the high end, that's more than four times 2024 consumption.
Utilities, regulators and data-center operators are already looking for ways to manage that demand without requiring quite as much new generation and grid infrastructure.
One proposed solution is “demand response” – temporarily reducing or shifting data-center electricity usage when the grid is under the most stress.
That's encouraging. But it also tells us something important.
The effects of AI reach beyond the technology sector. They're showing up in power-generation plans, utility infrastructure, land use, construction decisions and employment.
That's what happens when a technology becomes economically significant enough.
Its growing pains become everybody's growing pains.
You don't have to predict the future to prepare for uncertainty
So which future do I think we're heading toward?
Will AI disappoint, or set off a productivity revolution?
Will we discover that we've built too many data centers, or will we wonder how we ever lived without them?
I don't know.
And that's not a cop-out. Nobody knows!
Robert Rubin doesn't know either. Neither do the CEOs spending billions on AI infrastructure, the economists trying to measure its productivity benefits or the utilities attempting to forecast how much electricity all of this will require.
That's exactly the point.
AI success could generate enormous benefits, but not equally for all of us.
But success and failure produce different risks, and the sheer scale of the transformation means those risks increasingly reach beyond the technology industry itself.
That's a useful lesson for your own financial planning.
You don't have to correctly predict the future in order to prepare for uncertainty. In fact, that's one of the main reasons diversification exists in the first place.
When your retirement savings depend too heavily on any single economic assumption continuing to work, you're making your own version of that same circular bet.
Physical precious metals offer one way to diversify savings into tangible assets whose value isn't dependent on whether today's AI expectations ultimately prove too pessimistic, too optimistic or exactly right. If you'd like to learn more about diversifying retirement savings with physical precious metals, request your free Precious Metals Information Kit.
And if you've already done your research and would like to speak with Birch Gold about opening an account, call us at (800) 355-2116.
BGG - About the Author 
About Peter Reagan
Peter Reagan has covered precious metals and the economy for Birch Gold Group for over 15 years. His articles have been featured in several leading publications, including Newsmax and Zerohedge. At Birch Gold Group, Peter leverages his deep market insights to help educate website visitors and customers on why they should consider diversifying their savings into gold and other precious metals, and how that process works. Due to this commitment to education, his content has become a trusted source of information throughout the industry.
BGG - CTA Option 2 