AI creates the strongest moat in history? JP Morgan Asset Management: Two medium-term risks hide a liquidation crisis

If we were to summarize the current U.S. stock market in one sentence, it would be: AI is "smothering" everything.

According to the Wind Trading Desk, JPMorgan Asset Management (JPMAM) pointed out in its 2026 outlook report "Smothering Heights," released on January 1, that since the launch of ChatGPT in 2022, 65% to 75% of the returns, profit growth, and capital expenditures in the S&P 500 index have come solely from 42 companies related to generative AI. Without these 42 companies, the performance of the U.S. stock market is even worse than that of Europe, Japan, and China. More astonishingly, capital expenditures in the tech sector contributed 40%-45% to U.S. GDP growth over the past three quarters, while this proportion was less than 5% in the first three quarters of 2023.

This is an unprecedented gamble. The "moat" formed by NVIDIA's chip design, TSMC's manufacturing, and ASML's lithography seems unassailable. The market capitalization of just four semiconductor companies and four hyperscale cloud providers has ballooned from $3 trillion seven years ago to $18 trillion today, accounting for 16% of the global stock market capitalization. However, when the market is extremely concentrated and at historical highs, the question investors must ask is no longer "What other positives are there?" but rather "What could go wrong?"

JPMorgan Asset Management explicitly pointed out two mid-term risks that could trigger a market reckoning in the report: First, the profit realization crisis following massive capital expenditures (similar to a "metaverse" collapse), and second, the physical bottlenecks in U.S. power infrastructure. For investors, 2026 may replay the script of 2025: after experiencing a 10%-15% pullback due to profit taking and growth panic, the market will ultimately close higher by the end of the year, but before that, the questioning of AI profitability and energy supply will hang over like the sword of Damocles.

Risk One: The "Metaverse Moment" of Trillion-Dollar Capital Expenditures — Where Are the Returns?

This is the most pressing threat to the current "moat." Since the fourth quarter of 2022, the four major cloud providers (Microsoft, Alphabet, Amazon, Meta) have poured $1.3 trillion into capital expenditures and R&D, most of which is related to generative AI. The core issue is: can these massive investments be converted into corresponding profits? If not, these tech giants may face a "metaverse reckoning" similar to that of 2022, when the stock prices of the "Magnificent Seven" (Mag7) were halved.

The report points out that although the adoption rate of AI by enterprises is rising, the debate over return on investment (ROI) is extremely heated. Research from the Massachusetts Institute of Technology (MIT) shows that despite companies investing $30-40 billion, 95% of projects yield zero returns, and CEOs' confidence in AI strategies has significantly declined. While Goldman Sachs expects AI to boost productivity, currently, apart from infrastructure providers, very few companies have achieved significant excess returns through AI.

Moreover, there are also concerns about the financial health of cloud providers; in terms of profit transparency, aside from Microsoft, which has disclosed clear AI revenue, the profit paths of other giants remain unclear As free cash flow profit margins gradually decline and cash reserves decrease, the market is eagerly anticipating a clearer profit model. Meanwhile, some cloud vendors are optimizing their financial statements by extending the depreciation period of GPUs and network equipment—from 3 to 4 years to 5 to 6 years. If the depreciation period returns to normal due to adjustments in accounting standards or accelerated iterations of new-generation chips, the earnings per share and profit margins of related companies may face downward pressure of 6% to 8%.

As a key driving force behind artificial intelligence narratives, OpenAI also harbors risks. Despite its rapid revenue growth, the company needs to secure an additional 30 gigawatts of power supply to achieve its 2030 goals, and currently, 72% of GPT queries are still related to non-commercial scenarios. These factors make OpenAI a potentially more concentrated single enterprise risk point than NVIDIA, further highlighting the sustainability issues that need to be addressed behind the fervent investments in the AI field.

Risk Two: Hard Constraints of the Physical World—The Imminent Power Shortage in the U.S.

As OpenAI CEO Sam Altman stated, "The cost of AI will ultimately converge on energy costs." The report warns that physical world limitations are becoming the biggest bottleneck for AI development.

Data centers currently account for only 4%-8% of U.S. electricity demand, but they are expected to occupy two-thirds of future load growth. Due to the exponential growth in computing power demand offsetting improvements in chip energy efficiency, the power supply is facing severe challenges.

This challenge is first reflected in the huge supply-demand gap. Just the four partnerships announced by OpenAI require an additional 30.5 gigawatts of new power, equivalent to 75% of the peak capacity of new nuclear power added in the U.S. over the past five years. Meanwhile, the construction of critical infrastructure is severely lagging. The delivery times for core equipment such as gas turbines and transformers have reached three to seven years, with costs continuously rising, while the median waiting time for grid interconnection has exceeded 70 months.

Regional power crises have also emerged. In the data center-intensive PJM grid region of the U.S., electricity capacity prices are skyrocketing due to the accelerated retirement of fossil fuel power plants and the surge in data center loads. California has even seen data centers being forced to remain idle after construction due to a lack of available power. Although solar energy and battery storage are seen as part of the solution, considering the stability of supply and overall costs, it is expected that 60% of new data centers will still rely on natural gas power generation in the future. This means that the pace of AI expansion will no longer depend solely on the efficiency of algorithms and code, but will be more significantly constrained by the carrying capacity of the aging U.S. power grid and the construction speed of traditional energy infrastructure such as natural gas pipelines.

While solar and batteries are one of the solutions, considering stability and costs, 60% of new data centers will still rely on natural gas power generation in the future. This means that the pace of AI expansion will no longer depend on the efficiency of code, but rather on the carrying capacity of the aging U.S. power grid and the speed of natural gas pipeline construction.

Market Valuation and Debt Concerns: More Than Just Price-to-Earnings Ratio

Despite facing the aforementioned risks, the report believes that while current technology stock valuations are high, they have not reached the frenzied levels of the 2000 internet bubble. The current PEG (Price/Earnings to Growth ratio) is only 1-3 times, far below the 4-8 times of that era Currently, the profit margins of technology giants are extremely high, which is fundamentally different from the "young and unprofitable companies" (YUCs) that once flooded the market.

However, the subtle changes in financing structures are worth noting. Although the net debt levels of AI giants are very low, some companies have begun to utilize the debt market to finance data centers. For example, Oracle, despite having weaker cash flow compared to other giants, has borrowed heavily to meet OpenAI's computing power demands; Meta is constructing data centers through off-balance-sheet financing tools in partnership with Blue Owl. Although Standard & Poor's has maintained its rating, if these hidden debts are consolidated, Meta's leverage ratio will significantly increase. This indicates that the AI arms race is quietly shifting from "cash flow-driven" to "debt-driven," which undoubtedly increases systemic fragility