The artificial intelligence revolution has triggered unprecedented capital spending, with Big Tech firms planning to invest $5.2 trillion over five years. While markets have rewarded this spending so far, historical analysis reveals a concerning pattern: Infrastructure booms typically result in overinvestment, excess competition, and poor stock returns.
Kai Wu, author of the October 2025 research paper “Surviving the AI Capex Boom,” conducted a comprehensive historical analysis spanning multiple dimensions.
Historical Infrastructure Booms
Wu, the founder and chief investment officer of Sparkline Capital, examined major capital expenditure cycles throughout history, including:
Railroad expansion in the 1860s-1890sTelecom fiber optic buildout in the late 1990sCurrent AI infrastructure spending (2023-present)
He compared the scale of these investments relative to the gross domestic product and analyzed how shareholders in infrastructure-building companies fared during and after each boom.
The Investing Risk You Might Be Overlooking When Buying Popular StocksFirm-Level Stock Performance
Using decades of market data from 1963 to 2025, Wu analyzed the:
Returns of companies with high asset growth versus low asset growthPerformance of firms rapidly increasing capital expendituresResults across all 10 market sectors and major geographic regionsRelationship between capital intensity and stock returnsThe Magnificent Seven’s Transformation
The study specifically tracked how Apple AAPL, Microsoft MSFT, Amazon.com AMZN, Meta Platforms META, Google GOOGL, Nvidia NVDA, and Tesla TSLA are transitioning from asset-light business models to capital-intensive operations. It examines:
Historical capital expenditure trendsChanges in return on invested capitalFree cash flow deteriorationRising debt levels and circular financing arrangementsKey Findings
1. High Capital Spending Predicts Poor Returns
The research uncovered a stark pattern: Companies aggressively growing their balance sheets underperformed conservative peers by 8.4% annually from 1963 to 2025. This “asset-growth anomaly” held true across:
All 10 market sectorsMultiple geographic regions (US, Europe, Asia)Both boom and bust periodsDifferent types of capital spending
Firms rapidly increasing capital expenditures showed similar underperformance, with the effect accelerating during the dot-com bust but remaining consistent even outside bubble periods.
2. The AI Boom Is Historically Massive
Current AI spending already exceeds the internet boom’s peak relative to GDP. When adjusted for the shorter useful life of AI chips versus physical infrastructure:
AI spending surpasses even the railroad buildout of the 1860s-1870s.Big Tech firms are on track to spend nearly $400 billion in 2025 alone.AI capital spending accounts for an estimated half of US GDP growth.
This scale of investment requires generating $2 trillion in annual revenue by 2030 to justify costs, yet current AI revenues stand at only $20 billion—requiring a 100-fold increase.
Is Private Credit Drowning in Capital? These Are the Strategic Implications for Investors
3. The Magnificent Seven Face Unique Risks
These companies succeeded through asset-light models leveraging intangible assets, achieving 22.5% returns on invested capital. However, they’re now becoming asset-heavy:
Capital expenditures have surged from 4% to 15% of revenue since 2012.Meta, Microsoft, and Alphabet each plan to spend 21% to 35% of revenue on capital expenditure.This exceeds both current global utility sector averages and the spending of AT&T T at the peak of the telecom bubble.
The research shows asset-heavy firms have consistently underperformed asset-light ones. This effect also exists within sectors, with asset-heavy firms lagging their asset-light sector peers.
4. Deteriorating Fundamentals Signal Trouble
Several concerning trends emerged:
Free cash flow is declining as capital spending accelerates.Circular financing deals mirror dot-com era practices (that is, Nvidia investing $100 billion in OpenAI, which then buys Nvidia chips).Debt levels are rising, including Meta’s $27 billion off-balance-sheet financing.Depreciation charges could climb from $150 billion to $400 billion annually over five years.Useful life assumptions may be overly optimistic given rapid GPU replacement cycles.
5. The AI Prisoner’s Dilemma
Big Tech faces a game theory problem. While the optimal strategy would be moderate, coordinated investment, each company fears being left behind. This forces all players into aggressive spending, potentially destroying the collective profit pool even if individual firms succeed technologically.
The AI race collapses previously separate markets (search, social media, shopping) into one winner-take-all competition, eliminating the comfortable oligopoly structure that made these companies so profitable.
6. Past Boom Winners Were Often Not the Infrastructure Builders
Historical analysis revealed:
Railroad companies suffered through multiple panics and bankruptcies before stabilizing decades later.Telecom stocks crashed 92% after the dot-com bust and remain 60% below their peak 25 years later.Railroad builders captured only a tiny fraction of the economic value they created.
The real winners were often the customers who benefited from subsidized infrastructure. Excess capacity drove bandwidth costs down 90% after the dot-com bust, fueling the rise of Netflix NFLX and Facebook.
How Sustainable Is US Exceptionalism?
7. Valuation Risk Extends Beyond Infrastructure
Even asset-light early adopters face danger from excessive valuations. An analysis of the performance of “dot-com darlings” from 2000 to 2019 showed:
Despite achieving 12% annual sales growth as promised, these companies lost 80% of their value from 2000 to 2002.The problem was starting valuations of 33 times sales that compressed to 5 times.It took 18 agonizing years for investors to break even despite strong fundamental performance.
8. A Value Approach Offers Protection
The research demonstrated that incorporating both tangible and intangible assets into valuation models allowed investors to:
Maintain structural exposure to innovative technologiesDynamically rotate from overvalued to undervalued opportunitiesOutperform during both boom and bust periods
During the dot-com era, this approach favored Cisco Systems CSCO and Microsoft initially, rotated to Progressive PGR and FedEx FDX as tech became overvalued, then returned to beaten-down Amazon after the crash.
Key Investor Takeaways
1. Don’t Confuse Technology Potential With Investment Returns
AI will likely prove transformative, but that doesn’t guarantee good returns for infrastructure builders. As the railroad and internet examples show, AI can revolutionize society while still delivering poor returns for investors in the companies building it. Thus, it is important to separate your belief in AI as a technology from your investment thesis.
2. Watch Capital Intensity, Not Just Growth
The Magnificent Seven’s (Alphabet, Amazon.com, Apple, Meta Platforms, Microsoft, Nvidia, and Tesla) transition to capital-intensive models is concerning, as this model is historically associated with lower returns. Asset-heavy firms require constant reinvestment just to maintain their competitive position and face easier competitive replication. Wu’s research shows that AI early adopters currently trade at only a 13% valuation premium versus 137% for infrastructure players.
3. Diversify Beyond Obvious AI Plays
With the Magnificent Seven now comprising about 35% of the S&P 500—exceeding dot-com concentration levels—investors face significant single-theme risk. AI beneficiaries exist across all sectors and geographies, offering similar exposure with better diversification. Examples include companies using AI to improve operations (such as JPMorgan Chase JPM, Caterpillar CAT, Walmart WMT) rather than building infrastructure.
4. Use Robust Valuation Metrics
Traditional value metrics miss intangible assets, causing value investors to miss innovation. But ignoring valuation entirely leads to dot-com-style disasters. A comprehensive approach considering both tangible and intangible capital offers the best balance. Wu’s research shows that cheap (value) AI stocks have consistently outperformed, while expensive ones delivered disappointing returns despite the technology’s success.
5. Expect Capacity Overshoots to Benefit Adopters
Historical patterns suggest the current buildout may create excess AI computing capacity, driving costs down and effectively subsidizing early adopters. This parallels how excess fiber-optic capacity enabled Netflix and Facebook’s rise.
6. Monitor Free Cash Flow and Balance Sheets
As capital spending accelerates, watch for deteriorating fundamentals: declining free cash flow, rising debt levels, circular financing arrangements, and aggressive useful life assumptions that mask true depreciation costs. Be wary of companies whose capital spending is growing faster than revenue, especially if they’re issuing debt to finance expansion or entering into circular investment deals with customers and suppliers.
7. Consider Geographic and Sector Diversification
AI beneficiaries aren’t limited to US tech giants. Taiwan, Korea, the Netherlands, Israel, Germany, Japan, and Switzerland all offer significant exposure through different parts of the AI value chain.
8. Remember the 100-Fold Revenue Gap
For the current spending to make economic sense, AI revenues must grow from $20 billion to $2 trillion annually by 2030. This enormous gap suggests either spending will slow, revenues will disappoint, or margins will compress from competition. Maintain skepticism about infrastructure builders’ ability to monetize investments at current valuations. The math simply doesn’t work without unprecedented revenue growth or exceptional pricing power.
9. Stay Invested, But Be Selective
Despite all these concerns, cheap AI stocks have outperformed the market consistently. The goal isn’t to avoid AI but to maintain exposure while managing valuation and capital intensity risks.
ConclusionThe AI revolution is real and will likely transform the economy. However, history teaches us that the path from technological promise to investor profits is often punctuated by overinvestment, excess capacity, and disappointing returns for infrastructure builders.
With the Magnificent Seven embarking on history’s largest capital spending spree while accounting for about 35% of the S&P 500, investors face concentrated exposure to companies transitioning from winning asset-light models to historically challenged asset-heavy operations.
The solution isn’t to abandon AI but to think more broadly about how to capture its benefits. By combining AI exposure with value discipline, focusing on early adopters with lower capital requirements, and maintaining diversification across the AI value chain, investors can participate in the AI revolution while managing the substantial risks that historical capital cycles suggest lie ahead.
The winners of past infrastructure booms were often not those who built the infrastructure but those who used it wisely. Today’s smart money may lie not in the companies spending hundreds of billions on AI data centers, but in those positioned to benefit from the innovation—and potential oversupply—that this unprecedented spending will create.
Wu’s firm offers two exchange-traded funds that implement his findings: Sparkline Intangible Value ETF ITAN and Sparkline International Intangible Value ETF DTAN. Morningstar shows Sparkline Intangible Value ETF’s current P/E to be just 14.4 (as compared to 22.9 for Vanguard S&P 500 ETF VOO) and Sparkline International Intangible Value ETF’s to be just 12.6. Like the funds of such families as AQR, Avantis, Bridgeway, and Dimensional, Sparkline’s investment strategies are based on empirical research and are implemented in a systematic, transparent, and replicable manner.
Larry Swedroe is the author or co-author of 18 books on investing, including his latest, Enrich Your Future. He is also a consultant to RIAs as an educator on investment strategies.
Larry Swedroe is a freelance writer. The opinions expressed here are the author’s. Morningstar values diversity of thought and publishes a broad range of viewpoints.