Jensen Huang glaubt nicht an Blasen, OpenAI braucht einen "Gelddruckmaschine".

“Even leading companies like OpenAI have to resort to various financial maneuvers to maintain their valuations. The burst of the AI bubble is not far off.” said a Google engineer.

Previously, companies such as OpenAI, AMD, and NVIDIA established a complex cooperation relationship of “equity for orders.”

OpenAI CEO Sam Altman tightly “embraced” top Silicon Valley figures like Jensen Huang, Lisa Su, and Jensen Huang, jointly painting a story of “reshaping the power structure in Silicon Valley,” which is a very rare sight.

Lisa Su, Sam Altman, and Jensen Huang joined forces to “restructure Silicon Valley.” The picture was generated by AI.

This kind of alliance has also raised industry concerns, including the monopoly risks behind it and the bubble fueled by trillions of dollars in investments.

Compared with the destructive power of monopoly, once the bubble bursts, investors will be left with a mess. Similar scenarios have played out in the Internet bubble in 1999 and the real - estate bubble in 2008.

A key question is, do figures like Jensen Huang and Sam Altman, who frequently appear in various podcasts and interviews, preach about AI, and talk extensively about AI capital expenditure, count as a group “blowing bubbles” when they activate the capital machine and paint a trillion - dollar story?

“Equity for orders,” reshaping the power structure in Silicon Valley

Competition among tech giants has always mainly revolved around “talent wars” and “patent wars.” Occasionally, there are cooperation forms such as “cross - shareholding” or “equity for orders,” but true “resource restructuring” is relatively rare. In this regard, the most notable recent case is the strategic cooperation between OpenAI and AMD.

According to the agreement, OpenAI will deploy data centers with a total capacity of 6GW based on AMD chips, and AMD will symbolically grant up to nearly 10% of its shares to OpenAI for $0.01.

In terms of resource allocation scale, a 1GW data center requires approximately 500,000 GPUs of the GB200/300 series, and the hardware cost alone exceeds $20 billion. A total of 6GW corresponds to approximately 3 million GPUs, and the total hardware investment is expected to exceed $120 billion (excluding data center infrastructure and other operating expenses).

Facing such a large and certain order, the capital market reacted quickly. After the official announcement of the cooperation, AMD's stock price rose by a cumulative 43% over three consecutive trading days, and its market value increased by more than $100 billion.

Put simply, the cooperation between OpenAI and AMD can be understood as “the customer places an order with me, and I grant options to the customer,” which is comparable to “rebounding one's own shot” on the basketball court. Some people also describe it as “the internal cycle of the US technology industry.”

However, this kind of internal cycle has happened to NVIDIA many times.

Previously, NVIDIA officially announced that it would invest a total of $100 billion in OpenAI over the next decade. Even earlier, NVIDIA invested in the emerging US cloud provider CoreWeave, which is regarded as its “prodigal son.”

The difference between AMD's and NVIDIA's cooperation with OpenAI is that NVIDIA's story has become “I buy the customer's equity, and the customer places an order with me.”

So, when Jensen Huang evaluated the deal between OpenAI and AMD, he used the words “surprised” and “smart.” But in essence, both companies are pinning their hopes on OpenAI, a potential customer with an expected annual revenue of $200 billion in the future, to continuously contribute to their revenue.

The certainty has made Jensen Huang successful repeatedly, and he decided to replicate the same scenario with xAI. Jensen Huang said in an interview with CNBC, “I want to invest in anything that Elon Musk is involved in.”

When talking about the above “internal cycle,” an insider at NVIDIA told the author, “Everyone wants the money they invest to come back quickly. In fact, after such operations, the invested money has already been recovered. The more money you invest, the more you earn.”

When NVIDIA officially announced its investment in OpenAI, its stock price also rose by nearly 4% at the close of the day, and its market value increased by nearly $180 billion. The “investment - profit effect” is even more obvious with CoreWeave, its “prodigal son.”

According to the 13F filing submitted by NVIDIA in August, it holds approximately 24.28 million shares of CoreWeave, with a holding market value of approximately $3.959 billion. In the previous quarter, the holding market value was $897 million.

For OpenAI, NVIDIA and AMD are not the end of the “internal cycle.” Sam Altman also previewed in a recent interview that there will be more similar cooperations in the next few months.

There is a bubble, but this time it may be different

The complete narrative framework of AI can be summarized into six levels: energy, computing power, middleware, models, applications, and users.

The public hopes that OpenAI can tell a business story focusing on the interaction among models, applications, and users, with growth prospects and transformative power.

However, the progress of each link in this narrative is based on huge investments.

The reality is that OpenAI, with an annual revenue of only a little over $10 billion, cannot afford such huge investments. Jensen Huang also said bluntly in an interview, “They have no money.” The gap between the “meager” revenue and the aggressive strategic investment has given rise to market concerns about the industry bubble.

“The bubble does exist, which can be seen from the current product forms.” A Chinese AI entrepreneur said. In his opinion, the bubble is concentrated in the application layer, mainly because many startup teams have insufficient understanding of AI. “The recruitment ideas and job descriptions (JDs) of domestic enterprises for Agent development are incorrect. Many teams recruit AI Agents as independent positions, but in fact, it is just an extension for a specific scenario.”

He told the author that financial risk - control developers need to understand the logic of financial business, and medical diagnosis developers need to understand the medical knowledge system, rather than being general “AI Agent engineers.”

Lu Yi, a partner at Brilliant Phoenix, believes that the bubble is mainly manifested in small companies in the secondary market and medium - sized companies in the primary market.

He believes that companies with unclear profit models, lack of computing power support, and relying only on concept hype need to “delve into the industry to solve problems” as soon as possible, because “the demand for horizontally general problems has been covered by the capabilities of large models,” while leading enterprises can continue to raise funds to expand their computing power and thus expand their advantages, as “computing power is actually converted into training and inference capabilities.”

Regarding computing power, Jensen Huang, who has been frequently traveling between China and the US this year, is the most active preacher. He has repeatedly emphasized that AI is entering the “thinking” stage, and the computing power consumption in each link of pre - training, post - training, and inference is increasing exponentially. “It is about to increase by a factor of one billion,” Jensen Huang said in a recent blog on BG2.

Especially during the National Day holiday, Sora 2, known as the AI version of TikTok, and its independent application were launched, with over one million downloads in five days, directly refreshing the upper limit of computing power demand. Sam Altman also posted that the user generation rate far exceeded expectations. Jensen Huang commented in an interview, “Look at all the generated videos. It's crazy.”

Sora 2 is a huge traffic pool, but OpenAI's goal is not just video generation. From the demonstration of ChatGPT at the developer conference, Sam Altman wants it all.

Ben Thompson, the founder of Stratechery, compared OpenAI to Windows in his article, indicating that they will evolve into a system.

This concept was once envisioned by Elon Musk, his “arch - enemy,” when he acquired Twitter. He hoped to turn it into a super - app.

Now, ChatGPT will be backward - compatible with all apps. Users will directly complete code programming, video production, shopping, etc. here. Such an app or system is the underlying logic for the ten - billion - fold increase in computing power.

Most people do not deny the existence of the bubble, but the story of growth in the bubble is also worthy of affirmation.

A Silicon Valley AI entrepreneur said, “It doesn't matter if it's a bubble. It's just a natural stage. Real entrepreneurs will do it no matter what the situation is. (The bubble) is what speculators care about.”

Lu Yi believes that only with a bubble can technology progress. “This is like credit. The greater the confidence, the more loans you can get.”

“(Investors) will definitely be scared, but when everyone is scared, the bubble won't burst.” An investor who tracks tech stocks said when talking about the bubble. In his opinion, everything depends on OpenAI and Musk's robots. If this vision cannot be realized within 1 - 3 years, a huge bubble will form. Now, these two companies have tied NVIDIA and AMD to them. “It's like tying up ships with iron chains. If there's a problem, the fire will spread for 800 li.”

A research report by US market research firm Bernstein also mentioned that Sam Altman has the ability to set the global economy back by decades. However, they also emphasized that he also has the opportunity to lead humanity into a dreamland, but it's not clear which is more likely to happen.

In an article titled “How big is the artificial intelligence bubble,” Charu Chanana, the chief investment strategy analyst at Saxo Bank, quoted the bullish view and emphasized: “This bubble is different.” She wrote, “Despite the soaring stock price and market value of NVIDIA, its valuation multiple has decreased. NVIDIA's expected price - to - earnings ratio is 32 times, lower than 36 times at the end of 2022. This is not a sign of weakness - it reflects that the soaring profitability is catching up with the growth of the stock price.”

Don't fight against capital expenditure

“A bubble means there is production but no orders,” said a researcher who has long tracked Sino - US tech policies.

In other words, under the internal cycle logic of “equity for orders,” NVIDIA and AMD have orders, and AI companies can obtain chips. Since both supply and demand exist, there is no bubble.

In terms of supply, TSMC's monthly CoWoS production capacity is estimated to be 900,000 wafers, and NVIDIA's demand accounts for half of it. According to the dual - die design of the GB200, theoretically, 6.75 million GPUs can be produced in 2025. This figure is basically consistent with the revised annual production estimate of 6.9 million GPUs by UBS. Based on the assumption that 1GW corresponds to 500,000 GPUs, NVIDIA's GPU production capacity in 2025 corresponds to a data center capacity of nearly 14GW.

In terms of demand, according to statistics from The Information, OpenAI's computing power agreements exceed $1 trillion. Based on the assumption that the hardware cost of a 1GW data center is $20 billion, it corresponds to a hardware demand for a 50GW data center.

NVIDIA's annual supply is far from meeting OpenAI's demand, not to mention the demand from star companies such as Google, Meta, Amazon, and Anthropic.

“The arms race has not reached its peak yet. Currently, only OpenAI has fired the first shot.” The aforementioned tech - stock investor said.

Now, everyone knows that OpenAI and computing power suppliers have adopted the “equity for orders” model. Regarding the question of “where the money comes from,” Sam Altman previewed in an earlier interview that “in the near future, OpenAI's investment in data center construction is expected to reach trillions of dollars. We can design a very interesting new type of financial instrument to finance computing - power - related businesses.”

Potential financial instruments also include data center REITs - packaging the completed data centers into securities products and recovering construction funds by selling asset rights.

“Oracle and OpenAI are turning their capital cooperation into REITs,” a senior industry insider said earlier. “(Through) secondary distribution, it is a mature derivative. Its prospects depend on the US financing environment.”

OpenAI, which is short of funds, relies on various financial instruments, while well - funded Silicon Valley giants are generally increasing their annual capital expenditures.

According to relevant performance data, the total AI capital expenditures of Meta, Alphabet, Amazon, and Microsoft in 2025 will be close to $300 billion, and the upward trend will continue in 2026, with an expected increase to over $400 billion.

Regarding the growth trend of capital expenditures by tech giants, Bobby Molavi, a managing director and senior trader at Goldman Sachs, wrote in a recent research article, “This trend of capital investment is like a giant ship. It takes time to start and stop, and it cannot turn overnight. Therefore, in the five - year investment cycle, the performance of one quarter hardly matters.”

Bobby Molavi said that in the past, people always emphasized not to fight against the Federal Reserve. Now it's “don't fight against capital expenditure.”

The discussion about the bubble has always been there, but tech companies are still making moves.

Sam Altman recently hinted in a podcast interview with a16z that the aggressive investment in computing power stems from the potential of the next - generation models. “If it were only based on the current models, we wouldn't go this far,” Altman said.

Now, the market has pushed NVIDIA's market value to over $4 trillion based on an AI assistant with thinking and reasoning abilities. No one knows what the next - generation models can do, but if one day AI can cure cancer, become the core labor force, and achieve miracles in life sciences, will it be a story worth trillions of dollars?

Regarding the growth space of AI, Apple is also a company worthy of attention. This giant with over 2 billion active users has just started to make efforts.

This article is from the WeChat official account “Tencent Technology.” Author: Su Yang, Editor: Zheng Kejun. Republished