A One In A Lifetime Crash Is Coming (3 Warning Signs)

AI, the Stock Market, and Systemic Risk: A Deep Dive into JP Morgan’s “Smothering Heights”

Key Concepts:

• Capital Concentration: The disproportionate contribution of a small number of stocks (specifically 42 in the S&P 500) to overall market returns.
• Energy Risks: The escalating power demands of AI infrastructure and the potential limitations of electricity supply.
• Geopolitical Risks: The dependence on Taiwan for advanced chip manufacturing and China’s efforts to build an independent AI ecosystem.
• Hyperscalers: Large-scale data center operators (Meta, Amazon, Google, Microsoft) driving AI infrastructure investment.
• Metaverse Moment: A scenario where AI hype outpaces actual profitability, leading to a market correction.
• Choke Points: Critical dependencies in the supply chain, like Taiwan’s chip manufacturing, that create systemic vulnerabilities.

I. Market Concentration & The AI Premium

Since the release of ChatGPT in late 2022, a remarkably small segment of the stock market – 42 companies within the S&P 500 – has accounted for 78% of the index’s total returns. These companies fall into three categories: direct AI players (Nvidia, AMD, Google, Palantir), AI-linked utilities (NRG, Vistra), and AI equipment manufacturers (Eaton, Quanta, MCOR). These 42 stocks have experienced 190% growth, 153% earnings growth, and a 68% average increase in capital expenditure (capex) and research & development (R&D) spending.

In contrast, the remaining 458 companies in the S&P 500 have only risen by 26% over the same period, translating to an annual return of under 8%. This underperformance means that most US stocks are lagging behind markets in Europe, Japan, and China.

Eight companies – Nvidia, AMD, Meta Platforms, Amazon, Google, Microsoft, TSMC, and ASML – have collectively grown from a $3 trillion market cap in 2018 to over $18 trillion today, representing approximately 20% of all equity markets globally. This means roughly 20 cents of every dollar invested in stocks worldwide is concentrated in these eight entities.

Despite this concentration, JP Morgan’s paper argues that current AI valuations aren’t necessarily in a bubble, citing higher profit margins and returns on capital compared to the average S&P 500 company, and valuations that are not as extreme as those seen during the dot-com bubble. However, the paper emphasizes the need to focus on potential risks to these AI companies rather than solely on their earnings.

II. The Spending Spree: Capital Expenditure & Debt

Hyperscalers (Meta, Amazon, Google, and Microsoft) have collectively spent approximately $1.3 trillion on AI hardware, R&D, and infrastructure since the end of 2022 – exceeding the combined costs of the Manhattan Project, the Apollo program, and the entire US highway system.

The median S&P 500 company allocates around 10-20% of revenue to capex and R&D, while Meta is currently spending nearly 70% of its revenue on AI. Google, Amazon, and Microsoft are also increasing their ratios.

This spending is increasingly financed through debt. Long-term borrowing has increased tenfold from under $20 billion in 2024 to $200 billion in 2025. However, the lack of corresponding revenue is a concern. Microsoft is the only hyperscaler publicly reporting AI revenue, and it represents a small fraction of their overall costs. An MIT study cited in the paper found that 95% of organizations are seeing no measurable returns from their AI investments. CEO confidence in AI strategy has fallen from 82% in 2024 to 49% in 2025. AI models still exhibit a 20-30% “hallucination” rate, with some models reaching 50% or higher, hindering their adoption for critical tasks.

JP Morgan notes that accounting practices, such as extending the depreciation window for AI servers from 3 to 6 years, can artificially inflate reported earnings by 6-9 percentage points. Free cash flow margins are also declining, contrary to expectations during a “super cycle.”

III. Energy Constraints & The Cost of Intelligence

The AI boom is fundamentally an energy story. OpenAI alone could require approximately 30 gigawatts (GW) of new power generation by 2030 to support its current roadmap – exceeding the entire US grid’s capacity added in 2024. Data centers are projected to drive roughly two-thirds of all new electricity demand.

The US is only adding around 25 GW of grid capacity annually, creating a significant shortfall. Data center buildouts announced by Microsoft and Meta (6-10 GW each) raise questions about their power sources. Currently, natural gas is the primary solution, but supply chain bottlenecks for gas turbines and critical grid equipment (transformers, switches) mean new plants can take 3-7 years to come online.

Solar and battery storage are not currently cost-competitive with natural gas for 24/7 power supply, even with a 40% tax credit. Therefore, access to reliable power, not GPUs or memory, is becoming the primary bottleneck. The cost of intelligence is increasingly tied to the cost of energy, suggesting that the winners will be those with the most efficient chips and the best power deals.

IV. Geopolitical Risks: China & Taiwan

The AI ecosystem faces significant geopolitical risks, primarily centered around China and Taiwan. China is actively building an independent AI stack, investing heavily in innovation, power generation, and domestic chip manufacturing.

Since 2019, China has built 11 times more power generation capacity than the US. They have a $48 billion chip fund and offer subsidies of 20-30% of corporate profits in key AI and semiconductor sectors. Huawei has doubled the yields of its Ascend 910C AI chip and now controls over 75% of China’s AI chip production. A team in Shenzhen has even developed a prototype EUV lithography machine, aiming for production before 2030.

While Nvidia currently outperforms Huawei in chip performance (Blackwell Ultra delivers 2.9x the compute, 2.5x the memory bandwidth, and 2x better power efficiency), the gap is narrowing. Huawei’s latest racks are three times more efficient than Nvidia’s previous generation. China is willing to trade some power efficiency for independence in the world’s largest AI market.

Taiwan’s dominance in advanced chip manufacturing represents a critical vulnerability. TSMC produces over 90% of the world’s most advanced chips. Relocating production to the US is challenging; TSMC’s 5nm wafers made in Arizona have 62% gross margins compared to 8% in Taiwan, and even those chips are shipped back to Taiwan for packaging and testing. By 2030, the US may only be able to produce one-third of the advanced chips it consumes.

Taiwan’s reliance on imports (90% of energy, 66% of food) makes it vulnerable to a blockade. A blockade could cut global economic output by nearly 3% in the first year alone – worse than the 2008 financial crisis.

V. Potential Scenarios & Conclusion

The JP Morgan paper outlines several potential risks that could trigger a market correction:

• A “Metaverse Moment”: AI hype outpacing profitability, leading to a significant drop in AI stock valuations (30-50%).
• Energy Constraints: Inability to secure sufficient power to support AI infrastructure growth.
• Geopolitical Escalation: Conflict over Taiwan disrupting the chip supply chain.

The paper’s base case remains positive, with the US maintaining its lead in AI and continued earnings growth. However, it emphasizes the importance of monitoring these risks and being prepared for potential disruptions. The core message is to base investment decisions on data and a realistic assessment of risks, rather than solely on hype and optimism. The author concludes by reiterating the importance of investing in oneself, particularly through skills development in the rapidly evolving field of AI.