Song Xuetao 2026 US Stock Market Outlook: The Internal Melting Point and External Turning Point of the AI Bubble

In 2025, the U.S. stock market experienced a historic year intertwined with tariff shocks, fiscal shifts, and industrial waves. The "Deepseek Moment" and the "Independence Day Tariff" in April triggered market tremors, but the resilience of the U.S. stock market continues to emerge after the shocks. Since the third quarter, the OBBBA Act and the Federal Reserve's dovish shift have brought benefits on both fiscal and monetary levels. OpenAI announced a series of significant investment agreements with companies such as NVIDIA and Oracle, with the artificial intelligence boom driving market sentiment to new highs.

However, recently, the AI narrative has begun to face skepticism again. Tech giants are "spending at all costs" to ignite a capital expenditure boom—cash flows are shrinking while external financing is increasing. At the same time, the complex relationships of mutual investment, related transactions, circular financing, and interlinked chains have exacerbated the market's irrational exuberance in a self-reinforcing positive feedback loop. Is there really a bubble in AI investments? How can its extent be quantified? Will there be a specific point in 2026 that significantly amplifies the market's vulnerability?

The Existence of a Bubble is Justifiable

Many viewpoints argue that there is currently no bubble in the AI investment field, citing that compared to the numerous unprofitable companies during the 2000 dot-com bubble, today's tech giants have high revenues, healthy cash flows, and acceptable leverage. However, this kind of rigid comparison of indicators overlooks the fundamental differences between entities and the main players.

Today, the scale and concentration of AI investments far exceed those of 2000. The investment scale of AI giants in the economy and the positive externalities they bring are incomparable to those in 2000. This means that if these AI giants encounter problems, the impact on the entire financial and technological ecosystem will be catastrophic, which cannot be measured by simple revenue or valuation metrics. The mechanisms for the formation of each bubble are similar, but the manifestations and carriers of systemic risk are different.

From an industrial perspective, the value of AI in enhancing overall societal productivity will be very gradual. Although AI has some benefits in coding and certain processes within tech companies, its contribution to productivity enhancement will be very limited in the short term for most industries, as the transformation of organizations and processes lags behind the technology itself. Just as in the early days of the electrical revolution in the late 19th century, where electric motors simply replaced steam engines on factory spindles without improving efficiency, if AI is not deeply restructured with the organizational structure, incentive mechanisms, and decision-making processes of enterprises, its value will be wasted. Today, AI cannot complete the decision-making loop; most still "assist human predictions" rather than "replace human decisions," which makes high-level decision-making a bottleneck. Furthermore, the outsourcing benefits of AI for low-skilled positions are limited, and to realize the value of AI, it must replace high-skilled positions, which requires a long time to bridge the "AI gap." However, compared to the distant industrial prospects, investing in AI has become one of the market's consensus. The speed of bubble expansion and when industrial productivity lands are two different matters; we cannot deny the rationality of AI investment simply because the industry needs time. Currently, multiple parties have the motivation to inflate the bubble: technology companies are "all in AI" to avoid being eliminated, financial institutions are profiting from loose liquidity, media outlets are accelerating the bubble through traffic dissemination, and American residents (middle and upper classes) are binding AI stock bubbles through pensions and other means.

Even if the bubble bursts, it is not necessarily a bad thing, as new organizational revolutions often require low-cost soil to nurture. After the internet bubble burst in 2000, new things truly began to grow. The excessive infrastructure brought about by the bubble (such as fiber optics and servers) became cheap after the burst, providing fertile ground and extremely low operating costs for the rise of later internet giants. After the AI bubble bursts, the cheap computing power, electricity, and infrastructure left behind will also become fertile soil for future new business models and small company innovations. After costs are significantly reduced and industry standards are unified, the organizational revolution that reconstructs the operational logic of enterprises around AI can truly unleash productivity.

For the United States, pushing the AI bubble to the end is not only an economic behavior but also tied to national destiny.

On one hand, American household wealth (especially among the top 1% and top 10% of the wealthy) is heavily concentrated in the U.S. stock market, and the prosperity of the U.S. stock market is also the foundation of the dollar's credit. To support the dollar and the massive fiscal deficit, the U.S. has no reason to actively burst the bubble; it must ensure the continued prosperity of the U.S. stock market.

On the other hand, the AI narrative is also a covert way to reduce debt. By financializing AI, raising the value of stocks (such as the seven giants) and assets (like overvalued collateral as underlying assets for credit), it attracts international investors and allied governments to foot the bill. The U.S. "sells cards, sells stocks, sells dreams," transferring debt and risk to allies and international capital, allowing them to bear the construction costs of computing power infrastructure. In the geopolitical game, the U.S. initiates a "land-grabbing movement" through monopolizing ecosystems and consolidating itself, with costs being borne by allies through debt transfer mechanisms. Once the bubble bursts, allies, as the main payers and builders of infrastructure, will allow the U.S. to obtain a large amount of electricity, computing power, and new infrastructure at a lower cost, laying the foundation for future innovations. Foreseeing the outcome of "if the dead friends do not die, the poor will," the U.S. will inevitably push the AI bubble to the end.

For the U.S. stock market, the impact of AI is already self-evident; the core support for valuation is the market's belief that AI technology can shape a bright future comparable to the industrial and information revolutions. However, the "S&P 493 Index" after excluding the Mag7 has seen zero growth for two years, and the suppression of traditional sectors by high interest rates continues. Whether the value of AI can ultimately benefit society as a whole remains unknown and can only be judged retrospectively based on results.

"Iron Chain," Where is the Most Vulnerable Link?

The artificial intelligence industry can be divided into three levels: chip manufacturers, cloud service providers, and model developers. Chip manufacturers provide AI hardware, benefiting first from revenue and having the most abundant cash flow. Cloud service providers offer computing power facilities and services for model development, with costs mainly in hardware procurement and energy consumption, while revenue comes from cloud computing rentals. Model developers focus on AI model development and training, with major expenses in computing resource procurement and revenue from API service subscriptions.

In the past year, multiple cross-level and eye-catching large investment transactions have emerged among companies within these three levels. The market integration of the industrial chain, from chip manufacturing and cloud computing to AI applications, although helpful in integrating industrial chain resources, improving chip supply, computing power support, and application scenarios, has temporarily driven performance and valuation increases and even improved financing capabilities. However, this trend also blurs the boundaries of traditional industries, potentially creating a false sense of demand and leading to vulnerabilities in the industrial chain. If the business profit expectations of AI giants cannot generate sufficient profits and cash flow, and the liquidity environment deteriorates, the entire chain may face significant risks due to "damaged faith."

Currently, there is a serious lack of information disclosure regarding circular investments in related transactions and customer concentration. Giants should be regarded as acting in concert within the complex network of capital and business relationships (such as cross-shareholding, joint investments, strategic cooperation, etc.). In circular investments, inadequately disclosed related relationships may make it difficult for investors to see the real risks, and some revenues may be double-counted, potentially exaggerating the monetization scale of the AI ecosystem, which is only temporarily masked by loose liquidity. At the same time, giants should more clearly disclose their dependence on key major customers; for example, Oracle should explicitly state in its financial report that the surge in its RPO is mainly due to a single contract with OpenAI.

In the ambiguous circular investment model, negative public sentiment related to performance often amplifies market sensitivity. For instance, on October 7, internal documents from Oracle revealed that the gross margin of its cloud business related to NVIDIA was only 14% (overall gross margin about 70%), raising market concerns about its severe dependence on a few major customers and weak bargaining power, causing Oracle's stock price to drop by as much as 7.1% during the day. Additionally, Microsoft's third-quarter report showed a $3.1 billion loss from its investment in OpenAI, an increase of 490% compared to the same period last year. Based on Microsoft's 32.5% stake in OpenAI, this means OpenAI incurred losses exceeding $12 billion in a single quarter In addition, from the perspective of the AI industry chain, the profitability of upstream and downstream players is clearly differentiated. Represented by NVIDIA, upstream chip manufacturers are the first to enjoy high profits, benefiting from the explosive demand for AI chips, with strong product pricing power and order visibility. Midstream cloud service providers also have clear business models. Amazon, Google, and Microsoft have built resilient business models, deeply integrating AI into their core businesses, forming a solid moat. In the past two years, the revenue share of cloud businesses for these three giants has also shown a gradual upward trend. Oracle has seized the enormous computing power demand required for AI training and inference, locking in huge revenues for the coming years through its cloud infrastructure and high-value contracts with leading AI companies like OpenAI, Meta, and xAI. However, competition among downstream model providers is fierce.

Profitability has shown significant differentiation. General large model providers like OpenAI need to bear exorbitant R&D and computing costs, while enterprise application vendors like Salesforce and Adobe can overlay AI on their mature SaaS products, resulting in lower marginal costs. From the profitability and valuation contribution rates of AI giants' stock prices this year, it can be seen that chip manufacturers have the highest profit contribution rate, followed by cloud service providers, with model providers being the weakest.

Meta belongs to the most unique category. Unlike Microsoft, Google, and Amazon, which have cloud service businesses and profit from AI capabilities as tools and services, Meta has the largest exposure to economic fundamentals, with 99% of its revenue relying on digital advertising (related to the U.S. real economy). It has invested heavily in building a powerful AI social engine, but its commercial returns depend more on the advertising demand of the real economy and the future prosperity of the business ecosystem.

The U.S. is experiencing a typical "stagflation" environment, with the wealth gap continuing to widen and consumption polarizing. The affluent class, who own more assets and stocks, has become richer in the AI bull market, while the lower and middle classes, burdened with student loans, car loans, and mortgages, are under greater living pressure. From the third-quarter reports of U.S. stocks, high-end consumption (such as luxury goods and first-class airline sales) remains strong, while low-end consumption continues to downgrade, with more people turning to McDonald's "value meals," Walmart, or even cheaper supermarkets. When the weakness of the U.S. real economy allows Meta's advertising clients to cut budgets, it may be a more vulnerable moment for the AI chain.

The Fragility of Trillion-Dollar Capital Expenditures

Starting in 2025, capital expenditures of U.S. tech companies are showing competitive increases, raising questions about sustainability. In the third quarter of 2025, the combined capital expenditure of the five leading companies heavily investing in AI (the "AI Five": Microsoft, Meta, Amazon, Google, Oracle) reached $105.773 billion, a year-on-year increase of 72.9%. The enormous capital expenditures have brought cash flow challenges; as of the third quarter of 2025, the average Capex (capital expenditure) / CFO (operating cash flow) ratio for the AI Five was 75.2%, an increase of 29.7 percentage points from a year earlier; the average Capex/revenue ratio was 28.1%, up 12.3 percentage points from a year earlier.

From the perspective of free cash flow (CFO-Capex-net debt repayment), as of Q3 2025, Oracle among the five heavily investing AI giants has seen its free cash flow go "underwater," making it difficult to support the substantial capital expenditures during the same period, relying instead on depleting existing cash and increasing external financing to maintain operations.

In terms of the coverage ratio of average cash reserves at the end and beginning of the period against necessary expenditures (Capex + net debt repayment + dividend payments + repurchase expenditures), as of Q3 2025, the average for the five companies was 94.4%, a decrease of 39 percentage points from a year earlier, with Meta at only 37.3%, indicating that the safety cushion of existing cash reserves may become even thinner in the future.

Based on this, we make the following calculations:

[Assumption 1] Projecting the capital expenditures, operating cash flow, and revenue of the AI Five using the average growth rate from the past year: by the second quarter of 2027, the average Capex/CFO will reach 95.9%, approaching the peak of the highest among the "Four Tech Giants" (Microsoft, Intel, Cisco, IBM) after the bubble burst; by the third quarter of 2026, the average Capex/revenue of the AI Five will reach 39.5%, exceeding Intel's peak after the bubble burst.

[Assumption 2] Calculating the capital expenditures of the AI Five based on the market expected median compound annual growth rate (CAGR) from 2025 to 2028, with revenue and net profit based on Bloomberg consensus expectations, while maintaining the trend ratio of operating cash flow to net profit: by the third quarter of 2026, the average Capex/CFO of the AI Five will reach 96.9%, equivalent to Intel's peak after the bubble burst; By the fourth quarter of 2026, the average Capex/revenue of the five AI giants will reach 38.7%, approaching Intel's peak after the bubble burst.

Overall, the vulnerability of capital expenditures may gradually intensify in the second half of next year. However, considering that technology companies will continue to "go all in on AI" to avoid being eliminated, there is a rigidity in capital expenditures. When companies cut other expenses (such as dividends, buybacks, and equity incentives), it may become a turning point in the narrative.

From the perspective of free cash flow, turning negative in free cash flow could be a moment of deepening vulnerability. Assuming the trend ratio of operating cash flow to net profit remains unchanged, with net profit based on Bloomberg consensus expectations, Capex based on the median compound growth rate of market expectations, and net borrowing extrapolated from the average of the past five years, by the fourth quarter of 2026, Meta will face a free cash flow crisis. When the already fundamentally weak Meta falls into a deeper crisis, doubts about the narrative may be pushed to new heights.

In addition, as the giants have significantly increased capital expenditures for data center construction over the past year, depreciation will not be recognized until they are officially put into use, and its impact on the income statement has not yet manifested. If we assume that from Q4 2024, capital expenditures will gradually be transferred to fixed assets and depreciated linearly over a six-year period, by the third quarter of 2025, the ratio of Capex potential depreciation to net profit for the five AI giants will reach 11.8%, and will rise exponentially in the future.

Using [Assumption 1] for calculation, implied depreciation from Capex will grow from $14.9 billion in Q3 2025 to $114.5 billion by the end of 2028, approximately a 7.7-fold increase. By the end of 2026, 2027, and 2028, the implied depreciation from Capex to expected net profit will reach 37.6%, 60.2%, and 82.0%, respectively.

Using [Assumption 2] for calculation, implied depreciation from Capex will grow from $14.9 billion in Q3 2025 to $123.9 billion by the end of 2028, approximately an 8.3-fold increase. By the end of 2026, 2027, and 2028, the implied depreciation from Capex to expected net profit will reach 37.0%, 60.5%, and 87.7%, respectively.

High Leverage and Off-Balance-Sheet Financing Risks

In the first 11 months of this year, the total issuance of corporate bonds by U.S. hyperscaler companies reached $103.8 billion (excluding loans and private credit), which is more than five times the total issuance for the entire year of 2024 ($20.1 billion), and the weighted average interest rate rose from 4.75% to 4.91%. The surge in supply has already pushed up bond spreads; from October 1 to November 18, the 5-year CDS of Oracle and Coreweave rose by 49 basis points and 304 basis points, respectively, while the OAS spreads of U.S. investment-grade (IG) tech corporate bonds and speculative-grade (SG) tech corporate bonds also increased.

The market generally believes that merely issuing public corporate bonds is unlikely to fill the huge funding gap faced by these giants. It is predicted that from 2025 to 2028, global data center construction will generate $2.9 trillion in Capex demand, of which $1.5 trillion will come from external financing (including $200 billion in corporate bonds, $150 billion in ABS and CMBS products, $350 billion in PEVC and sovereign capital, and $800 billion to $1.2 trillion relying on the private credit market). The opacity of private credit product ratings and holders will pose significant risks.

Taking Meta as an example, a set of off-balance-sheet financing plans was designed for the $27 billion Hyperion data center project—establishing a joint venture called Beignet Investor, 80% owned by investment management company Blue Owl Capital, and issuing $27.3 billion in bonds. Meta holds only a 20% stake and does not consolidate the financial statements, meaning the massive debt does not appear directly on its balance sheet. However, Meta provides substantial implicit guarantees for the joint venture, becoming a contingent liability.

Meta is not an isolated case; companies like xAI and Anthropic have also adopted similar SPV financing models. This reflects the common dilemma faced by tech giants in the AI arms race, needing to meet astronomical funding demands while maintaining attractive financial statements and credit ratings However, such off-balance-sheet financing operations carry significant potential financial risks. When these operations reach the trillion level, systemic risks cannot be ignored. If the technological iteration speed of AI chips and data centers exceeds expectations, it means that the assets held by the SPV may depreciate significantly before generating sufficient returns, and the risks will ultimately be passed on to bond investors.

Historically, off-balance-sheet financing tools have been associated with major crises such as the Enron bankruptcy in 2001 and the subprime mortgage crisis in 2007. Currently, the capital demand for AI investments is enormous. If a large number of companies rely on such hidden leverage, a single default event could trigger systemic risks through highly interconnected capital chains when the technology bubble bursts or the market turns. Even though the current tech boom is different from the internet bubble, with giants enjoying high profit margins, strong earnings growth, and mature and diversified core businesses, the opaque private credit market and off-balance-sheet financing methods could still amplify market volatility and risk transmission effects.

Political uncertainty triggers liquidity tightening, posing external risks to the AI bubble

Narratives are often chosen by liquidity. A key factor for the sustainability of the AI narrative is the incremental liquidity brought by the 150 basis points rate cut since last September, as well as the widespread under-allocation of funds from U.S. institutions and retail investors, with AI being one of the few growth assets available.

In the first 11 months of this year, the stock price increases of the seven giants (and even the entire information technology sector) were driven by earnings, while the rises in traditional sectors such as utilities, real estate, and consumer goods included valuation contributions to varying degrees. This also explains the high concentration of the U.S. tech sector and the crowded nature of market trading.

Looking ahead to next year, both the AI narrative and the real economy in the U.S. are seeking looser monetary policies. Trump's "Everything for the Midterms" will require more relaxed fiscal stimulus to address the "Affordability Crisis" faced by American residents, which in turn will limit the space for monetary policy easing to demonstrate a tough stance against inflation. Trump's efforts to avoid forming a more obvious "re-inflation expectation" before the midterms essentially lead to a conflict between "votes" and "stocks" in 2026, making it inevitable for U.S. stocks to bear greater volatility.

For the Federal Reserve, there is now "no way out," and it can only shoulder the political responsibility to continue cutting interest rates. However, under the pressure of "votes" in 2026, micro-managing will become more difficult. If the new Federal Reserve chair adopts a dovish stance after taking office, the "side effects" of interest rate cuts will be hard to control in an environment of stagflation characterized by supply contraction and demand expansion. Once "both goods and money loosen" triggers re-inflation, even without raising interest rates, the rise in rates will exert liquidity pressure on U.S. stocks, and the upward risk of long-term U.S. Treasury yields will become increasingly prominent Whenever liquidity tightens at the margin, the market tends to adopt a more cautious attitude towards the AI narrative. In October this year, the U.S. government shutdown led to a backlog in the TGA account, preventing fiscal funds from flowing out, which caused a passive tightening effect on the market. The U.S. dollar index briefly broke through the 100 mark, and there has been an increasing number of voices holding an "objective neutral" stance towards U.S. stocks. In November, due to Trump's declining approval ratings in local elections, some Federal Reserve officials turned hawkish, further increasing the adjustment pressure on U.S. stocks.

The uncertainty of liquidity next year essentially stems from the uncertainty of the midterm elections. If Trump's approval ratings continue to decline, his control over fiscal and monetary policy will also weaken, and his influence over allies who have made substantial investment commitments will diminish, leading to a passive increase in the fragility of the AI narrative. In other words, the chain of "politics - liquidity - narrative" may be the root cause of volatility in the U.S. stock market. When Trump shifts to "midterm election mode" will also be the most important macro node in the first half of next year, with the baseline scenario likely occurring after his visit to China in April, transitioning to "domestic affairs" with relevant outcomes; however, this is a dynamic process, and if Trump's approval ratings remain sluggish, the timeline will be forced to advance.

Xuetao Macro Notes

Risk Warning and Disclaimer

The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investing based on this is at one's own risk