Behind the Trillion-Dollar Capital Gamble: Time Is Becoming AI's Counterparty
Unprovoked panic and worry always spread for no apparent reason.
This is by no means rare in history.
The "Tulip Mania" 600 years ago passed in the blink of an eye; the South Sea Bubble 300 years ago was just a fleeting cloud; the dot-com bubble around the turn of the millennium has not yet fully dissipated; and the collapse of Lehman Brothers during the 2008 financial crisis still has lingering echoes to this day...
There is still no conclusive answer as to where the "Tulip Mania" came from
Right now, we are experiencing an unprecedented AI "gamble" in history.
Most people don't realize that unlike the wins and losses of the past hundreds of years, AI is a gamble that all participants simply cannot afford to lose.
And this time, it is not just a matter of commercial success or failure, but an ultimate game concerning humanity and civilization.
More often than not, success is just a coincidence
In 2024, Leopold Aschenbrenner, who was fired from OpenAI, founded the hedge fund Situational Awareness. At that time, the fund was roughly $200 million in size. By July this year, the size of Situational Awareness had grown to a staggering $450 billion.
If the story had stopped there, this German post-2000 generation would undoubtedly have been portrayed as a child prodigy, just like SBF, the founder of FTX back then. SBF's story took a sharp turn for the worse in the turmoil in November 2022, and he eventually ended up in prison.
Leopold started his career in AI and made his fortune from AI. Situational Awareness used leveraged positions to heavily hold stakes in storage suppliers such as SanDisk, Micron, and SK Hynix.
In the current AI gamble, these once aging hardware suppliers have become the biggest beneficiaries.
In the second quarter of this year, Samsung Electronics' revenue increased by 130% year-on-year, its operating profit surged more than 18 times year-on-year, and the semiconductor division contributed 99% of the company's total profit. SK Hynix's revenue rose 257% year-on-year, and its operating profit jumped 557% year-on-year.
Micron's revenue in the third quarter of fiscal 2026 increased by 345.7% year-on-year, and its GAAP net profit increased nearly 14 times year-on-year. Kioxia's revenue in the first quarter of fiscal 2026 grew by 415.5% year-on-year, and its net profit soared more than 45 times year-on-year.
SanDisk's revenue in the fourth quarter of fiscal 2026 increased by 372% year-on-year, its GAAP net profit exceeded 6.9 billion US dollars, compared with a net loss of 23 million US dollars in the same period last year, and its gross profit margin reached an astonishing 84.6%.
However, by July, the situation took a sudden downturn.
The stocks of these manufacturers were sold off on a large scale, and SanDisk's share price alone fell by nearly 47% in July.
The high-spirited Leopold finally ushered in the turning point of his own story. Situational Awareness lost 350 billion US dollars in one month, surpassing Archegos run by Bill Hwang in 2021, setting a loss record in the history of hedge funds.
The market has its own perception of the situation.
No matter how much we advocate the rationality of the market, no matter how many intellectual geniuses there are, or even how many models and artificial intelligence we introduce as tools, we must admit that the foundation of market operation is still people who cannot remain completely rational forever.
As a result, a certain trader takes profit, a certain institution withdraws its gains, and more market participants follow suit. Eventually, it gradually evolves into a sell-off storm sweeping the entire market.
The question is, is this just a coincidental "luck"?
As the old saying goes, things are changing?
In the United States, the social media youth addiction litigation (MDL 3047) is exploding. Parents and teenagers accuse Meta, TikTok, Snapchat and YouTube of designing addictive platforms that harm adolescents' mental health, leading to depression, anxiety, eating disorders and self-harm. As of August 25, the number of such lawsuits has skyrocketed from less than 1,000 at the beginning of this year to more than 3,100.
In the first three years after the lawsuits were filed, none of the defendants paid any compensation, and the large companies being sued often used the Section 230 immunity clause in the U.S. Communications Decency Act as a defense.
However, the situation changed in the first quarter of 2026.
In January, Snap and TikTok reached a secret settlement with the plaintiffs before the trial of the bellwether case. In March, the Los Angeles Superior Court jury found Meta and Google negligent in their design choices, and ruled that Meta and Google should bear a total of 6 million US dollars in compensation. In May, the first federal MDL bellwether case reached a settlement before trial, with Snap, TikTok and YouTube reaching a $27 million settlement with the plaintiffs.
Earlier this year, Mark Zuckerberg attended a hearing in Los Angeles
On August 18, the lawsuit jointly filed by 29 states against Meta first opened in Oakland. The theoretical maximum penalty is as high as 1.4 trillion US dollars, which is almost equivalent to Meta's market value.
This is an earth-shattering change in the business logic of AI.
What does this mean?
Over the past decade, user-facing services, platforms and products have always evaded responsibility under the excuse that "algorithms are innocent" and "algorithms can have no values".
However, after the rise of AI, companies and entrepreneurs can no longer use the same excuses to exonerate themselves from the risks and drawbacks associated with AI.
Once you use AI to write code and develop products, the company needs to take responsibility if something goes wrong. If AI-related content posted by users infringes copyright and the rights of others, the platform shall also bear corresponding secondary liability.
In short, the U.S. institutionalized governance of AI still has great uncertainty in legal practice, which is the most prominent destabilizing factor in the entire current AI capital gamble.
With the support of the ruling faction in the White House and the federal government, companies represented by Google, Microsoft, OpenAI, Meta, Anthropic and Pentair are trying their best to consolidate their first-mover advantages in the AI competition. However, the struggles within factions and systems are intentionally or unintentionally exerting resistance.
Uncertainty is not only the root cause of the sharp short-term fluctuations in the AI capital market, but also its inevitable trend.
However, if we extend our vision to a scale of five or ten years, we will realize that there is a greater inevitable crisis in it.
In the second quarter, Google recorded a negative free cash flow of 5.9 billion US dollars for the first time in its history
In the first half of this year, Google's operating revenue reached 80.5 billion US dollars, but its free cash flow dropped from about 24.3 billion US dollars to only 4.3 billion US dollars.
From 2024 to 2025, Google's capital expenditure increased from 52.5 billion US dollars to 91.4 billion US dollars, and its expenditure in 2026 will reach 200 billion US dollars, which will continue to increase significantly in 2027.
In 2025, the total expenditure of large companies such as Amazon, Google, Meta, Microsoft and Oracle on artificial intelligence was only 121 billion US dollars.
UBS analysts predict that global enterprises will add as much as 900 billion US dollars in new debt in 2026, while Morgan Stanley and JPMorgan Chase are more bold in predicting that the technology industry may need to issue up to 1.5 trillion US dollars in new debt in the next few years to fund the construction of artificial intelligence and data center infrastructure.
In 2023, Sam Altman, CEO of OpenAI, attended a hearing in Congress
OpenAI's goal is to invest a total of 600 billion US dollars by 2030, and the previous expectation was as high as 1.4 trillion US dollars.
Promoting development by taking on debt, even if it cannot be called drinking poison to quench thirst, is at least a real sense of involution.
So what is the result of involution?
In 2023, OpenAI's revenue was about 2 billion US dollars. In March this year, the company stated that its monthly revenue reached 2 billion US dollars. OpenAI estimates that it will achieve profitability in 2030.
In May this year, after completing a $65 billion financing, Anthropic announced that its revenue run rate as of early May had reached 47 billion US dollars. Three months later, the company claimed that its revenue in 2028 would be between 190 billion and 200 billion US dollars. The company's latest prediction is that its total addressable market (TAM) exceeds 30 trillion US dollars.
At present, OpenAI's valuation exceeds 840 billion US dollars, and Anthropic's valuation exceeds 965 billion US dollars, and both have secretly submitted draft IPO prospectuses.
The capital market will inevitably compare them with SpaceX, which currently has a market value of more than 1.85 trillion US dollars.
The company's revenue from April to June this year was 7.8 billion US dollars, a year-on-year increase of more than 90%, and its artificial intelligence business revenue increased by about 250% year-on-year. However, SpaceX's capital expenditure surged from 2.83 billion US dollars in the same period last year to more than 18 billion US dollars, and its artificial intelligence capital expenditure increased significantly from 749 million US dollars to 15.83 billion US dollars.
Obviously, the development of the AI market is heavily dependent on continuous capital infusion, which at first glance seems like a game that makes friends with time.
However, an obvious problem that everyone has ignored is that according to the profitability deadline set by OpenAI, the company only has four years left. Don't forget, four years ago, NVIDIA was still a gaming graphics card company with a share price of less than 20 US dollars. In the past five years, the share price of Beyond Meat, the once-popular artificial meat company, has plummeted by more than 99%.
AI capital and the market are more like rivals against time.
Time is money, a maxim that capitalist society holds as an unshakable creed, which is extremely appropriate to describe the development trend and trajectory of AI. Through unprecedented scale of investment in history, the participants of this gamble try to use money to gain time advantage, or at least delay certain disadvantages.
This is exactly the current situation in the United States.
Due to the rise of short-term and long-term interest rates in the past few years, the interest cost of national debt has risen sharply. The Congressional Budget Office predicts that the interest cost in 2026 will exceed 1 trillion US dollars. By 2027, interest costs will surpass defense, Medicare and Medicaid to become the second largest expenditure item of the government, and will reach 2.1 trillion US dollars by 2036. In the ten years from 2026, the U.S. debt interest cost will reach an astonishing 16.2 trillion US dollars.
It is worth noting that the Congressional Budget Office predicts that if the interest rate on new debt rises, the average interest rate will exceed the economic growth rate (R>G) by 2029, and this gap will reach 75 basis points by 2036.
The rise in interest rates pushes up debt, the rise in debt pushes up interest rates, and the rise in interest rates in turn continues to push up interest costs. It is not alarmist to say that this vicious cycle will eventually trigger a debt crisis.
This crisis is the fundamental reason why the current U.S. capital market is investing in the AI field with full force at a lightning speed. Capital firmly believes that AI, like the Internet in the last century, will once again help the United States obtain and consolidate its position as the sole leader.
Let's imagine such a possibility: if the real technological singularity is not AI, or the scale and revenue of the AI market in the next ten years are not as huge as capital estimates, and the dual capital crises at the market level and the national level break out at the same time, then how should we deal with such a situation?
This article is from the WeChat official account "AI Value Officer", author: RELIEX, published with authorization from 36Kr.