The AI Inflection Point: A Critical Assessment of the 2025 ‘Bubble’ Hypothesis

by Gemini 3.0, Deep Research. Warning, LLMs may hallucinate!

Executive Summary

In December 2025, the global financial markets stand at a precarious juncture, dominated by a single, monolithic theme: Artificial Intelligence. The release of Howard Marks’ Oaktree Capital memo, “Is It a Bubble?”, marks a significant milestone in the ongoing debate regarding the sustainability of asset prices and capital deployment in the AI sector.¹ Marks, whose celebrated “bubble.com” memo in January 2000 presciently identified the excesses of the dot-com era, returns to the subject exactly twenty-five years later to interrogate the current market psychology.² His analysis arrives amidst a backdrop of unprecedented market concentration, where the S&P 500 has effectively become a proxy for a handful of mega-cap technology firms, and where corporate credit markets have begun to exhibit signs of strain under the weight of financing the most capital-intensive infrastructure build-out in human history.⁴

This report serves as a comprehensive institutional analysis of the thesis presented by Howard Marks. By synthesizing macroeconomic data, sector-specific capital expenditure (CapEx) trends, credit market signals, and technical hardware constraints available as of late 2025, we aim to rigorously stress-test Marks’ conclusions. Our analysis largely concurs with Marks’ identification of bubble-like characteristics, specifically regarding the migration of risk from equity to credit markets and the potentially flawed assumptions underpinning the amortization of AI hardware.⁶

However, this report also identifies critical divergences from the 2000 analogy. Unlike the revenue-light entities of the dot-com era, the primary architects of the 2025 AI build-out—the “Magnificent Seven”—possess robust cash flows and are delivering tangible productivity gains that are already visible in US GDP data.⁷ The risk, therefore, may not be a systemic collapse of equity values for the market leaders, but rather a severe dislocation in the credit markets and a “capex winter” that devastates the secondary ecosystem of hardware vendors, data center developers, and “neocloud” providers who have leveraged their balance sheets against rapidly depreciating silicon assets.⁹

The following sections provide an exhaustive deconstruction of the current market environment, examining the psychological, financial, and physical dimensions of the AI trade to determine the validity of the bubble hypothesis.

1. The Oaktree Framework: Deconstructing the “Bubble” Psychology

1.1 The Definition of Irrationality

To evaluate the current market, one must first establish the definitions set forth by Marks. He argues that a bubble is not merely a phenomenon of high valuations, but a psychological state characterized by the suspension of disbelief and the rejection of historical norms.² In his December 2025 memo, Marks observes that investors have once again embraced the dangerous philosophy that “this time is different,” a mindset that justifies limitless valuations based on the perceived infinite potential of a revolutionary technology.⁴

Marks draws a critical distinction between “mean-reversion bubbles” and “inflection bubbles.” The former, such as the South Sea Bubble or the subprime mortgage crisis, offer no lasting social utility and result in pure wealth destruction. The latter, which includes the railroad mania of the 19th century and the dot-com boom of the late 20th century, fundamentally transform the economy but still lead to massive losses for investors who overpay for the transformation.⁴ His classification of AI as an “inflection bubble” is vital; it acknowledges the reality of the technological breakthrough—potentially the “railroad of the 21st century”—while warning that the financial vehicles used to ride this railroad may crash before reaching the destination.⁴

1.2 The Role of Uncertainty and Imagination

A central pillar of Marks’ argument is the interaction between novelty and valuation. When a technology is “new and seemingly revolutionary,” there is no historical data to restrain the investor’s imagination. In the absence of constraints, the future appears limitless, and asset prices detach from any metric of predictable earning power.⁶ In 2025, this dynamic is evident in the valuation of companies promising Artificial General Intelligence (AGI). The “trillion-dollar promises” of AGI serve as a narrative shield, deflecting scrutiny of near-term profitability and justifying capital outlays that dwarf historical precedents.⁴

Marks argues that in such an environment, the boundary between optimism and irrationality becomes blurred. He cites the “piling in” behavior of investors who fear missing out (FOMO) on the next great wealth creation event, a behavior that bypasses the necessary due diligence regarding risk-adjusted returns.⁶ This observation is corroborated by the behavior of retail and institutional flows into AI-themed ETFs and indices, which have continued unabated despite rising interest rates and geopolitical friction.¹¹

1.3 The Aggressive Shift to Debt

Perhaps the most specific and technically astute warning in the Oaktree memo concerns the evolution of financing structures. Marks identifies a dangerous transition in the AI sector: the move from equity-funded growth, derived from the immense operating cash flows of the tech giants, to debt-funded growth.⁶ He posits that while debt is an appropriate tool for financing stable, predictable outcomes, it is toxic when applied to ventures where the outcome is “purely a matter of conjecture”.⁶

This warning is critical because it highlights a structural fragility that did not exist in the early phases of the AI boom (2023-2024). In the initial phase, the build-out was funded by the “house money” of Google, Microsoft, and Meta. By late 2025, however, the capital requirements had exceeded even their massive free cash flow capabilities, forcing them—and more dangerously, their satellite partners—into the bond markets.⁶Marks explicitly notes that aggressive debt financing for technological revolutions “has the potential to magnify all the negative effects,” creating a leverage trap that could turn a correction into a crisis.⁶

2. Macro-Valuation Framework: Testing the “Irrationality” Hypothesis

To determine whether the market has truly crossed the threshold into irrationality, we must conduct a rigorous quantitative analysis of valuation metrics as of December 2025, contrasting them with historical bubble peaks.

2.1 The Elevation of Price-to-Earnings Multiples

As of December 10, 2025, the S&P 500 trades at a forward Price-to-Earnings (P/E) ratio ranging between 23.1x and 27.88x, depending on the specific consensus estimates utilized.¹³ This represents a significant deviation from the long-term historical average of approximately 16x. While this valuation is elevated, it remains below the stratospheric peaks of ~30x seen during the height of the dot-com mania in March 2000. However, relying on the headline index number masks the extreme bifurcation within the market.

The Nasdaq 100, which serves as the primary vehicle for AI exposure, trades at a forward P/E of approximately 32.35x.¹⁶ This premium implies that investors are pricing in not just growth, but a flawless execution of the AI roadmap. A closer examination of the “Magnificent Seven” reveals a dispersion that complicates the bubble narrative. For instance, Meta Platforms trades at roughly 25.8x forward earnings, a valuation that could be considered reasonable given its growth profile.¹⁷ In contrast, Tesla trades at roughly 200x, and Nvidia maintains a multiple of 51.5x, despite its massive market capitalization.¹⁷ This variance suggests that while pockets of the market are undeniably frothy, the broad-based mania where every tech stock traded at triple-digit multiples (as seen in 1999) is not fully replicated in 2025.

Table 1: Comparative Valuation Metrics (Historical Peaks vs. December 2025)

2.2 The Shiller CAPE Ratio: A Historical Warning

The Cyclically Adjusted Price-to-Earnings (CAPE) ratio, a metric favored by economic historians for its ability to smooth out business cycle volatility, stands at 39.42 in December 2025.¹⁸ This is a profound data point that strongly supports Marks’ thesis. A CAPE ratio approaching 40 has only been observed once before in U.S. history: during the dot-com bubble. This metric suggests that current prices are historically stretched relative to the long-term earnings power of corporate America.

The implication of a CAPE ratio at this level is that future returns over the subsequent decade are statistically likely to be low or negative, as valuations eventually revert to the mean. This aligns with the Oaktree memo’s concern that asset prices are not justified by “predictable earning power”.⁶ The market is effectively pricing in a permanent upward shift in profitability—a “productivity miracle”—that has yet to fully materialize in the GAAP earnings of the broader S&P 500, beyond the technology sector.

2.3 The Buffett Indicator: The GDP Disconnect

The Buffett Indicator, defined as the ratio of Total Market Capitalization to GDP, provides perhaps the most alarming signal in late 2025. The ratio has reached approximately 224%, shattering previous records.¹⁹ This compares to a peak of around 140% in 2000 and roughly 200% in late 2021. Marks’ assertion that valuations have gone “beyond past norms” is mathematically undeniable through this lens.⁶

Proponents of the AI trade often argue that the Buffett Indicator is flawed because the S&P 500 is a global index while GDP is a domestic measure. They contend that the global revenue streams of U.S. tech giants justify a higher ratio. However, even when adjusting for international revenues, a ratio exceeding 200% implies a conviction that corporate profit margins will remain permanently elevated above historical norms. This belief is predicated on the assumption that AI will drive a massive reduction in labor costs, thereby expanding margins.²⁰ Should these margin gains fail to materialize, or should they be competed away, the valuation compression could be severe.

2.4 Concentration Risk and the “Passive Bubble”

A distinct feature of the 2025 market, which differs structurally from 2000, is the extreme level of concentration. The top 10 companies now account for 42% of the S&P 500’s total value.⁵ This “hollowing out” of the index means that the performance of the entire U.S. equity market is effectively a derivative of the AI capex cycle.

This concentration creates a mechanical feedback loop. As investors pour money into passive index funds, they are forced to allocate 42 cents of every dollar to these top 10 names, regardless of valuation. This aligns with Marks’ observation that in bubbles, investors “pile in” without regard for whether the price offers a “reasonable return with a tolerable amount of risk”.⁶ The structural dominance of passive investing in 2025 exacerbates the bubble dynamic, creating a fragility where a reversal in sentiment towards just three or four companies could trigger a systemic market decline.

3. The Capex Conundrum: The “Field of Dreams” Build-out

The central economic debate of late 2025 revolves around Capital Expenditure (Capex). Is the massive, synchronized spending on data centers, GPUs, and energy infrastructure a rational investment in the future of the economy, or is it a speculative mania comparable to the fiber-optic glut of 2001?

3.1 The Scale of Investment

The numbers are staggering. As of late 2025, the “Hyperscalers” (Amazon, Alphabet, Meta, Microsoft) are allocating nearly 70% of their operating cash flows to AI-related investments.²¹ Aggregate AI capital expenditure for these firms is projected to exceed $405 billion in 2025 alone.²² This represents a fundamental shift in the business models of these companies, moving from “asset-light” software platforms to “asset-heavy” industrial utilities.

Marks notes that this “installation phase” of a technology cycle is historically characterized by “necessary but not necessarily financially wise investments”.⁶ The infrastructure is being built on the “Field of Dreams” premise: if we build it, they will come. The sheer velocity of this spending—doubling and tripling within a span of 24 months—echoes the railway mania, where parallel tracks were laid between the same cities by competing firms, leading to inevitable bankruptcies despite the long-term utility of the rail network.⁴

3.2 The Sequoia “Revenue Gap”

In June 2024, Sequoia Capital partner David Cahn published a seminal analysis identifying a $600 billion gap between the revenue required to justify the AI infrastructure build-out and the actual revenue being generated by the AI ecosystem.²³ By December 2025, Cahn’s analysis has been updated, and the conclusion is sobering: the gap has not closed; it has widened.

While Hyperscalers report strong growth in their cloud divisions—Google Cloud, for instance, delivered double-digit growth in Q3 2025 driven by AI ⁷—the overall revenue generated by end-user AI applications remains a fraction of the cost of the infrastructure supporting it. The “return on investment” (ROI) paradox is evident in enterprise adoption data. Reports from MIT and Enterprise Technology Research indicate that only a small percentage (single digits to low teens) of enterprises are seeing sustained, scalable ROI from their generative AI deployments in late 2025.²⁴This “productivity J-curve” suggests that while the benefits of AI may be real, they are lagging significantly behind the capital deployed to achieve them.

3.3 The “Circular Financing” Risk

A critical vulnerability in the 2025 ecosystem is the phenomenon of “circular financing,” explicitly warned about by Marks.⁴ This dynamic involves Big Tech firms investing in AI startups and “neocloud” providers, who then use that capital to purchase cloud computing services back from the Big Tech investors.

For example, a Hyperscaler might invest $1 billion in an AI model developer. That developer then signs a committed use contract to spend $1 billion on the Hyperscaler’s cloud compute. On paper, the Hyperscaler records $1 billion in revenue and cloud growth. In reality, they have simply moved cash from their balance sheet to their income statement, intermediated by a startup. This creates an illusion of organic demand. If venture capital funding tightens, or if the startups fail to find sustainable business models, this circular revenue stream will evaporate, revealing the true, lower level of end-market demand. This aligns perfectly with Marks’ skepticism regarding whether the current enthusiasm is “merited or irrational”.¹

4. The Debt Supercycle: Where the Bubble is Most Acute

While equity valuations garner the headlines, this report identifies the credit marketsas the locus of the most acute risk in 2025. It is here that Marks’ warning about “aggressive debt financing” is most prescient and where the data most strongly validates the “bubble” designation.

4.1 The Shift from Equity to Debt

Marks emphasizes that “companies are committing amounts that require debt financing” and that for some, this leverage is “aggressive”.⁶ The data from late 2025 confirms this shift. Historically, the tech giants were bastions of fiscal conservatism, holding massive net cash positions. However, the voracious capital appetite of the AI arms race has forced them into the bond market.

In the second half of 2025, we witnessed a surge in mega-bond issuances. Oracle issued $18 billion in debt, Meta sold $30 billion, and Alphabet issued $15 billion.¹² This transition from funding growth via operating cash flow to funding via leverage introduces a new dimension of risk: interest rate sensitivity. For the first time, the core engines of the U.S. economy are becoming sensitive to the cost of capital, exposing them to refinancing risks that were previously irrelevant to their business models.

4.2 Spreads and Credit Risk Signals

The bond market, often considered the “smart money,” has begun to price in this elevated risk profile. In late 2025, credit spreads for technology issuers widened significantly. Notably, the spread on Oracle’s 5-year bonds doubled to 104 basis points over Treasurys, and the cost of insuring its debt via Credit Default Swaps (CDS) surged.²⁶

This widening spread indicates that fixed-income investors are demanding a higher risk premium to hold the debt of companies that are “betting the farm” on AI. It validates Marks’ guideline that it is “not okay” to use debt where the outcome is uncertain.⁶ The bond market is effectively signaling that it views the future cash flows from these AI investments as far less certain than the equity market does.

4.3 The Data Center ABS Boom and Residual Value Risk

A critical and rapidly expanding niche in the 2025 financial landscape is the market for Data Center Asset-Backed Securities (ABS). This market has grown to over $100 billion, becoming a primary funding mechanism for the physical infrastructure of AI.²⁷These securities package the lease payments from data centers into tradable bonds, sold to institutional investors.

The structural flaw in this market lies in the assumption of residual value. Marks notes that investors are beginning to question the “residual value risk of data centers when the bonds mature”.⁶ An ABS deal typically assumes that the data center—and often the specialized equipment within it—will retain significant value at the end of the lease term (often 5-7 years). However, given the rapid pace of technological change (discussed in Section 5), a data center optimized for 2023-era H100 GPUs may be functionally obsolete by 2028. If the residual value of these assets approaches zero, or if the cost to retrofit them is prohibitive, the equity tranches of these ABS structures could be wiped out, and senior tranches could suffer impairments. This represents a classic “duration mismatch,” where long-term debt is secured by assets with shrinking economic lives.

4.4 The “Neocloud” Shadow Banking System

Further compounding the credit risk is the emergence of “GPU-backed lending.” In a practice reminiscent of the subprime era, AI startups and “neocloud” providers (such as CoreWeave and Lambda Labs) are borrowing billions of dollars using their stockpiles of Nvidia GPUs as collateral.¹⁰

Short-sellers and skeptical analysts have flagged this as a critical failure point. The value of a GPU is highly volatile and subject to rapid depreciation upon the release of a newer generation. If the market value of an H100 chip drops by 50% due to the release of Nvidia’s next-generation architecture, the collateral value backing these loans collapses. This would trigger margin calls, forcing the liquidation of chips into a saturated market, further depressing prices. This “asset-backed lending” on rapidly depreciating consumer electronics is the very definition of “aggressive debt financing” that Marks warns against.⁶

5. The Physical Constraints: Thermodynamics & Obsolescence

To fully understand the financial risk, one must look beyond the balance sheets to the physical reality of the hardware itself. The 2025 AI bubble is constrained by the laws of thermodynamics and the relentless march of semiconductor innovation.

5.1 The Depreciation Trap and “Phantom Earnings”

Corporate accounting standards typically allow for the depreciation of server hardware over a period of 5 to 6 years. However, the “economic life” of a cutting-edge AI GPU in 2025 is estimated to be closer to 2 to 3 years. This acceleration is driven by Nvidia’s aggressive product roadmap, which has moved to a one-year release cycle (Hopper to Blackwell to Rubin).²⁸

This discrepancy creates a massive “phantom earnings” problem. Companies are reporting profits based on a 6-year amortization schedule, while in reality, they must replace their capital stock every 3 years to remain competitive. Analysts at Barclays have noted this discrepancy, cutting earnings forecasts for AI firms in late 2025 to reflect “more realistic depreciation assumptions”.²⁹ If the entire industry were forced to mark-to-market the value of their GPU clusters and adopt a 3-year depreciation schedule, corporate earnings for the tech sector would be significantly lower, and P/E ratios would be even higher than they currently appear. This supports Marks’ concern that valuations are based on “predictable earning power” that may be illusory.⁶

5.2 The Liquid Cooling Retrofit Wall

A major, underappreciated capital destroyer in 2025 is the thermal limit of current data center infrastructure. The transition to Nvidia’s Blackwell architecture and beyond requires liquid cooling, as traditional air cooling is insufficient to manage the heat generated by racks with power densities exceeding 100kW (compared to the 10kW legacy standard).³⁰

Retrofitting existing air-cooled data centers for liquid cooling is technically difficult and prohibitively expensive, often costing 7-10% more than building new facilities from scratch.³¹ This implies that a vast swathe of the existing data center capacity—which forms the collateral for the ABS market discussed in Section 4—is at risk of becoming “stranded assets.” These facilities cannot support the latest AI hardware without capital injections that destroy the economics of the original investment. This physical constraint reinforces the “overbuild” narrative found in historical infrastructure bubbles, where early infrastructure becomes obsolete before it can pay for itself.

5.3 Power Density and the Energy Ceiling

The demand for power has outstripped the grid’s ability to supply it. Goldman Sachs estimates that AI will consume 19% of total data center power demand by 2028.³²This insatiable hunger for electricity has led to a scramble for “behind-the-meter” power deals, including the restart of nuclear reactors and massive investments in renewable energy.

While this validates the importance of AI (it is a technology vital enough to reshape national energy policies), it imposes a hard ceiling on growth. The lead time for new power generation (years) is misaligned with the lead time for chip deployment (months). This energy bottleneck drives up the marginal cost of compute, compressing the margins of AI operators and making the high valuation multiples even harder to justify.

6. The Counter-Narrative: Productivity & The “Golden Age”

While the financial structure (debt and valuation) strongly suggests a bubble, the economic data offers a potent counter-narrative that complicates the “irrationality” claim. To provide a balanced analysis, we must consider the arguments of the “AI Bulls.”

6.1 GDP and Tangible Economic Impact

Unlike the cryptocurrency bubble of 2021, which produced little in the way of tangible economic output, the AI boom is clearly visible in the macroeconomic data. In the first half of 2025, AI-related capital expenditures contributed a measurable 1.1% to U.S. GDP growth.⁸ Furthermore, Goldman Sachs forecasts that AI will boost potential U.S. GDP growth to approximately 2.4% by 2027.³³

This data suggests that the spending, while massive, is generating real economic activity. It is hiring engineers, pouring concrete, and manufacturing high-value components. This “real economy” impact distinguishes AI from purely speculative financial bubbles. The “railroad” analogy used by Marks cuts both ways: yes, railroad investors lost money, but the railroads did get built, and they did power the American economy for a century. The same may be true for the GPU clusters of 2025.

6.2 The Deflationary Thesis

Cathie Wood of ARK Invest represents the vanguard of the bullish argument. She posits that the “bubble” is actually a prelude to a massive deflationary boom. In her view, AI will drastically lower the cost of knowledge work, leading to a productivity surge that justifies the current capex.³⁴

If AI agents can indeed automate 30-40% of enterprise tasks, as some optimistic projections suggest, the resulting margin expansion for the S&P 500 would be substantial. This margin expansion would compress the current high P/E ratios, validating the current stock prices. From this perspective, the debt being incurred today is a rational investment in a lower-cost future economy.

6.3 The “Winner-Takes-Most” Dynamic

Oaktree’s own Bob O’Leary argues that technological revolutions are “winner-takes-all” games.⁶ This implies that for the winners, the rewards are astronomical. The performance of the “Magnificent Seven” in 2025 supports this view. While smaller players struggle, companies like Alphabet and Microsoft continue to report robust growth in their core businesses, subsidized by their AI investments.⁷

It is plausible that the “bubble” is concentrated in the second-tier companies (the “losers” in the winner-takes-most scenario) and the debt instruments backing them, while the equity of the market leaders remains defensible, albeit expensive. The market may be correctly pricing the dominance of these firms, even if it is mispricing the risk of the broader ecosystem.

7. Geopolitics and Regulation: The External Pricking Mechanisms

A bubble, no matter how inflated, requires a needle to burst. In late 2025, geopolitical friction and regulatory intervention serve as the most likely catalysts for a reversal in market sentiment.

7.1 The Tariff and Trade War Context

The 2025 market is navigating a treacherous geopolitical landscape. “Trade policy” and “tariffs” are frequently cited by investors as primary sources of economic uncertainty.³⁵ The fracturing of the global semiconductor supply chain, driven by U.S. restrictions on chip exports to China and reciprocal measures, threatens to disrupt the flow of critical components.

If U.S. tech firms are cut off from the Chinese market—which remains a significant source of revenue for chipmakers—or if tariffs increase the cost of imported hardware, the Return on Invested Capital (ROIC) for AI projects will plummet. The current valuation models generally assume a frictionless global market for AI services; a “Splinternet” reality would require a massive repricing of these assets.

7.2 Sovereign AI and Regulatory Headwinds

On the flip side, governments are treating AI as a matter of national security, leading to the rise of “Sovereign AI” investments.³⁷ While this ensures a baseline of demand (as governments are less price-sensitive than corporations), it also invites heavy-handed regulation.

The proliferation of AI-generated misinformation and “slop” ³⁸ has led to calls for strict regulatory caps on deployment and liability frameworks. If regulations slow the rollout of AI agents in critical sectors like healthcare and finance, the revenue growth needed to fill the “Sequoia Gap” will be delayed, potentially causing the debt-laden financing structures to collapse under the weight of their own interest payments.

8. Conclusion: Verdict on the Oaktree Memo

Based on an exhaustive analysis of the financial, economic, and technical data available in December 2025, this report agrees with the core analytical framework and conclusions presented in Howard Marks’ memo “Is It a Bubble?”, while offering specific nuances regarding the nature of the risk.

Areas of Concurrence

1. The Credit Bubble is the Critical Risk: We fully endorse Marks’ warning regarding “aggressive debt financing.” The migration of risk from equity to credit—manifested in the surge of corporate bond issuance, the widening spreads of tech debt, and the proliferation of precarious ABS structures—is the most dangerous signal in the market. The financialization of rapidly depreciating hardware assets (GPUs) creates a systemic fragility that mirrors the subprime crisis, albeit on a smaller scale.

2. Valuations Pricing in Perfection: With a Buffett Indicator at ~224% and a Shiller P/E near 40, the market is historically stretched. The assumption of a “limitless” future has decoupled asset prices from rational, risk-adjusted earning power.

3. The “Installation Phase” Overbuild: The physical evidence of the data center construction boom, juxtaposed with the “Sequoia Gap” in revenue, confirms that the industry is in a classic speculative overbuild phase. Much of the capital currently being deployed will likely be written down or destroyed.

Areas of Nuance and Divergence

1. The Durability of the Incumbents: Unlike the fragility of the dot-com leaders (many of whom had no revenue), the “Magnificent Seven” in 2025 are fundamentally robust businesses with massive cash moats. A bursting of this bubble is likely to result in a painful compression of valuation multiples (e.g., a 30-40% correction) rather than a total wipeout of equity value for these specific firms.

2. The Productivity Floor: The tangible contribution of AI to GDP (1.1%) and the potential for long-term productivity gains suggest that the technology is not “vaporware.” The economic utility provides a floor to the bubble that did not exist in purely speculative manias like the 17th-century Tulip craze.

Final Verdict

The analysis confirms that the AI sector in late 2025 exhibits the classic hallmarks of an “Inflection Bubble.” The technology is real and transformative, but the financial mechanisms used to fund its deployment have become speculative and dangerous.

Investors should heed Marks’ counsel for “prudence and selectivity.” The risks are not evenly distributed; they are concentrated in the credit markets, the hardware supply chain, and the secondary equity players who lack the balance sheet fortitude to survive a “capex winter.” The market is not necessarily facing a zero-sum collapse, but it is facing a reckoning where the cost of capital will exceed the return on capital for a significant portion of the current AI build-out.

Oaktree’s analysis stands as a necessary corrective to the prevailing euphoria, identifying the structural cracks in the foundation of the AI cathedral.

Works cited

1. Is It a Bubble? - Oaktree Capital Management, accessed December 10, 2025, https://www.oaktreecapital.com/insights/memo-podcast/is-it-a-bubble

2. Memos from Howard Marks: On Bubble Watch | Brookfield Oaktree Wealth Solutions, accessed December 10, 2025, https://www.brookfieldoaktree.com/insight/memos-howard-marks-bubble-watch

3. On Bubble Watch - Oaktree Capital Management - Commentaries - Advisor Perspectives, accessed December 10, 2025, https://www.advisorperspectives.com/commentaries/2025/01/11/on-bubble-watch

4. AI bubble or euphoria? Howard Marks warns you not to bet the house on future, accessed December 10, 2025, https://m.economictimes.com/markets/stocks/news/ai-bubble-or-euphoria-howard-marks-warns-you-not-to-bet-the-house-on-future/articleshow/125890304.cms

5. S&P 500 now a 10-stock show — market concentration hits record 42%, smashing Dot-Com peak - The Economic Times, accessed December 10, 2025, https://m.economictimes.com/news/international/us/sp-500-now-a-10-stock-show-market-concentration-hits-record-42-smashing-dot-com-peak/articleshow/125113020.cms

6. Is It a Bubble? - Oaktree Capital Management, accessed December 10, 2025, https://www.oaktreecapital.com/insights/memo/is-it-a-bubble

7. Alphabet Announces Third Quarter 2025 Results, accessed December 10, 2025, https://s206.q4cdn.com/479360582/files/doc_financials/2025/q3/2025q3-alphabet-earnings-release.pdf

8. Is AI already driving U.S. growth? | J.P. Morgan Asset Management, accessed December 10, 2025, https://am.jpmorgan.com/us/en/asset-management/adv/insights/market-insights/market-updates/on-the-minds-of-investors/is-ai-already-driving-us-growth/

9. Data Center Risk if AI Promises Fade | S&P Global, accessed December 10, 2025, https://www.spglobal.com/en/research-insights/special-reports/look-forward/data-center-frontiers/data-center-risk-if-ai-promises-fade

10. AI Cloud Firms Face Debt Risks Backed by Nvidia GPUs - Markets.com, accessed December 10, 2025, https://www.markets.com/news/ai-cloud-companies-gpu-backed-debt-risk-3011-en

11. US stock market in 2025: A volatile year led by tech and AI, accessed December 10, 2025, https://m.economictimes.com/markets/stocks/news/us-stock-market-in-2025-a-volatile-year-led-by-tech-and-ai/articleshow/125884649.cms

12. Updates from Two Big Tech Firms Land This Week. What It Could Mean for the AI Trade, accessed December 10, 2025, https://www.investopedia.com/updates-from-two-big-tech-firms-land-this-week-what-it-could-mean-for-the-ai-trade-avgo-orcl-11865254

13. Highest Forward 12-Month P/E Ratio For the S&P 500 in More Than 5 Years, accessed December 10, 2025, https://insight.factset.com/highest-forward-12-month-p/e-ratio-for-the-sp-500-in-more-than-5-years

14. S&P 500 P/E Ratio (Quarterly) - United States - Historical … - YCharts, accessed December 10, 2025, https://ycharts.com/indicators/sp_500_pe_ratio

15. US - S&P 500 - Forward PE Ratio | Series - MacroMicro, accessed December 10, 2025, https://en.macromicro.me/series/20052/sp500-forward-pe-ratio

16. Nasdaq PE Ratio Trends - YCharts, accessed December 10, 2025, https://ycharts.com/companies/NDAQ/pe_ratio

17. The ‘Mag 7’ stocks these top fund managers own | Trustnet, accessed December 10, 2025, https://www.trustnet.com/news/13461013/the-mag-7-stocks-these-top-fund-managers-own

18. S&P 500 Shiller CAPE Ratio (Monthly) - United States - Hist… - YCharts, accessed December 10, 2025, https://ycharts.com/indicators/cyclically_adjusted_pe_ratio

19. US - Buffett Indicator | US Stock Market | Collection - MacroMicro, accessed December 10, 2025, https://en.macromicro.me/collections/34/us-stock-relative/406/us-buffet-index-gspc

20. AI: Stairway to Heaven or Heartbreaker? - Charles Schwab, accessed December 10, 2025, https://www.schwab.com/learn/story/ai-stairway-to-heaven-or-heartbreaker

21. Charts of the week: Artificial Intelligence is Leading a CapEx Boom, accessed December 10, 2025, https://millcreek.com/perspectives/charts-of-the-week-artificial-intelligence-is-leading-a-capex-boom/

22. Big Tech’s $405B Bet: Why AI Stocks Are Set Up for a Strong 2026 - IO Fund, accessed December 10, 2025, https://io-fund.com/ai-stocks/ai-platforms/big-techs-405b-bet

23. AI’s $600B Question - Sequoia Capital, accessed December 10, 2025, https://sequoiacap.com/article/ais-600b-question/

24. AI factories face a long payback period but trillions in upside - SiliconANGLE, accessed December 10, 2025, https://siliconangle.com/2025/11/09/ai-factories-face-long-payback-period-trillions-upside/

25. Alphabet plans $15 billion US bond sale, €6.5 billion in Europe - Investing.com, accessed December 10, 2025, https://www.investing.com/news/stock-market-news/alphabet-plans-15-billion-us-bond-sale-65-billion-in-europe-93CH-4327932

26. 3 analysts on why AI debt isn’t a bubble - Quartz, accessed December 10, 2025, https://qz.com/ai-bubble-debt-market-corporate-bonds-oracle

27. Rating the Cloud: Evolving Criteria for Data Center ABS - Linklaters, accessed December 10, 2025, https://www.linklaters.com/en/knowledge/publications/alerts-newsletters-and-guides/2025/october/17/rating-the-cloud-evolving-criteria-for-data-center-abs

28. Why I don’t worry (as much) about big tech’s depreciation schedule - MBI Deep Dives, accessed December 10, 2025, https://www.mbi-deepdives.com/why-i-dont-worry-as-much-about-big-techs-depreciation-schedule/

29. Lifespan of AI Chips: The $300 Billion Question - CITP Blog, accessed December 10, 2025, https://blog.citp.princeton.edu/2025/10/15/lifespan-of-ai-chips-the-300-billion-question/

30. Chill Factor: NVIDIA Blackwell Platform Boosts Water Efficiency by Over 300x, accessed December 10, 2025, https://blogs.nvidia.com/blog/blackwell-platform-water-efficiency-liquid-cooling-data-centers-ai-factories/

31. Data Center Cost in 2025-2026? Liquid Cooling Drives Budgets Even Higher, accessed December 10, 2025, https://datacenter-forum.ro/en/data-center-cost-in-2025-2026-liquid-cooling-drives-budgets-even-higher/

32. Generational Growth AI, data centers and the coming US power demand surge - Goldman Sachs, accessed December 10, 2025, https://www.goldmansachs.com/pdfs/insights/pages/generational-growth-ai-data-centers-and-the-coming-us-power-surge/report.pdf

33. What Is the US Economy’s Potential Growth Rate? | Goldman Sachs, accessed December 10, 2025, https://www.goldmansachs.com/insights/articles/what-is-the-us-economys-potential-growth-rate

34. December 5, 2025 — In The Know With Cathie Wood - Ark Invest, accessed December 10, 2025, https://www.ark-invest.com/videos/market-commentary/december-2025-in-the-know-cathie-wood

35. The (Not So) Magnificent Seven? | Charles Schwab, accessed December 10, 2025, https://www.schwab.com/learn/story/not-so-magnificent-seven

36. CoBank releases 2026 year ahead report – forces that will shape the US rural economy, accessed December 10, 2025, https://markets.businessinsider.com/news/stocks/cobank-releases-2026-year-ahead-report-forces-that-will-shape-the-us-rural-economy-1035634135

37. Global AI Spending November 2025: $400B Annual Investment Meets Investor Reality Check - Lunabase.ai, accessed December 10, 2025, https://lunabase.ai/blog/global-ai-spending-november-2025-400-b-annual-investment-meets-investor-reality-check

38. AI researchers are to blame for serving up slop, accessed December 10, 2025, https://www.theguardian.com/technology/2025/dec/09/ai-researchers-are-to-blame-for-serving-up-slop

While the tools and methods may differ, the fundamental dynamic remains the same. The Church promised eternal life through spiritual salvation, while AI promises a future where technology can...

11 SEPTEMBER 2024

The Analogy Between Churches Accumulating Wealth and AI's Billion-Dollar Investments: Promises of Transcendence

Read full story