The $10 Trillion AI Infrastructure Gamble: Who Wins, Who Loses?
AI could transform civilization. It could also destroy a few balance sheets along the way.
Wall Street has discovered a remarkable business model: Spend trillions today, promise extraordinary profits tomorrow, and hope nobody asks too many questions about what happens in between.
Remember The Big Short? Some of the investors associated with that story are now betting against companies helping finance the $10 trillion AI boom. Their concern isn’t necessarily that artificial intelligence will fail. It’s that AI could succeed spectacularly while some of the companies building it lose billions.
Welcome to the artificial intelligence revolution, where Wall Street is betting trillions on the future, and some investors who have seen this movie before are betting against the companies financing it.
Data Center Campuses
The world’s largest technology companies are racing to build sprawling data centers, secure electricity and purchase specialized chips.
Consider the scale: A single large AI data center can consume as much electricity as 100,000 American homes. Some proposed mega-campuses could eventually require enough power for 2 million households – approaching the combined number of households across Miami-Dade, Broward and Palm Beach counties in South Florida. That’s an extraordinary amount of infrastructure to finance before the long-term profits are assured. (Source: International Energy Agency)
Meanwhile, startups are borrowing against computing equipment that may age faster than the loans used to purchase it. More than 1,000 American communities have reportedly sought restrictions on data-center development. (Source: Financial Times)
Here’s the paradox: AI could become one of the most useful technologies in modern history while leaving a surprising number of its early financiers disappointed. History has seen this movie before.
History of Industrial Revolutions
Railroads transformed America but bankrupted shareholders. Fiber-optic networks helped build the internet, although many companies that financed them didn’t survive the dot-com collapse. The technology succeeded. Some of the investors didn’t.
Now the stakes are even larger. Research by Columbia Business School economist Stijn Van Nieuwerburgh, published by Brookings, estimates roughly $10.3 trillion in U.S. AI infrastructure investment from 2025 through 2032. That averages about 3.6% of annual GDP, compared with approximately 2.2% for the railroad expansion and 1.1% for the telecom-and-fiber boom. (Brookings Institution, Stijn Van Nieuwerburgh)
Industrial revolutions: What would their buildouts cost in today’s economy?
Using 2026 U.S. GDP of approximately $32.6 trillion, the chart below puts history’s great infrastructure booms into today’s economic perspective. (GDP-equivalent estimates; Brookings and U.S. BEA.)
AI’s projected $10 trillion buildout sounds staggering, but consider the competition. Measured in today’s GDP-equivalent dollars, America’s railroad expansion of the late 1800s would represent roughly $15 trillion, interstate highways of the 1950s–1970s nearly $7 trillion, and the telecom-and-fiber boom of the 1990s–2000s approximately $3 trillion.
The difference isn’t just how much we’re spending, but how quickly. AI’s projected annual investment represents 3.6% of GDP, compared with 2.2% for railroads and 1.1% for telecom. By that measure, AI is the most intense infrastructure boom of the group.
There’s another important distinction. During the 1990s telecom boom, companies laid enormous amounts of fiber-optic cable betting that internet demand would eventually catch up. Much of that capacity initially went unused, contributing to a wave of bankruptcies. The fiber wasn’t obsolete; the business models and financing assumptions were.
AI presents almost the opposite challenge: Companies are racing to build enough computing capacity to keep up with extraordinary demand.
But history offers a warning. Demand can be real, infrastructure can be essential, and investors can still lose billions if they overbuild, overborrow or overpay.
AI Arms Race: The Most Expensive Race Nobody Can Afford to Lose
Imagine you run a major tech company. Your competitors are spending billions to control the next generation of computing. Spend too little and risk losing customers and relevance. Spend too much and you could end up with costly buildings full of yesterday’s chips.
What do you do? Apparently, you spend. That’s the uncomfortable logic behind the AI arms race. Each company may be acting rationally, even if their collective spending creates more capacity than the market can profitably absorb.
And the physical world has a way of interrupting digital dreams. The Financial Times, citing RegLayer, reported at least 47 major U.S. data-center projects delayed, withdrawn or blocked in 2026. Bernstein estimated that only 35% of the planned U.S. data-center pipeline was viable. (Financial Times, Oct. 19, 2026.)
Separately, FT research found more than 1,000 communities had sought restrictions on AI infrastructure development. AI may be software on your phone, but to your neighbors it means electricity demand, water consumption, transmission lines and industrial facilities. A project can win Wall Street’s applause long before it wins a zoning vote.
These bottlenecks aren’t uniformly bad for business. Delayed competitors may strengthen existing operators’ pricing power. Utilities may welcome grid investment while customers question who pays. Chip suppliers may celebrate orders even as developers struggle to finance them.
The lesson for investors: Announced spending is not completed construction, and completed construction is not necessarily profitable capacity. Those are three separate milestones.
The Debt Behind the Digital Gold Rush
The AI boom began with an enormous advantage: Some of its biggest builders were also among the richest corporations in history. Profits from advertising, cloud computing and software helped finance new infrastructure.
But the capital demands have grown so large that the financing by the mega-cap tech companies is spreading into bonds, private credit, joint ventures, leases and specialized data-center companies.
S&P Global recently warned that the true extent of AI-linked obligations may be harder to see than conventional debt figures suggest. That is not the same as predicting widespread defaults; it is a warning to look beyond the headline balance sheet. (S&P Global, reported by Reuters, Oct. 2026.)
FIGURE 2. AI financing moves toward debt | Source: Texas A&M Real Estate Research Center; reported figures.
Neoclouds: The Middle Layer of the Ai Expansion
Behind the AI revolution is a new breed of company called a neocloud. Think of it as a landlord for AI computing power. Companies like CoreWeave own or lease data-center space, install expensive Nvidia chips, and rent that capacity to technology giants such as Microsoft, Meta and OpenAI.
CoreWeave began as a cryptocurrency-mining company before pivoting to AI. Rather than mining Bitcoin, it now rents computing power to customers. It’s a capital-intensive infrastructure business dressed up as a cloud technology company.
Consider CoreWeave, with approximately $50 billion in reported debt and financing obligations against a $32 billion stock market valuation. Recent Financial Times reporting described one of its bonds trading in distressed territory, alongside financing setbacks affecting other AI infrastructure developers. (Sources: CoreWeave financial disclosures; Financial Times. Financing obligations are not the same as outstanding debt.)
The concerns have even attracted investors associated with The Big Short. Seawolf Capital, founded by traders featured in the book and film, reportedly took a bearish position in CoreWeave. Their concern isn’t that AI will fail. It’s that borrowing billions to lease rapidly depreciating computer chips may prove a difficult business model, even when demand is booming. (Source: Business Insider, October 27, 2026.)
The challenge? Data-center buildings may last 30 years, while the GPUs generating revenue could become outdated in just three to five years. Even functioning chips can lose value as newer models deliver more computing power at lower costs.
That creates three overlapping risks: rising borrowing costs, depreciating technology and continuous reinvestment. A company can have enormous customer demand and still struggle to generate enough cash flow to justify its investment.
Yet this isn’t necessarily another dot-com bubble. Today’s largest AI customers include some of the world’s most profitable companies. The risks facing a heavily indebted neocloud are fundamentally different from those facing Microsoft, Meta or Alphabet.
Invesco recently cited a Gartner forecast suggesting that more than 80% of neocloud providers could cease operating independently by 2030, through consolidation or failure. That’s not a prediction that 80% will go bankrupt, but it underscores how dramatically the industry could change. (Source: Invesco, Brian Levitt, October 26, 2026, citing Gartner.)
The greatest risk may not be that AI fails. It’s that AI succeeds spectacularly while some of the companies borrowing billions to build its infrastructure fail to make money.
Credit Markets: A Warning, Not a Crisis
The bond market offers a useful reality check. While financing stress has emerged at CoreWeave, broader investment-grade and high-yield credit spreads have remained relatively contained, according to Invesco’s October 26 commentary.
High-yield spreads recently stood near 3.4%, below their 4% historical average. In other words, lenders are becoming more cautious about certain AI borrowers without indiscriminately abandoning corporate credit. (Clearonomics/Bloomberg high-yield chart, Oct. 23, 2026.)
That’s an important distinction. Financial trouble among heavily leveraged AI infrastructure companies doesn’t necessarily signal a broader credit crisis. But widening spreads, higher Treasury yields and mounting refinancing costs deserve attention.
The question isn’t simply whether AI demand will keep growing. It’s whether companies can generate sufficient cash flow to support the enormous debt financing its expansion.
The $10 Trillion Question: Where Are the Profits?
Jon here. Here’s the statistic I find hardest to ignore. J.P. Morgan Asset Management found that roughly two-thirds of S&P 500 companies discuss AI on earnings calls. About 17% describe specific benefits. Only 2% quantify an earnings impact.
FIGURE 3. AI adoption versus quantified returns | Source: J.P. Morgan Asset Management; rounded figures.
That doesn’t mean the other 98% are getting nothing from AI. It means the bridge from excitement to measurable profits is still under construction, even as spending races ahead.
Consider programmers completing projects 30% faster. Does management cut costs, develop more software or tackle a backlog? All are useful, but not all immediately boost earnings. American Express has described using productivity gains to address development needs, while Netflix has discussed reinvesting efficiencies into content.
Here’s the catch: Wall Street often values AI infrastructure companies as though customers will pay almost anything for computing power. But customers want precisely the opposite: More computing power at lower prices. As technology improves and competition intensifies, pricing pressure could undermine the returns investors expect.
During a gold rush, investors focus on miners and sellers of picks and shovels. But what if the lasting winners are hospitals, insurers, manufacturers and other businesses using AI without financing the entire technological frontier?
Higher bond yields complicate the equation, raising construction costs and reducing the present value of future profits. That helps explain why investors are increasingly distinguishing cash-rich technology giants from heavily indebted infrastructure hopefuls.
Brookings offers another sobering calculation: Under its baseline model, AI infrastructure would need to support approximately $3.7 trillion in annual revenue by 2032 to justify the investment. For perspective, S&P Global estimates the entire AI infrastructure market generated just $337 billion in revenue in 2025. The measures aren’t directly comparable, but the gap illustrates how much future growth investors are counting on. (Sources: Brookings; S&P Global Market Intelligence.)
The danger isn’t necessarily that AI fails. It’s that investors correctly identify the next great technology—and still pay too much for its first generation of winners.
The Bottom Line: A Revolution Doesn’t Guarantee a Return
We remain constructive on AI’s long-term economic potential, including opportunities beyond the technology giants. But technological inevitability is not an investment strategy.
The $10 trillion question isn’t whether AI will transform the economy. It’s whether the companies financing its expansion can generate enough profits to justify the enormous spending. Rising borrowing costs, rapidly aging technology and ambitious earnings expectations leave little room for disappointment.
For investors, that means focusing on financial strength, sensible valuations and businesses capable of turning AI productivity into lasting profits. A billion-dollar data center isn’t necessarily a billion-dollar investment opportunity.
We’re not predicting another dot-com collapse. Broader credit markets remain relatively stable, and today’s technology leaders have substantial profits and financial resources. But if financing problems spread, earnings weaken or the largest customers pull back, the risks could become much more serious.
The railroads transformed America. The internet changed the world. Both left behind bankrupt companies, disappointed shareholders and infrastructure that remained valuable long after its original investors lost money.
AI may follow a similar path. The revolution could be extraordinary. The investment returns may be anything but.
For more information on our firm or to request a no-obligation investment and retirement check-up with Jon W. Ulin, CFP®, please call us at (561) 210-7887 or email jon.ulin@ulinwealth.com.
Note: Information provided on tax and estate planning is not intended to be a substitute for specific individualized tax or legal advice. We suggest that you discuss your specific situation with a qualified tax or legal advisor. This material was prepared with the assistance of AI. All content has been reviewed, edited, and approved by Ulin & Co Wealth Management prior to use.
You cannot invest directly in an index. Past performance is no guarantee of future returns.
Diversification does not ensure a profit or guarantee against loss. The information given herein is taken from sources that are believed to be reliable, but it is not guaranteed by us as to accuracy or completeness.
This is for informational purposes only and in no event should be construed as an offer to sell or solicitation of an offer to buy any securities or products. Please consult your tax and/or legal advisor before implementing any tax and/or legal related strategies mentioned in this publication as NewEdge Advisors, LLC does not provide tax and/or legal advice. Opinions expressed are subject to change without notice and do not take into account the particular investment objectives, financial situation, or needs of individual investors.
