AI’s Insatiable Appetite for Capital, Energy, and Data: Assessing Bubble Risks and Systemic Implications

Executive Summary

The artificial intelligence industry stands at an inflection point, characterized by extraordinary capital commitments, escalating resource consumption, and mounting valuation pressures that warrant serious examination of bubble dynamics from economic, social, and political perspectives.

Global AI infrastructure investment is projected to reach $1 trillion by 2030, with the top 11 cloud providers alone committing $390 billion in 2025 alone—a 67% increase over 2024 levels. This capital intensity now approaches 30% of hyperscaler revenues, roughly triple historic norms, even as payback periods extend to a decade or longer.

Simultaneously, environmental concerns intensify: by 2030, US AI data centers could generate 24-44 million metric tons of annual carbon dioxide emissions (equivalent to adding 5-10 million vehicles to roadways) while consuming 731-1,125 million cubic meters of water annually.

From financial markets to geopolitical competition to labor market disruption, the AI sector exhibits characteristics of both genuine technological transformation and speculative excess, creating risks that warrant careful monitoring across multiple dimensions.

Introduction

Part I: The Economic Dimension—Capital Intensity, Payback Periods, and Financing Fragility

The Scale of AI Capital Expenditure

The magnitude of AI infrastructure investment has reached dimensions rarely seen in industrial history.

The collective capital commitment announced by major hyperscalers—Alphabet, Meta, Microsoft, Amazon, Apple, and other cloud providers—totals approximately $400 billion for 2025 alone, representing a 67% increase from 2024.

McKinsey projects cumulative data center capital expenditure will reach $5.2 trillion through 2030, while the broader ecosystem (power infrastructure, transmission, networking, and ancillary systems) will require $7 trillion or more.

JPMorgan Chase projects that financing the AI boom will require $1.5 trillion in investment-grade bonds alone over the next five years, with total financing needs across all capital markets potentially exceeding $5 trillion by 2028.

This investment intensity represents something approaching 1.3% of global GDP currently, though still below the late 1990s tech boom (which reached approximately 2% of GDP) and substantially below the late 19th-century railroad boom (which exceeded 3% at peaks).

However, the critical distinction lies not in aggregate size but in temporal concentration and financing structures.

The infrastructure being deployed must be constructed and become operationally productive within an extremely compressed timeframe—typically three to five years—to avoid technological obsolescence from rapid AI capability advancement.

The Capex-to-Revenue Problem

One of the most significant economic warning signals involves the growing gap between capital expenditure and current revenue generation.

OpenAI, widely regarded as the flagship AI company and primary beneficiary of investment, generated $4.3 billion in revenue during the first half of 2025 while operating at a $13.5 billion loss, representing a loss-to-revenue ratio of 314%—essentially the company loses $3.14 for every dollar of revenue.

This loss structure is unsustainable absent projected dramatic revenue acceleration or continued extraordinarily high funding.

More broadly, AI capex intensity among hyperscalers has reached approximately 30% of revenues, roughly triple historical norms.

This reflects a fundamental economic imbalance: companies are consuming capital at rates that assume dramatic future revenue growth that has not yet materialized.

In contrast, Apple, the most capital-efficient of the major technology companies, maintains capex at approximately 2% of revenue.

The differential between Apple’s capital intensity and hyperscaler AI capex intensity reflects two competing models: Apple’s approach of renting compute from cloud providers versus hyperscalers’ strategy of self-constructing owned infrastructure at scale.

The payback period analysis is particularly illuminating. Analysis of AI factory capex suggests breakeven scenarios ranging from 8 to 25+ years depending on productivity gains realization.

In accelerated scenarios (probability-weighted at 25%), payback might occur in 8 years with 35% annual revenue growth.

In base case scenarios (40% probability), breakeven stretches to 12 years. In delayed adoption scenarios (25% probability), payback extends to 18 years.

Worst-case scenarios (10% probability) suggest 25-year payback periods with only 8% annual growth assumptions.

Given that most technology investments typically target 3-5 year payback periods, these timescales indicate extraordinary investment in speculative future productivity.

The Debt Financing Pivot: From Equity to Leverage

Critically, the financing structure of the AI boom is undergoing a historic transformation. For the first two years of the generative AI explosion, investment was predominantly equity-financed through corporate cash flows, venture capital, and private equity.

However, as capex requirements have accelerated beyond internal cash generation capacity, companies have increasingly turned to debt financing.

In Q3 and Q4 2025, debt issuance for AI infrastructure exploded. Aggregate borrowing in September and October 2025 reached $75 billion in bonds and loans—far exceeding historical quarterly patterns.

Bank of America estimates that private credit loans related to AI may have nearly doubled in the 12 months through early 2025, with the private credit sector now accounting for increasingly significant portions of data center financing.

Morgan Stanley projects that private credit markets could account for over 50% of the $1.5 trillion required for data center expansion through 2028.

Most concerning from financial stability perspectives is the emergence of complex debt structures including asset-backed securities (ABS), securitized data center rental payments, and opaque financing arrangements involving special purpose vehicles and circular funding loops.

Approximately $130 billion in ABS linked to data centers was issued across 27 transactions in 2025, marking a 55% increase from 2024 and climbing toward an estimated $420 billion by year-end.

These instruments repackage data center rental payments into tradable securities—essentially betting that major technology companies will maintain rental obligations on capacity used for AI infrastructure.

As economist Markus Brunnermeier noted in the Center for Economic and Policy Research analysis, “if AI investment drives risk appetite elsewhere in the system, it can lead to Minsky-type investment boom-to-bust cycles in GDP and crises.”

The critical vulnerability emerges when “bank lending joins the funding cycle. At that point, the AI bubble ceases to be a matter for investors alone and becomes a genuine policy concern.”

Early evidence suggests this transition is beginning: major banks including JPMorgan Chase and Goldman Sachs have substantially increased lending to data center operators and AI companies.

Valuation Metrics and Historical Comparison

The current AI market exhibits a complex valuation profile that simultaneously resembles both the dot-com bubble and represents genuinely justified growth.

Unlike the dot-com era, current valuations are driven primarily by actual earnings growth rather than pure multiple expansion.

Technology company earnings growth has accelerated dramatically to 45% year-over-year, compared to 15% during the dot-com peak.

This earnings growth provides some fundamental justification for elevated valuations.

However, concerning signals are simultaneously evident.

The market has become severely concentrated, with the “Magnificent Seven” AI-related stocks accounting for 75% of S&P 500 returns, 80% of earnings growth, and 90% of capital spending growth since ChatGPT’s launch in November 2022.

This concentration exceeds dot-com era patterns and creates what financial regulators term “concentrated systemic risk.”

Palo Alto Networks’ market capitalization has reached $145 billion—larger than Kazakhstan’s entire GDP—despite cybersecurity representing a highly specialized domain.

Goldman Sachs strategists have warned that AI equity valuations resemble 1997 levels in the dot-com cycle—several years before the 2000 peak and subsequent bust.

This temporal positioning suggests that valuations could advance further before any correction, but also warns that inflection points often arrive with minimal warning.

The consensus among major financial institutions has shifted markedly: CEO Jamie Dimon of JPMorgan Chase, Bank of England Governor Andrew Bailey, and venture capital leaders are increasingly articulating concerns that “a lot of assets look like they’re entering bubble territory.”

Part II: The Environmental Dimension—Sustainability Crisis at Scale

Projected Carbon and Water Footprint

Recent research from Cornell University has provided detailed quantification of environmental impacts from AI infrastructure expansion.

The analysis, published in Nature Sustainability in November 2025, projects that electricity consumption for AI will surge by 7 to 17 times between 2024 and 2030, with carbon emissions potentially increasing by 24 to 44 times and water usage rising 6 to 13 times.

Specifically, the study forecasts that US AI data centers alone could emit between 24 million and 44 million metric tons of carbon dioxide annually by 2030—equivalent to the emissions from 5-10 million gasoline-powered vehicles.

Water usage could reach between 731 million and 1,125 million cubic meters annually, sufficient to fill 300,000 to 500,000 Olympic-sized swimming pools.

In regional hotspots, the impact is even more acute: Phoenix, Arizona—a booming data center hub—faces a projected 400% increase in water usage from data center electricity generation by 2030.

In Texas, data center water consumption alone is projected to rise from approximately 100 billion gallons in 2025 to 399 billion gallons by 2030, potentially accounting for 6.6% of statewide water usage.

These projections assume continued growth in AI server deployment at current trajectories.

Efficiency improvements could reduce impacts—best-case scenarios incorporating advanced siting strategies, grid decarbonization, and operational efficiency could theoretically reduce carbon and water footprints by 73% and 86% respectively compared to worst-case trajectories.

However, as Nature research emphasized, these reductions require “unprecedented reliance” on carbon removal, water restoration, and complex long-term offset mechanisms.

The paper notes that without additional coordinated interventions, “AI data centres are likely to generate substantial environmental impacts in the coming years,” requiring companies to shift toward transparent approaches involving third-party verification and governmental cooperation.

The Grid Decarbonization Prerequisite

The environmental sustainability of the AI boom depends entirely on parallel and accelerated grid decarbonization.

Yet actual decarbonization progress remains insufficient to support the projected energy demands.

The US Energy Information Administration projects grid decarbonization at rates substantially slower than would be required to offset AI-driven electricity demand increases.

Federal data center siting decisions have exacerbated this problem: despite Nebraska’s vast wind energy potential, state utilities have directed investments toward natural gas generation, while other states have approved data center development in regions lacking adequate renewable generation capacity.

This creates a fundamental mismatch between corporate sustainability commitments and operational reality.

Google, despite holding the world’s largest portfolio of renewable energy contracts, has seen its overall carbon emissions increase by 48% since 2019 (largely due to AI and data centers) while replenishing only 18% of the water it consumed.

Microsoft reported a 34% increase in water consumption year-over-year to 6.4 million cubic meters globally.

These numbers contradict corporate net-zero pledges and demonstrate that aggregate consumption growth is outpacing efficiency improvements and renewable sourcing.

The Federation of American Scientists estimates that companies’ true carbon footprints may be 662% higher than reported figures due to reliance on outdated measurement standards (Power Usage Effectiveness) and renewable energy credits that mask actual emissions.

This measurement opacity complicates policy evaluation and allows companies to claim net-zero progress while absolute environmental impacts continue escalating.

Part III: The Social Dimension—Labor Disruption and Data Justice

AI Exposure and Labor Market Disruption

The labor market impacts of AI adoption are beginning to manifest across occupational categories with striking correlation between AI exposure and unemployment increases.

Research from the St. Louis Federal Reserve found a 0.47 correlation coefficient between occupational AI exposure and unemployment rate increases between 2022 and 2025.

Computer and mathematical occupations—averaging approximately 80% AI exposure—experienced unemployment increases of 1.2 percentage points, among the steepest declines.

Clerical occupations (75% exposure) saw 1.0 percentage point unemployment increases, while manual labor with low AI applicability (20% exposure) experienced only 0.2 percentage point increases.

Goldman Sachs estimates that AI could displace 6-7% of the US workforce if widely adopted, though the institution emphasizes that displacement should prove temporary as new employment emerges.

However, historical patterns offer limited reassurance.

During the 1980s, automation displaced routine occupations, but each subsequent recession created increasingly prolonged jobless recoveries.

Today, AI is threatening non-routine cognitive occupations—scientists, engineers, designers, lawyers—that were historically immune to automation.

JPMorgan Chase research notes that “workers last employed in non-routine cognitive jobs have always accounted for the smallest share of the unemployed, until recently. This changing pattern might be indicative of rising unemployment risk for these workers going forward.”

Critically, the Brookings Institution’s analysis using labor market data from November 2022 through November 2025 found evidence that “occupations that embraced generative AI most intensively showed the largest unemployment gains, with a correlation coefficient of 0.57.”

While Yale’s Budget Lab found remarkable labor market stability in broad occupational categories, the St. Louis Fed’s analysis revealed clear occupational-level divergence, suggesting that concentrated sectoral disruption may be occurring while aggregate statistics mask underlying reallocation.

Unequal Distribution of Productivity Gains

A fundamental economic concern involves the distribution of productivity gains from AI deployment.

McKinsey research suggests approximately 75% of AI value accrues to four areas: customer operations, marketing and sales, software engineering, and R&D.

These are precisely the domains where high-skill, high-wage employment predominates. Meanwhile, workers displaced from routine occupations typically face transitions to lower-skill, lower-wage employment with less job security.

MIT’s finding that 95% of corporate AI projects fail to deliver measurable return on investment within six-month to one-year pilot periods suggests that widespread productivity gains remain speculative.

Companies frequently delay replacing departing staff members, which creates value through avoided hiring and training costs, but this “value” isn’t recorded as positive ROI in short-term pilot studies.

This measurement issue means that actual productivity benefits may be occurring but systematized evaluation frameworks fail to capture them—leaving companies and workers uncertain about whether AI is genuinely improving productivity or merely substituting for human labor.

Data Justice and Extraction Concerns

The training data underpinning generative AI systems represents another critical social and ethical concern.

Most large language models are trained on internet content—much of which was created by individuals and creators who did not explicitly consent to commercial AI training usage.

Content on Creative Commons licenses, academic publications, and user-generated platforms has been incorporated into commercial models without explicit permission or benefit-sharing.

The relationship between data providers and AI developers is fundamentally extractive: millions contribute creativity and knowledge to the internet commons, while a handful of corporations capture the economic value through proprietary models.

Copyright holders and content creators have increasingly challenged this model.

The US is examining fair use doctrines in AI training contexts, the EU’s AI Act includes transparency requirements regarding training data, but fundamental questions remain unresolved globally: Who owns training data?

How should benefits from AI systems be distributed to creators? What constitutes adequate consent for data usage?

These unresolved questions create ongoing legal and ethical tensions that could trigger retroactive regulatory action affecting AI companies’ training approaches and market valuations.

Part IV: The Political Dimension—Geopolitical Fragmentation and Regulatory Divergence

US-China Competition and Tech Decoupling

The competition for AI dominance has become explicitly framed as a matter of national security and geopolitical power.

The United States has pursued an aggressive strategy of “tech decoupling,” systematically tightening export controls on advanced semiconductors and AI infrastructure to China.

By mid-2025, the US had banned even specialized AI chips designed to meet earlier export rules, effectively closing remaining chokepoints for advanced GPU access.[weforum]

China has countered with a multipronged strategy emphasizing bilateral engagement with Global South countries, providing cost-effective AI solutions and research partnerships that promote technological reliance on Chinese approaches.

The “Digital Silk Road” represents China’s effort to build coalitions with developing nations, establishing technological frameworks that compete with US-designed systems.

This geopolitical bifurcation creates enormous inefficiencies. Technology companies must increasingly navigate fractured supply chains, maintain separated R&D infrastructure, and optimize for competing standards.

Rather than converging on globally optimal technology architectures, investment is diverted toward redundant infrastructure designed to avoid technological reliance on geopolitical adversaries.

From an economic perspective, this fragmentation represents pure value destruction—capital committed to solving geopolitical problems rather than technological problems produces no productivity benefit and increased costs for all participants.

Regulatory Divergence and Governance Fragmentation

The US and EU have pursued fundamentally different regulatory approaches to AI governance, creating lasting structural divergence.

The EU’s AI Act, enacted in August 2024 and phasing into implementation through 2026, establishes comprehensive, risk-based regulation applicable across all member states with extraterritorial scope.

The framework categorizes AI systems according to risk levels and mandates stringent requirements for high-risk applications, including comprehensive risk management, technical documentation, bias testing, data quality assurance, and transparency measures.

The US, by contrast, has pursued a fragmented, sector-specific approach emphasizing innovation and minimal regulatory constraint.

Executive policy guidance explicitly frames regulation as “onerous” and a “barrier” to innovation, with policy agendas focused on “rescinding Biden Executive Order 14110” and reviewing “FTC investigations commenced under the previous administration.”

This reactive deregulation approach means that AI policy can shift dramatically with political administration changes, creating instability and uncertainty for investment planning.

For multinational organizations, this divergence creates a complex compliance challenge.

The pragmatic path forward is to “build to the EU standard and adapt downwards where necessary”—essentially, companies that meet stringent EU requirements can operate anywhere, while companies that meet only US standards may face barriers to EU market access.

This creates a de facto regulatory harmonization around the highest bar, driven by market access requirements rather than deliberate policy coordination.[dataversity]

The lack of regulatory convergence compounds policy challenges. Data governance questions—What constitutes fair use of copyrighted material in AI training?

How should training data be sourced and shared?—have dramatically different answers in different jurisdictions.

Energy efficiency requirements mandated in the EU lack equivalent federal requirements in the US. Transparency mandates in one jurisdiction clash with trade secret protection in another.

This regulatory fragmentation simultaneously protects neither consumers nor workers effectively while creating compliance complexity that favors large, well-resourced companies capable of navigating multiple regulatory regimes.

Smaller companies and startups face disproportionate compliance burdens, potentially accelerating market consolidation toward the largest technology companies.

Part V: Bubble Assessment and Systemic Risk Evaluation

Bubble Characteristics: Present but Not Definitively Conclusive

Assessing whether the AI sector constitutes a bubble requires careful calibration of multiple indicators.

Historically, financial bubbles are characterized by asset valuations fundamentally disconnected from revenue generation and cash flow metrics, speculative excess driven by momentum rather than fundamentals, and widespread adoption of leverage to amplify gains.

The current environment exhibits some but not all classic bubble characteristics:

Bubble-like Indicators

Extreme concentration

54% of global investors believe AI stocks are in a bubble.

Valuation disconnects

Private AI companies (Palantir at 700x earnings) and startups exhibit valuations bearing minimal relationship to revenue or profitability

Rapid multiple expansion in secondary markets: venture capital valuations for AI startups have accelerated despite early-stage revenue that barely materializes

Emerging leverage dynamics

the shift from equity to debt financing introduces fragility that wasn’t present in equity-funded boom periods

Factors Moderating Bubble Concerns

Earnings growth fundamentals: Unlike dot-com, current valuations are driven by actual earnings growth (45% vs. 15% dot-com era) rather than pure multiple expansion

Large-cap tech valuations remain reasonable: Nvidia at multiples closer to historical norms when earnings growth is factored in

Technology adoption curves validate AI productivity: enterprise AI adoption is occurring at meaningful scale rather than remaining speculative

Critically Uncertain Factors

Future payback period realization

Whether 8-25 year payback assumptions prove accurate or optimistic depends entirely on AI productivity gains that have not yet materialized

Capital market sentiment

Any major disappointment regarding AI productivity could trigger rapid sentiment reversal given the momentum-driven nature of recent gains

Financial Stability and Contagion Risks

The Bank of England has specifically warned about financial stability risks from AI-driven valuations.

The central bank identified “concentrated risk” from heavy reliance on a small number of companies and “model homogeneity” risks where similar AI systems deployed across financial institutions create correlated behaviors during market stress.

During financial dislocations, synchronized algorithmic trading and AI-driven responses could amplify volatility rather than dampening it.

Additionally, the circular nature of AI financing creates interconnection risks: venture capital funds invest in AI startups, technology companies invest in AI infrastructure, infrastructure companies (CoreWeave, etc.) raise capital from financial markets backed by technology company commitments.

If any node of this circular flow experiences disruption, pressure propagates throughout.

For example, if major technology companies delay capex due to disappointing productivity realizations, infrastructure companies would immediately face revenue pressure and potential loan defaults, potentially triggering financial market contagion.

CEPR research notes that while equity-financed bubbles generate limited systemic risk because equity losses are absorbed by investors, the transition to debt financing creates genuine policy concerns: “Although that was initially the case [equity financing], the situation is changing.

AI firms are increasingly relying on debt-based and circular financing structures.

Circular funding loops of this nature imply that the same capital appears several times on different balance sheets. The consequence is hidden fragility, misleading investors, creditors and regulators.”

Part VI: Comparative Framework—AI Boom vs. Historical Precedents

Why This Differs From Dot-Com (But Not Entirely)

The AI boom exhibits important distinctions from the dot-com era that warrant attention. During the late 1990s, fraudulent accounting (WorldCom’s $11 billion fraud being perhaps the most extreme example) inflated perceived internet demand.

Modern audit standards and regulatory transparency have improved, reducing (though not eliminating) accounting manipulation.

Additionally, dot-com involved forced upgrade cycles (Y2K) and consumer adoption of novel internet services, both of which have limited parallels in current AI deployment, which targets productivity gains among existing business operations.

However, parallels are significant and concerning.

Both eras exhibit sector concentration, with a small number of stocks driving majority market gains.

Both feature valuations advancing faster than earnings growth in earlier phases (though the current cycle shows stronger earnings fundamentals).

Both attracted extraordinary capital commitments into infrastructure believed necessary to support future growth but increasingly speculative regarding actual demand.

Both involved geopolitical narratives about national competitiveness driving policy support for investment.

And most critically, both feature extended payback periods where investors must assume productivity gains will eventually justify current capital commitments.

The key distinction

the dot-com bubble was primarily a multiple-expansion phenomenon (valuations rising relative to earnings), while the AI cycle has been primarily an earnings-growth phenomenon (valuations rising proportionally to robust earnings growth).

This suggests underlying business models may be more sustainable than dot-com companies.

However, it simultaneously suggests that any disappointing deceleration in earnings growth could trigger dramatic multiple compression, as valuations would lose both the earnings growth foundation and momentum dynamics.

Part VII: Key Risk Indicators and Monitoring Framework

Research from Brookings Institution and other institutions has identified six critical indicators for monitoring AI bubble risks.

Capital Efficiency and Payback Periods

Tracking whether actual payback periods move toward base-case assumptions (12 years) or extend toward worst-case scenarios (18-25 years).

Lengthening payback periods signal deteriorating business case fundamentals.

Semiconductor Supply Constraints

Kearney’s 2025 State of Semiconductors report found that 42% of semiconductor leaders expect advanced-node shortages and only 65% expressed confidence in supply chain resilience (down from 82% the prior year).

Persistent supply constraints that prevent capex deployment would suggest either that demand assumptions were incorrect or that execution risks are larger than anticipated.

AI Adoption Rates and Productivity Realization

Large-scale real-world validation of AI productivity gains would support valuations.

Continued pilot studies showing 95% failure rates to deliver ROI would signal that expected productivity gains are not materializing.

Debt Market Pricing

Monitoring the cost of debt issuance for AI infrastructure is critical.

Rising spreads (the premium over risk-free rates) would indicate investor concerns about repayment capacity.

Conversely, continued tight spreads despite elevated leverage would indicate continued speculative appetite.

Venture Capital Valuations for Early-Stage AI Companies

While late-stage technology company valuations remain grounded in earnings growth, early-stage AI startup valuations (where we see 700x earnings multiples) provide leading indicators for speculative excess.

Deterioration in early-stage valuations would precede late-stage corrections.[reuters]

Financial System Interconnectedness

Monitoring the proportion of AI financing flowing through complex structures (ABS, private credit, special purpose vehicles) versus traditional bank lending or corporate cash flow provides signal about financial fragility.

Rapid expansion of opaque financing structures indicates increasing leverage and hidden fragility.

Conclusion

Innovation, Exuberance, and Systemic Vulnerability

The AI sector simultaneously embodies both genuine technological transformation and financial dynamics that exhibit bubble characteristics.

Capital being deployed is real and enormous. Environmental impacts are measurable and concerning. Labor displacement is beginning across occupational categories. Geopolitical competition is intensifying.

And the financing structure is undergoing a fundamental transformation from equity to debt that increases financial fragility.

From an economic standpoint

Current valuations depend entirely on sustained earnings growth and realization of productivity gains that remain speculative.

Extended payback periods (12-25 years) and capex-to-revenue ratios approaching 30% suggest that companies are making extraordinarily aggressive bets about future productivity.

The emergence of complex debt financing structures after a period of equity financing introduces systemic vulnerabilities not present in the early bubble phase.

From an environmental standpoint

Current deployment trajectories are incompatible with corporate net-zero commitments absent unprecedented and currently unobserved grid decarbonization.

Water consumption patterns in water-stressed regions could create genuine resource scarcity constraints on data center expansion.

From a social standpoint

Labor displacement is concentrating in high-skill, high-wage occupations while productivity gains accrue disproportionately to capital rather than labor, threatening to accelerate inequality.

Data justice concerns about extraction and consent remain fundamentally unresolved.

From a political standpoint

Geopolitical fragmentation is creating inefficiencies and reducing overall productivity, while regulatory divergence between the US and EU creates compliance complexity and reduces global standardization benefits.

The critical question is not whether the AI sector will experience corrections—the evidence suggests that some degree of valuation compression is likely as expectations are recalibrated to reality.

The question is whether and how severely this correction might propagate through financial systems.

Given the shift toward debt financing, the concentration of capital flows through a small number of companies, and the emergence of complex financing structures involving asset-backed securities and private credit, the conditions for financial contagion appear increasingly present.

The next 12-24 months will likely prove decisive in determining whether current investment levels produce genuine productivity gains that justify valuations, or whether the AI boom transitions from justified technological transformation into recognized speculative excess requiring significant correction.