<p>On October 29th， NVIDIA became the first company in history to surpass a market capitalization of 5 trillion US dollars as soon as the US stock market opened.</p>One day prior， at the GTC DC conference， NVIDIA CEO Jensen Huang showcased the robust demand stemming from Artificial Intelligence (AI).
For instance， he projected that its Blackwell and Rubin architecture chips would generate over 500 billion US dollars in revenue for 2025 and 2026， from an estimated 20 million GPUs. This projected revenue is an astounding five times the revenue generated by its Hopper architecture chips from 2023 to 2025.
During his presentation， Huang once again dismissed the notion of an “AI bubble.”
He had every reason to dismiss it. As the pivotal “pickaxe seller” in the AI era， NVIDIA holds significant pricing power over scarce resources. Essentially， NVIDIA’s market valuation is directly tied to the market’s optimism about the future of AI in the United States.
However， with NVIDIA’s market capitalization crossing the 5 trillion US dollar mark， voices questioning the “AI bubble” have also reached an unprecedented height.
This scrutiny can be traced back to a related-party transaction.
On September 22nd， NVIDIA announced a colossal investment of up to 100 billion US dollars in OpenAI. In return， OpenAI committed to a massive order for 10GW worth of GPUs.
This “hand-washing” transaction， where funds essentially circulate between closely linked entities， immediately amplified concerns in the capital market about an AI bubble.
Recently， in a speech in Washington， IMF Managing Director Kristalina Georgieva stated that global stock markets have surged on optimism about AI’s productivity-enhancing potential. However， she warned that financial conditions could “turn on a dime，” with current valuations “approaching levels seen during the dot-com boom 25 years ago.” A sharp market correction， she cautioned， would drag down global growth.
The Financial Policy Committee (FPC) of the Bank of England has echoed similar concerns. Its latest quarterly report indicated that the US stock market “shows overvaluation by several metrics，” particularly concentrated in AI-related technology companies. This high concentration makes the market particularly vulnerable to shocks when AI expectations cool down.
In this intricate game， companies like NVIDIA are both players and spectators， with money moving from one pocket to another. Is this a superhighway leading to the future， or is it a grander spectacle of fireworks than the dot-com bubble of 2000?
01 A Capital’s Internal Circulation
To discuss an AI bubble， one cannot avoid the concept of “circular financing.”
A cohort of tech giants is collectively inflating a massive valuation balloon using each other’s money.
Let’s examine how these giants are maneuvering in this capital game. The core logic of this game is: you buy my services， I invest in your company， and he then buys my chips. This is not merely simple business cooperation; it resembles a closed-loop capital system.
Let’s break down this closed loop:
1. NVIDIA， a chip supplier， invests 100 billion US dollars in OpenAI， an AI model company.
2. With the funds， OpenAI places a substantial order for 10GW of GPUs from NVIDIA. This means a significant portion of NVIDIA’s investment will flow back to NVIDIA as revenue.
3. To operate such an extensive GPU cluster， OpenAI requires massive cloud infrastructure. Consequently， it procures cloud services valued at 300 billion US dollars from Oracle， a cloud service provider.
Ultimately， to support such immense cloud services， Oracle’s primary choice is to purchase more chips from NVIDIA.
A perfect capital loop is thus established. Money circulates among these three companies， with each recording revenue and orders， making the narrative appear exceptionally compelling.
This is not the entirety of the story.
In October 2025， NVIDIA’s primary competitor， AMD， also entered this game.
AMD announced an AI infrastructure partnership with OpenAI， agreeing to deploy 6GW of GPU computing power for the latter. In exchange， AMD did not receive direct cash but instead granted OpenAI 160 million warrants， effectively giving OpenAI an indirect stake of nearly 10% in AMD.
In this capital game， which resembles the “chaining of boats” strategy in ancient warfare for mutual support， the boundaries between suppliers， customers， investors， and even competitors are becoming increasingly blurred.
As the “pickaxe seller，” NVIDIA also seeks to gain a share of the AI application development pie through its investments. These giants are tightly binding their fates through related-party transactions and hefty orders， collectively amplifying the narrative surrounding AI.
The realization period for this narrative is extended to the distant year of 2030， involving long-cycle， capital-intensive sectors like computing power， electricity， and energy storage， with uncertain return prospects.
The giants’ money is not inexhaustible. What happens when funds run low?
Debt， and then more investment.
If the hundreds of billions in capital expenditures for AI by these giants in the past couple of years were primarily funded by their excess operating cash flow and resources squeezed through cost-cutting measures and layoffs， then the players at the table are now resorting to riskier tools: “leveraging up.”
As of the second quarter of 2025， the combined free cash flow of the “Magnificent Seven” has decreased by 62.45% compared to the end of 2024.
(Data source: Wind)
The tech giants are no longer content to rely solely on free cash flow， elevating external financing to an unprecedented level. Companies like Meta are actively raising funds for AI data center construction through bond issuances， equity financing， and even private credit facilities.
For instance， Meta partnered with private equity giant Blue Owl to raise 27 billion US dollars through a private bond issuance for data center construction， setting a record for private bond issuance. Although this bond received an A+ rating from S&P， its yield was as high as 6.58%， approaching junk bond territory.
More significantly， a subtle “hand” appears to be at play behind this fervor.
The US government has designated AI as a core element of its national strategic competition， implying a substantial increase in its commitment to this field.
It is foreseeable that regulatory oversight on capital flow into the market will continue to be lenient， which will undoubtedly exacerbate the issues of leverage and overvaluation.
The combination of favorable financing prospects and historically low corporate credit spreads is enticing these giants to borrow more， betting on a future that could be immensely brilliant or utterly disastrous.
Regardless of how compelling the narrative becomes， an undeniable fact is that OpenAI continues to operate at a loss.
OpenAI’s revenue for the first half of 2025 was approximately 4.3 billion US dollars; however， the company incurred a loss of 13.5 billion US dollars during the same period， primarily due to research and development in AI and the operational costs of running ChatGPT.
As the era of deep learning models matures， pricing power in large AI models is shifting from “technological monopoly” to “market competition.” OpenAI’s large models are also being forced to reduce prices. To increase prices in the future， OpenAI must offer revolutionary capabilities that cannot be quickly replicated， or it will face a difficult struggle.
It’s not just OpenAI; other AI giants also face the awkward reality of “not making money.”
According to brokerage forecasts， Google’s Gemini large model is projected to generate only a few hundred million US dollars in monthly revenue in 2025. Doubao， ByteDance’s large model， is expected to have monthly revenues in the tens of millions to hundreds of millions of Chinese yuan. For their parent companies， with valuations in the hundreds of billions or trillions of dollars， this revenue is almost negligible.
However， a more serious issue than the profitability of the large models themselves lies in the downstream application ecosystem.
Recently， MIT (Massachusetts Institute of Technology) released a report on the generative AI field. It highlighted that 95% of investments in generative AI have failed to generate returns for businesses， with only 5% of projects achieving commercial success. Furthermore， S&P Global mentioned in early 2025 that 42% of generative AI projects were abandoned midway.
This clearly indicates that the commercialization progress of downstream applications is far from sufficient to support the upstream capital expenditure of trillions of dollars for computing power.
Even so， AI must succeed.
It is crucial to understand that AI will influence corporate standing， societal underlying logic， and global order in the future. The United States views AI as a weapon for control， while China sees it as a tool for development.
AI large model companies cannot generate their own revenue but require substantial funding. So， where will this money come from?
02 The US Stock Market Must Rise
This year， the S&P 500 has risen by 17.16%. While corporate earnings prospects are cited as the superficial driver， in reality， the US government is indirectly injecting capital into AI through the stock market， while simultaneously using the AI narrative to “hijack” the performance of US stocks and solidify its financial hegemony.
To grasp this， let’s first consider the utility of the US stock market to the United States.
As is well known， the United States has constructed a financial hegemony system by leveraging the strong position of the US dollar. In simple terms， **the US can exchange a piece of paper (the dollar) for a lifetime of labor from other nations， and they dare not refuse.**
Global trade relies on dollar settlements. When the US wants to spend money， it can simply print dollars. If the US spends too much， it issues Treasury bonds for other countries to purchase. If any country disobeys， the US can expel it from the “global payment group” (SWIFT)， rendering it unable to conduct transactions.
Within this system， the US stock market primarily provides liquidity for financial hegemony.
Ample liquidity facilitates the smooth global flow of dollars， promoting their outflow and inflow from the United States， thereby maintaining and strengthening the dollar-based financial hegemony.
As the world’s largest consumer market and importer， the US runs trade deficits with most major economies; in 2024， the US trade deficit in goods exceeded 1.2 trillion US dollars. The US exports dollars globally through trade. The US then reclaims dollars through US Treasury bonds and a long-term bull market in US equities， meaning other countries use dollars to buy US bonds， and global capital allocates dollars to US stocks.
This ensures the smooth circulation of the dollar hegemony system.
It is important to note that if liquidity becomes excessive or overly loose， and the US continuously exports dollars without being able to reclaim them through US bonds and stocks， it could lead to increased expectations of dollar depreciation， eroding the dollar’s foundational creditworthiness.
If market participants lose confidence in the dollar’s value stability， they may reduce their holdings of dollar assets and seek other more stable currencies or assets for reserves， thereby weakening the dollar’s status as an international reserve currency.
For the sake of dollar liquidity， the US stock market must maintain a long bull run.
The US stock market must continuously “create dreams，” selling persistent growth expectations to the world to attract real money and reclaim dollars， thereby further solidifying its financial hegemony.
Unlike in previous periods， the current “AI narrative” has driven the rise of US stocks to a far greater extent. This actually indicates that for the US economy and its stock market， AI is currently the only sector capable of generating significant incremental growth and imaginative potential.
Statistics show that the top 10 largest weighted stocks (market capitalization weighted) in the S&P 500 index in 2025 account for 41.43% of the total. These top 10 include: NVIDIA， Microsoft， Apple， Google (GOOG)， Google (GOOGL)， Amazon， Facebook， Broadcom， Tesla， and Berkshire Hathaway – essentially all related to AI.
(Data source: Wind)
Previously， the highest proportion of the top 10 weighted stocks was during the dot-com bubble in 2000， at 23.52%. At that time， the top 10 weighted stocks included: General Electric， ExxonMobil， Pfizer， Citigroup， Cisco Systems， Walmart， Microsoft， American International Group， Merck， and Intel.
Furthermore， the current “Magnificent Seven” of the US stock market – **Apple， Microsoft， Google， Amazon， NVIDIA， Tesla， and Meta** – account for 37.29% of the total market capitalization of the S&P 500 index， a figure that was less than 20% five years ago.
The primary drivers of the current S&P 500 rally have been a combination of improved corporate earnings， sustained optimism surrounding AI-related investments， and a macroeconomic environment conducive to risk assets， including expectations of Federal Reserve interest rate cuts.
Specifically， a breakdown of US stock returns shows that the inflated US stock market value is driven not only by corporate earnings but also by changes in price-to-earnings ratios and dividend buybacks.
According to data from Western Securities， the 23.1% annualized return of US stocks from September 2022 to September 2025 comprised 4.9% from corporate earnings growth， 17.3% from price-to-earnings valuation expansion， and 1.5% from dividends and buybacks.
This means that the rise in US stocks over the past three years has been primarily driven by valuation expansion， implying that the capital market has granted a certain premium. If this premium cannot be absorbed by earnings growth， it will develop into a bubble.
Of course， the US government， led by Donald Trump， would certainly not allow the US stock market bubble to burst.
On one hand， “the rise of US stocks” is a testament to Trump’s performance.
In the United States， most households indirectly own stocks through mutual funds， index funds， or retirement accounts like 401(k)s， with a smaller portion owning stocks directly.
Historically， the stock market bull run during Trump’s presidency (2017-2020)， which saw the S&P 500 increase by over 50%， became a key campaign plank for his 2024 bid. Some voters would support him due to the rising stock market and their replenished wallets.
With the congressional midterm elections approaching in November 2026， the impact of a rising stock market on his election prospects is undeniable.
On the other hand， the US government is also a direct beneficiary of the rising stock market.
In August 2025， the US government agreed to obtain nearly a 10% stake in chip giant Intel by converting a previous appropriation of nearly 9 billion US dollars into equity. Intel’s stock price has risen by nearly 100% since August 2025.
This also signifies a shift in the US government’s industrial policy towards AI， from a “subsidy-driven” approach to “capital control.”
03 AI: The Dominant Narrative in US Equities
Behind NVIDIA’s 5 trillion US dollar valuation lies another undeniable variable: the US’s persistent pursuit of an “America First” AI hegemony strategy. To achieve this goal， both US equities and bonds must perform robustly.
In July 2025， the White House released “Winning the Race: The US AI Action Plan，” outlining 90 specific policies with a singular core objective: “to ensure America’s sustained leadership in artificial intelligence.”
The US aims to translate its AI technological advantage into geopolitical influence. By leading global AI governance， the US intends to embed “American values” into international standards， compelling allies to align with US policies on a regulatory level， thereby creating a sticky relationship of “technological dependence-policy synergy” and ultimately establishing an AI technology alliance.
The plan explicitly states that the US will not only counter China’s influence in international organizations but also export a complete AI technology stack (hardware， models， software， applications， and standards) to its allies， preventing them from “turning to our competitors.”
The “Stargate” project is a crucial initiative in the US’s quest for AI dominance. Its core objective is to build the world’s largest AI data center network through an investment of 500 billion US dollars.
Under this “Stargate” plan， OpenAI， SoftBank， Oracle， and MGX will form a joint venture， the Stargate Project， to oversee it. Arm， Microsoft， and NVIDIA will also participate in the project company.
The most critical factor in this PowerPoint plan is the source of the 500 billion US dollars for the project.
Specifically， for the initial 100 billion US dollars tranche， OpenAI and SoftBank will each contribute 19 billion US dollars， while Oracle and the Middle Eastern sovereign wealth fund MGX will jointly contribute 7 billion US dollars， totaling 45 billion US dollars from these four entities. The remaining funds will be provided by new， yet-to-be-determined investors.
Mutual Investment believes that the own funds of these companies alone are insufficient. Debt financing will be required to ensure the smooth execution of the AI strategy， which in turn necessitates the continuous new highs in US equities.
Considering the list of Stargate project participants， most are already listed on US stock exchanges. US-listed companies rely on their stock market valuations for financing.
For instance， in terms of debt financing， once a company is publicly listed， its value is clearly established， making bank loans relatively accessible. Concurrently， companies can even use their stock as collateral for loans.
In addition to financing for expansion and business operations， US-listed companies can borrow money for stock buybacks. With a reduced number of outstanding shares， earnings per share increase if profits remain the same， thus inflating stock valuations.
Companies can borrow money to boost valuations， and then borrow more money based on those inflated valuations， creating a “financing-valuation increase-re-financing” cycle.
To bridge the funding gap for the “Stargate” project， more US-listed companies or capital will be needed， requiring US equities to continue their upward trajectory. Consequently， the AI bubble will inevitably grow larger.
Should the AI bubble burst， the impact will not be limited to the stock market; the crisis will likely transmit to the bond market and the banking credit system.
If liquidity in the entire financial system then tightens， it could trigger a larger financial crisis. The Federal Reserve’s six consecutive interest rate hikes to 6.5% from 1999 to 2000 were also a significant factor contributing to the dot-com bubble’s collapse.
Ultimately， the future of AI， built on trillions of US dollars， and this grand bubble， face an uncertain destination.
关于作者
