The Inside Story of AI's "Trillion-Dollar Closed Loop": The "Offensive and Defensive Strategies" of Contemporary Tech Giants like Altman, Jensen Huang, Nadella, and Masayoshi Son

OpenAI CEO Sam Altman is firmly binding the world's largest tech giants to his artificial intelligence vehicle through a series of carefully orchestrated deals. By cleverly leveraging these giants' ambitions and anxieties, Altman has not only secured seemingly endless computing power for OpenAI but also built a massive "too big to fail" business ecosystem that closely ties the fate of the entire tech industry to this unprofitable startup.

In recent months, a series of deal storms led by Altman have swept through Silicon Valley. In January of this year, SoftBank's Masayoshi Son and Altman announced a $500 billion "Interstellar Gateway" project at the White House, followed by NVIDIA proposing a similar project to OpenAI, ultimately finalizing a $100 billion collaboration last month; Oracle's stock surged nearly 40% after signing a $300 billion contract, making its chairman Larry Ellison the world's richest person for a time; AMD and Broadcom also announced deals worth tens of billions of dollars.

These collaboration announcements triggered a strong reaction in the capital markets, bringing a total market value increase of $630 billion to the related companies on the first trading day after the news broke, repeatedly driving U.S. tech stocks to set historical highs.

This high-stakes gamble around computing power is fundamentally a "finite game" initiated by Altman. He successfully brought one company after another into the game by exploiting the "fear of missing out" (FOMO) among competitors. However, this financial "one mountain higher than another" has also intensified market concerns: Is the enthusiasm for artificial intelligence evolving into a massive bubble that relies solely on one company—or even one person's vision?

As OpenAI signs computing power bills amounting to hundreds of billions of dollars while its annual revenue is only $13 billion, this enormous financial leverage is testing the nerves of all participants. Some deals even exhibit a "circular" characteristic, where partners invest to help OpenAI purchase their own products. Behind this is Altman's firm belief in the exponential growth logic of computing power; he has told employees that the company's long-term goal is to build 250 gigawatts of computing power by 2033, a massive plan that could rival Germany's total electricity consumption.

Masayoshi Son's Ambition: Igniting the FOMO Tipping Point

Before igniting this arms race for computing power, Altman faced setbacks. Last year, he proposed to Microsoft CEO Satya Nadella that Microsoft invest at least $100 billion to build a new data center for OpenAI called "Interstellar Gateway," but was rejected. Similarly, his $70 trillion global chip manufacturing plan proposed to TSMC also failed to gain support, with TSMC CEO C.C. Wei bluntly stating at the 2024 shareholders' meeting that Altman was "too aggressive and hard to believe."

After facing repeated obstacles, Altman turned to SoftBank CEO Masayoshi Son. This risk-taking investor, eager to make a significant bet in the AI field after experiencing investment failures like WeWork, recalled at an event in Tokyo that when Altman emphasized to him, "Without more computing power, everything (the grand vision of AI) is impossible," he began to think, "If more is better, then we should do more."To facilitate the deal, Masayoshi Son invited Altman to his mansion in the suburbs of Tokyo for several days last November. In this estate, referred to as "Versailles" by his colleagues, the two reached a consensus. Subsequently, they jointly announced a plan to invest $500 billion in the "Stargate" project at the White House, with SoftBank leading the financing and construction of data centers for OpenAI.

Despite media reports indicating that the joint venture project encountered some obstacles due to site selection disagreements, it successfully ignited market enthusiasm. SoftBank's stock price rose 11% following the announcement, and more importantly, it triggered a "fear of missing out" across the industry, paving the way for Altman's subsequent plans.

Microsoft's "Resilience": From Bystander to Returning to the Table

As OpenAI's exclusive cloud partner for six years, Microsoft was once its most steadfast supporter behind its rise. However, faced with Altman's endless demand for computing power, hesitation began to emerge within Microsoft. In January this year at Davos, when asked about the $500 billion Stargate plan, Nadella responded with a smile, "I only know that my $80 billion is fine," referring to Microsoft's data center budget for the fiscal year.

Nadella's cautious attitude was not an isolated case. He stated in a podcast interview:

"At some point, supply and demand must match. When you only numb yourself with the excitement of the supply side without truly understanding how to translate it into customer value, you may completely go off track."

Subsequently, a report from brokerage TD Cowen indicated that Microsoft had canceled some data center leases in the United States, which was interpreted as a reduction in support for OpenAI's single workload.

Microsoft granted OpenAI permission to seek other cloud service providers while turning to attract a more diversified customer base. At this time, Oracle seized the opportunity and signed a $300 billion contract with OpenAI. Upon the news, Oracle's stock price soared nearly 40%, bringing its market value close to the $1 trillion mark.

This victory led some Microsoft executives to privately criticize, believing that Oracle might not have the capability to complete such a massive project. However, the market's enthusiastic reaction could not be ignored. A week after Oracle's stock price surged, Nadella publicly announced the "world's most powerful AI data center" located in Wisconsin, with a clear indication that part of it would be used for training OpenAI models. According to insiders, Microsoft is currently discussing providing additional computing power for OpenAI.

According to a previous article by Jianwen, OpenAI's almost arrogant ambition convinced Microsoft that rather than taking on astronomical financial risks, it was better to let competitors take the gamble. Moreover, despite loosening exclusive terms, Microsoft remains the only cloud service provider offering supercomputers for training OpenAI models, which is the most critical aspect of AI developmentAt the same time, Microsoft continues to take a 20% cut from OpenAI's soaring revenue.

Jensen Huang's "Aggression": From Supplier to Deep Bundling

For NVIDIA CEO Jensen Huang, seeing Altman and Masayoshi Son announce a historic AI infrastructure project at the White House is undoubtedly a stimulus. According to media reports citing sources familiar with his thoughts, as the chip supplier that has provided core power to OpenAI for the past decade, Huang hopes to stand beside Altman.

Soon, NVIDIA secretly proposed a similar project to OpenAI, intending to replace SoftBank and help raise the funds needed to build new data centers. Ultimately, the two sides reached a massive $10 billion deal last month, which Huang called "the largest computing project in history."

However, the negotiation process was not smooth sailing. Reports indicate that discussions between the two sides stalled during the summer. The turning point occurred on June 27, when tech media The Information reported that OpenAI had begun renting Google's TPU chips to support ChatGPT. This news sent shockwaves through NVIDIA. According to insiders, Huang immediately called Altman to inquire about the authenticity of the report and expressed his willingness to get negotiations back on track.

The final agreement far exceeded initial expectations. NVIDIA agreed to lease up to 5 million chips to OpenAI, with costs reaching as high as $350 billion at current market prices. At the same time, NVIDIA has the right to invest up to $100 billion to help OpenAI cover this expense.

More notably, according to media reports citing informed sources, a previously unreported arrangement is that NVIDIA is discussing providing guarantees for part of the loans to build its own data centers for OpenAI. This means that if OpenAI fails to repay the debt, the chip giant will face billions of dollars in debt risk, binding the fates of both parties more deeply than ever before.

The Influx of Competitors: AMD and Broadcom's Gamble

Altman's "FOMO" strategy is contagious. AMD CEO Lisa Su is also eager to secure an "AI highlight moment" for her company. At AMD's "Advancing AI" conference in June, Altman was the closing guest, and Su posed a symbolic question to him on stage:

"Can I call you my AI idol?"

At that time, AMD was striving to catch up with NVIDIA, and OpenAI's endorsement was crucial. The two companies closely collaborated on AMD's latest chip, the MI450, and began negotiations for a deal worth billions of dollars. To seize this opportunity, Su was willing to offer up to 10% of AMD's future stock as a reward, effectively providing OpenAI with a substantial subsidy. After the deal was announced on October 6, AMD's stock surged 24% in a single day, marking one of the largest single-day gains in history.

The number of participants in this competition is still increasing. Shortly after NVIDIA announced its deal, Broadcom also announced an agreement with OpenAI to jointly develop new chips and computing systems, providing 10 gigawatts of computing power, comparable in scale to NVIDIA's agreement.

Every time a giant enters the arena, it forces other players to accelerate their follow-up actions, further betting on the future of OpenAI, and making this "trillion-dollar closed loop" crafted by Altman even more robust