Photo by Art Wall - Kittenprint on Unsplash
Photo by American Public Power Association on Unsplash
My read: The AI bubble debate is the wrong frame for most investors. Whether the bubble inflates further or eventually deflates, somebody has to build the physical infrastructure underneath it — and those builders are paying dividends right now.
The Common Belief — AI Risk Means Stay Away
$600 billion. That is the floor-level estimate for what five hyperscalers — Amazon, Alphabet, Microsoft, Meta, and Oracle — are projected to spend on capital expenditures in 2026 alone, with the ceiling reaching $720 billion, representing a 36% to 70% year-over-year increase. Approximately 75% of that range — between $450 billion and $540 billion — is directed specifically toward AI infrastructure, according to analysis reported by Google News citing Seeking Alpha's coverage of the capex waterfall thesis. Breaking it down by company: Amazon at $200 billion, Alphabet at $175 to $185 billion, Microsoft at $120 billion or more, Meta at $115 to $135 billion, and Oracle at $50 billion.
At the same time, MIT research cited across multiple financial outlets finds that 95% of enterprises report zero measurable ROI (return on investment — the financial return relative to what was spent) from generative AI investments, while OpenAI is projected to lose $14 billion in 2026 on roughly $13 billion in revenue. As of June 13, 2026, a Bank of America Global Fund Manager Survey conducted in November 2025 showed 45% of respondents identifying an AI equity bubble as the most significant current market risk. The Shiller CAPE ratio — a valuation measure comparing stock prices to ten years of averaged earnings — exceeded 40 in late 2025, a level previously seen only before the dot-com crash, with nearly 80% of S&P 500 gains in 2025 driven by AI-related companies.
The conventional conclusion: if you cannot see the ROI, do not fund the party. That logic is sound — but only if you are buying the companies throwing the party.
Where It Breaks Down
What the stay-away framing misses is the physicality of the AI buildout. Every dollar of that $600 to $720 billion has to flow into something concrete: electricity, steel, cooling equipment, fiber-optic cable, land, and construction labor. The companies supplying those inputs do not care whether the AI applications running inside those data centers ever generate commercial returns. They get paid on contract.
Consider the capex-to-sales ratios entering 2026: Oracle at 86%, Meta at 54%, Microsoft at 47%, Alphabet at 46%, Amazon at 25%. These are not companies cautiously rationing their AI investment — they are accelerating it, funded in part through debt. Futurum Group analysis notes that hyperscalers raised $108 billion in debt during 2025 alone, with projections suggesting $1.5 trillion in debt issuance over the coming years to sustain the buildout. That extraordinary volume of spending creates extraordinary, durable demand for physical inputs that have nothing to do with whether any particular AI model ever achieves commercial viability.
This is the capex waterfall thesis in plain English: capital pours in at the top (hyperscaler budgets), flows through intermediate layers (chip manufacturers, data center builders, grid operators), and arrives at the bottom as predictable, contracted revenue for non-tech infrastructure businesses. Dividend-focused utilities, industrial infrastructure REITs (real estate investment trusts that own income-generating properties like power substations and transmission lines), and power equipment manufacturers all sit downstream — and they collect regardless of what happens at the top of the stack.
The Power Numbers Behind the Thesis
Chart: 2026 projected capital expenditure by major hyperscaler. Combined floor-level spending exceeds $600 billion, with roughly 75% earmarked for AI infrastructure.
Goldman Sachs forecasts $7.6 trillion in cumulative AI infrastructure spending from 2026 through 2031, broken down as $5.1 trillion for compute, $2.1 trillion for data centers, and $358 billion for power infrastructure. Goldman's own analysts flag real uncertainty in that figure — shortening the assumed chip lifespan from five years to three years would increase annual depreciation costs by nearly $1 trillion, per Goldman's own analysis. The top-line projection is a scenario, not a commitment. But even discounted substantially, the power and data center construction layers represent multi-year contracted demand.
The global AI infrastructure market is projected at $1.37 trillion in 2026 alone, rising to nearly $3 trillion by 2028 and over $5 trillion by 2030, according to market research cited across financial publications. Nvidia is assumed to capture approximately 75% of all compute spending through 2031 — translating to roughly $3.8 trillion in cumulative revenue. But compute is only one layer. The data center construction and power layers beneath it are where dividend-focused ETFs (exchange-traded funds — baskets of stocks that trade like a single share) gain exposure.
As of June 13, 2026, the Energy Information Administration's May 2026 report projects that commercial power consumption will exceed residential consumption for the first time in 2027, driven primarily by AI data center demand. Data centers globally consumed 448 trillion watt-hours of electricity in 2025, and U.S. data center power demand is projected to grow from 4% of total electricity generation in 2023 to 9% by 2030. To fund the required grid upgrades, utilities will need 10% to 19% additional annual revenue above previous forecasts — a significant capital requirement, but one that translates directly into utility revenue over time.
The Smart AI Trends analysis of how Wall Street learned to value AI giants without a profit line is worth reading alongside this — it explains precisely why direct AI equity positions carry a structurally different risk profile than the infrastructure layer discussed here.
What It Means at Kitchen-Table Level
Think of the California Gold Rush. The miners betting everything on finding a vein mostly lost. The merchants selling pickaxes, denim, and room-and-board to those miners made reliable money — because demand for supplies existed regardless of who struck it rich. The AI-era equivalent of the pickaxe business includes electric utilities, grid operators, and data center construction firms.
As of June 13, 2026, the Vanguard Utilities ETF stands as a $9 billion fund with 62.4% of assets allocated to electric utility stocks — companies whose revenue increasingly comes from powering data centers. The Defiance AI & Power Infrastructure ETF (AIPO) was named Best New Thematic ETF at the ETF.com Awards 2026, signaling that institutional investors are formalizing the indirect-AI-beneficiary category as a distinct asset class. These are not speculative moonshots; they are bets that electricity consumption will keep rising — a considerably safer wager than betting on which specific AI company survives the coming shakeout. (Industry predictions cited in multiple financial outlets warn that 99% of AI startups will be gone by late 2026 as budgets flow only to products that demonstrate verifiable value.)
For a 38-year-old building a long-term portfolio, the math works out to this: dividend-paying utility ETFs generate quarterly income, exhibit lower price volatility than growth stocks, and are tied to a demand curve that expands as long as data centers keep operating — whether or not the AI software inside them ever turns a profit. The 78% of Americans who told surveyors they are concerned about higher electricity bills from data center construction are, without realizing it, describing the same revenue stream that utility shareholders are collecting.
One honest caveat before anyone rushes to rebalance: utilities are not riskless. That 10–19% additional annual revenue requirement is money utilities must raise upfront, often through rate increases or new debt of their own. If hyperscaler capex slows sharply — say, because enterprise ROI data stays as weak as the current MIT findings suggest — the demand growth curve could plateau faster than utility capital plans assumed. The picks-and-shovels thesis is more resilient than direct AI equity exposure, but a serious capex reversal would not leave infrastructure stocks untouched.
Photo by Markus Winkler on Unsplash
A Better Frame — Three Moves Worth Making This Week
Pull up your brokerage or ETF holdings and identify any overlap with Amazon, Alphabet, Microsoft, Meta, or Oracle. If you hold a broad S&P 500 index fund, you already carry meaningful AI concentration — nearly 80% of 2025's S&P gains were driven by AI-related companies. Understanding what you already own tells you whether infrastructure plays function as a hedge against your existing positions or as a new exposure entirely.
The Vanguard Utilities ETF and the Defiance AI & Power Infrastructure ETF (AIPO) are two documented starting points in this category as of June 13, 2026. For each fund you research, check: the top ten holdings, the expense ratio (the annual management fee expressed as a percentage of your investment), the dividend yield, and how the fund behaved during previous interest-rate-increase cycles. Utilities are interest-rate-sensitive — when rates rise, their dividend yields become relatively less attractive. This is a research checklist, not a recommendation.
The signal that matters most for the capex waterfall thesis is not whether a new AI model gets announced — it is whether enterprise ROI data starts improving from its current floor (95% reporting zero impact, per MIT research). Bookmark the EIA's quarterly electricity consumption reports and Goldman Sachs's periodic infrastructure spending updates. If commercial ROI materializes, capex sustains and the waterfall keeps flowing to the infrastructure layer. If it does not, the AI equity tier corrects sharply — but power infrastructure demand, already under contract, adjusts much more slowly.
Frequently Asked Questions
How can I invest in AI infrastructure without buying tech stocks directly?
The capex waterfall thesis points toward utilities ETFs, infrastructure REITs, and thematic ETFs focused on power infrastructure. As of June 13, 2026, examples include the Vanguard Utilities ETF — a $9 billion fund with 62.4% in electric utility stocks — and the Defiance AI & Power Infrastructure ETF (AIPO), which won Best New Thematic ETF at the ETF.com Awards 2026. These funds provide exposure to the electricity and construction demand that AI data centers generate without requiring a bet on which AI software company ultimately wins. Nothing in this article constitutes a recommendation to purchase any specific fund.
Is the AI spending boom really similar to the dot-com bubble, and does it matter for dividend ETF investors?
The parallels are real: the Shiller CAPE ratio exceeded 40 in late 2025, a threshold previously reached only before the dot-com crash, and nearly 80% of S&P 500 gains in 2025 were AI-driven. One structural difference is that today's spending is funded primarily by cash-generating profitable companies rather than unprofitable startups — though hyperscalers have also leaned heavily on debt, raising $108 billion in 2025 alone per Futurum Group, with projections of $1.5 trillion in debt issuance over coming years. For dividend-focused utility investors, the bubble question is somewhat secondary: if capex continues, data center power demand keeps rising; if capex collapses, demand growth slows but does not reverse overnight because existing data centers still consume power. The risk is real but structurally more muted than for direct AI equity exposure.
What happens to utility stocks when AI data center electricity demand grows rapidly?
Growing data center demand creates contracted revenue streams for utilities — but it simultaneously demands significant upfront capital investment. As of June 13, 2026, the Energy Information Administration projects commercial power consumption will surpass residential for the first time in 2027, driven primarily by AI data centers. U.S. data center power demand is projected to grow from 4% to 9% of total electricity generation between 2023 and 2030. Utilities serving this demand need 10% to 19% more annual revenue than previously forecast just to fund the required grid and transmission upgrades. That additional revenue is a tailwind for utility earnings over time — but the upfront capital requirements are a headwind in the near term, and 78% of Americans surveyed expressed concern that data center growth will raise their electricity bills as utilities pass those costs along.
Disclaimer: This article is for informational purposes only and does not constitute financial advice. The content reflects editorial commentary and analysis of publicly reported financial information — no independent product testing or personal investment experience is represented. Readers should conduct their own due diligence and consult a qualified financial advisor before making any investment decisions. Research based on publicly available sources current as of June 13, 2026.
No comments:
Post a Comment