U.S. stocks are playing a new type of AI roulette game.

Written by: Wuyi Fan

Recently, there has been a joke circulating in the US stock market community:

“OpenAI invests $100 billion in Oracle for cloud computing services; Oracle invests $100 billion in Nvidia for graphics cards; Nvidia reinvests $100 billion in OpenAI to layout AI systems. Question, who actually spent this $100 billion?”

Of course, the above is just a joke; the amounts and facts are quite different, and it is not that the three companies are circling around the same sum of money, but it does reflect a closed loop of a new type of capital narrative.

In this closed loop, each link is a real contract or investment, and every action is amplified by the capital market, thereby bringing about a trillion-dollar increase in market value.

On September 11, Oracle's stock price surged by 36%, marking the largest single-day increase since 1992. Overnight, the company's market value soared to $933 billion, and founder Larry Ellison even briefly surpassed Musk to become the world's richest person.

On September 22, NVIDIA and OpenAI announced a strategic partnership, with NVIDIA planning to invest up to $100 billion in OpenAI. NVIDIA's stock rose nearly 4%, surpassing a market value of $4.46 trillion, triggering a rally in the entire tech stock market, with all three major U.S. stock indices hitting record highs.

1 trillion dollars seems like a lot, but it can easily drive up the U.S. stock market by trillions overnight, definitely a small investment for a big impact.

The US stock market is playing a new type of AI roulette game.

Triangle Cycle: How does money circulate?

In the real-world investment maze, three names form a perfect capital closed loop: OpenAI, Oracle, and Nvidia.

First Ring: OpenAI is hungry for computing power

The core protagonist of the story is OpenAI. As the creator of ChatGPT, OpenAI handles requests from 700 million users every day. Such a scale of AI computation requires massive computing power.

This year, OpenAI signed the largest technology contract in history, a 5-year $300 billion cloud computing agreement with Oracle. Under this contract, OpenAI is required to pay Oracle approximately $60 billion per year, which is equivalent to 6 times the company's current annual revenue.

What can this money buy? 4.5 gigawatts of data center capacity, equivalent to the electricity consumption of 4 million American households. Oracle is building data center parks for OpenAI across five states, including Wyoming, Pennsylvania, and Texas.

For OpenAI, this guarantees a place and computing power to run models; for Oracle, this is revenue certainty for the next five years.

Second Ring: Oracle needs chips

After receiving a huge order from OpenAI, Oracle faces a problem: how to build these data centers?

The answer is chips, a large number of chips. Oracle plans to invest hundreds of billions of dollars in the StarGate project to purchase NVIDIA's GPUs. According to industry estimates, 4.5 gigawatts of computing power will require more than 2 million high-end GPUs.

Oracle CEO Safra Catz stated bluntly: “The vast majority of our capital expenditure investments go towards purchasing revenue-generating equipment that will enter data centers.”

The “income-generating devices” mainly refer to Nvidia's H100, H200, and the latest Blackwell chips.

Oracle has become one of Nvidia's largest clients.

Third Ring: Nvidia's Reinvestment

At the same time that Oracle is frantically purchasing chips, Nvidia announced an astonishing decision: to invest $100 billion to support OpenAI in building a 10-gigawatt AI data center.

This investment will be carried out in phases, with Nvidia investing corresponding funds each time OpenAI deploys 1 gigawatt of computing power. The first phase is planned to start in the second half of 2026, using Nvidia's Vera Rubin platform.

NVIDIA CEO Jensen Huang said in an interview: “A 10-gigawatt data center capacity is equivalent to 4 to 5 million GPUs, which is about our total shipment volume for this year.”

At this point, a perfect capital cycle has formed:

OpenAI pays Oracle for computing power, Oracle uses the money to buy chips from Nvidia, and Nvidia invests the money earned back into OpenAI.

Wealth amplifier between reality and virtuality

The long-term contracts worth 300 billion dollars brought Oracle a single-day market value increase of over 250 billion dollars, while an investment of 100 billion dollars led to a single-day increase of 170 billion dollars for Nvidia.

The three companies support each other and endorse each other, resulting in a resonance in stock prices.

There is a rationality behind the rise in stock prices.

What is most scarce for the capital market is certainty about the future.

The contract between Oracle and OpenAI means that part of its cloud revenue for the next five years is locked in, leading investors to naturally give a higher valuation.

In addition, NVIDIA is using “GW (gigawatts)” as the unit of measurement this time. 1GW is roughly equivalent to the scale of a super data center. 10GW means that NVIDIA and OpenAI are building a new generation AI factory. This new narrative language is more imaginative than “how many GPUs were purchased,” making it easier for the market to be driven.

NVIDIA's investment in OpenAI is equivalent to saying, “I recognize it as a super customer of the future”; OpenAI signing a contract with Oracle is equivalent to saying, “Oracle has the capability to support my future cloud computing needs,” allowing OpenAI to secure more funding; Oracle purchasing NVIDIA GPUs is equivalent to saying, “NVIDIA's chips are in short supply.”

This is a stable and prosperous industrial chain.

This loop seems flawless at first glance, but a closer look reveals its intricacies.

OpenAI currently has an annual revenue of about 10 billion dollars but has committed to paying Oracle 60 billion dollars each year. Where will this huge gap be filled?

The answer lies in one round of financing after another. In April, OpenAI completed a $40 billion financing round and is expected to continue raising funds.

In fact, OpenAI pays Oracle with investors' money, Oracle uses that money to purchase Nvidia chips, and Nvidia then reinvests part of the revenue back into OpenAI. This is a cycle driven by external capital.

In addition, these astronomical contract figures are mostly based on “commitments” rather than immediate delivery, which can be delayed, renegotiated, or even canceled under certain conditions. What the market sees are the numbers of commitments, not the actual cash flowing.

This is the magic of the modern financial market: expectations and commitments can create a multiplicative wealth effect.

Who will foot the bill?

Returning to the initial question of the joke: “Who exactly put out this 100 billion?”

The answer is investors and the debt market.

SoftBank, Microsoft, Thrive Capital, and other investment institutions are the direct payers of this game. They invested hundreds of billions of dollars into OpenAI, supporting the entire capital cycle. In addition, banks and bond investors also provided funding support for Oracle's expansion, while ordinary people holding related stocks and ETFs are the “silent payers” at the end of the chain.

This AI capital rotation game is essentially a financial engineering of the AI era. It leverages the market's optimistic expectations for the future of AI to create a self-reinforcing investment cycle.

In this cycle, every party is a winner: OpenAI gains computing power, Oracle secures orders, and Nvidia sees sales and investment opportunities. Shareholders watch their paper wealth grow, and everyone is happy.

But this joy is built on one premise: the commercialization process of AI in the future can support these astronomical investments. Once this premise is shaken, the beautiful cycle may turn into a dangerous spiral.

Ultimately, the buyers of this game are every investor who believes in the future of AI, betting today's money on tomorrow's AI era.

I hope the music won't stop.

Disclosure: The author holds shares in NVIDIA and AMD.