How Sam Altman Tied Tech’s Biggest Players to OpenAI

Jensen Huang was wrapping up a trip celebrating the Lunar New Year in Asia this past January when Masayoshi Son, the chief executive of SoftBank , stood alongside Sam Altman at the White House to unveil what was billed as the largest AI infrastructure project in history.

For almost a decade, the Nvidia chief executive had supplied the AI chips that powered OpenAI’s rise. Huang wanted to be the one unveiling such a giant deal with OpenAI’s CEO, according to people familiar with his thinking.

Læs også: Derfor kan du nu læse artikler fra The Wall Street Journal på Euroinvestor

Soon after, Nvidia secretly pitched OpenAI on a similar project, offering to effectively sideline SoftBank and help raise the funds needed to build new data centers itself. The talks culminated in a giant $100-billion deal between the two companies, announced at Nvidia’s Santa Clara headquarters last month.

“This is the largest computing project in history,” Huang said.

Altman might not have intended to give Huang FOMO, or the fear of missing out, but he had that effect. And it is getting contagious.

To achieve his vision of securing seemingly endless computing power for OpenAI, Altman has gone on a dealmaking blitz, playing the egos of Silicon Valley’s giants off one another as they race to cash in on OpenAI’s future growth.

The resulting game of financial one-upmanship has tied the fates of the world’s biggest semiconductor and cloud companies—and vast swaths of the U.S. economy—to OpenAI, essentially making it too big to fail. All of them are now betting on the success of a startup that is nowhere near turning a profit and facing a mounting list of business challenges.

Investors aren’t bothered.

On four separate days over the past two months, the stock prices of Oracle , Nvidia, Advanced Micro Devices , and Broadcom soared after they disclosed OpenAI-related deals—adding a combined $630 billion to their market value in the first day of trading after the announcements. A broader rally in tech stocks followed each time, helping lift the U.S. stock market to record highs.

“The most successful people I know believe in themselves almost to the point of delusion,” Altman once wrote in a 2019 blog post titled “How To Be Successful,” before adding: “Self-belief alone is not sufficient—you also have to be able to convince other people of what you believe.”

Ever since the viral launch of ChatGPT, Altman has pumped up AI’s world-changing potential more than any other tech CEO, prophesying a world where the technology finds a cure to cancer, provides customized tutoring to every student on earth, and becomes an endless gusher of profits for the companies behind it.

He wants OpenAI to be at the center of this transformation, and recently told employees that the company’s long-term goal was to build 250 gigawatts worth of computing capacity by 2033 in order to make it happen. Such a plan would cost more than $10 trillion by today’s standards and be enough to power a mid-sized country like Germany.

OpenAI is set to generate $13 billion in revenue this year—a tiny fraction of the $650 billion in computing bills that it is signing up for between its Nvidia and Oracle deals alone, according to calculations made by The Wall Street Journal. The cost could be closer to the trillion-dollar mark when factoring in agreements with AMD, Broadcom, and other cloud providers like Microsoft .

The commitments to build such massive amounts of chips and data centers before OpenAI can afford it are fueling fears that enthusiasm for AI is turning into a bubble hinging on the success of just one company—and increasingly the vision of one person. Some of Altman’s partners are even helping OpenAI pay for their chips, creating circular deals that prop up demand.

Altman says that OpenAI is dealing with computing shortages so severe that it is having to delay product releases as a result. He believes that revenue will grow with more computing power. This month, OpenAI’s new Sora video app rocketed to the top of the App Store, showing that the company’s penchant for building viral products goes beyond ChatGPT.

“I have learned over my career, again and again and again, you just have to trust the exponential,” Altman said at an event in Tokyo earlier this year. “We’re not built to conceptualize it, but you just have to trust.”

Breakfast at ‘Versailles’

The 40-year-old AI pioneer has never been afraid to bet big. He dropped out of Stanford University to start his first company, at one point using his poker winnings to fund college expenses. After becoming rich, Altman put large chunks of his wealth in startups chasing far-flung technologies like cryptocurrency and nuclear fusion, and loaded up on debt to increase the size of his investments, convinced of their unlimited upside.

Other CEOs were not always on board with his AI ambitions.

Last year, Altman asked Microsoft chief executive Satya Nadella if his company would invest at least $100 billion building new data centers for OpenAI, calling the project Stargate. He said no.

He got the same response when pitching leaders at the Taiwanese chip-making giant TSMC on a $7-trillion project building new chip manufacturing plants across the world. Altman is “too aggressive for me to believe,” chief executive C.C. Wei said at the company’s annual shareholder meeting in 2024.

Altman then turned to Son, the risk-taking chief executive of SoftBank, who was looking to go big in AI after a series of bungled startup bets, including a failed $16-billion investment in WeWork.

Son had grown obsessed with the singularity, the hypothetical moment where AI becomes smarter than humans and transforms civilization, and was dreaming up ideas such as building AI-powered robot factories across the world. When Altman met Son for dinner in 2023, he emphasized that none of this would be possible without a lot more computing power.

“I started thinking, well, if more is better, we should do a lot,” Son recounted at a Tokyo event with Altman.

Hoping to strike a deal, Son invited Altman last November to stay for several days as a guest in his home near Tokyo, a sprawling, countryside estate lined with statues of Roman emperors that associates call Versailles for its striking resemblance to the former royal residence of the French King Louis XIV. (One entrance features a black-and-gold iron gate with a crest featuring an “M” with a crown on top).

Over plates of scrambled eggs, Altman laid out his vision for building a new cloud company with SoftBank that would finance and build data centers across the U.S. for OpenAI. Son agreed to lead what they later unveiled as a planned $500 billion effort, also called Stargate.

The joint venture recently hit some snags after the two sides disagreed over which sites to build on, The Wall Street Journal reported. OpenAI and SoftBank announced two new data-center sites last month, but they are a fraction of the capacity that OpenAI is securing elsewhere.

But Altman’s wooing of Son succeeded in kicking off an investment mania. SoftBank’s stock shot up 11% after the White House announcement, as did those of other tech partners named on the project. In the coming weeks and months, OpenAI received hundreds of proposals for potential sites to build on, setting the stage for its next act.

Microsoft cedes ground

In January, Microsoft’s Nadella was attending an unseasonably warm edition of the World Economic Forum in Davos, Switzerland when he was asked in a TV interview about the $500-billion Stargate announcement.

“Look, all I know is, I’m good for my $80 billion,” he said with a chuckle, referencing how much Microsoft said it would spend building data centers for the fiscal year. The sentiment echoed those shared by two of Altman’s biggest rivals, Anthropic CEO Dario Amodei and Elon Musk , both of whom cast doubt on the project.

For years, Microsoft’s stock had boomed thanks to its six-year partnership with OpenAI, which included its lucrative role as the startup’s exclusive cloud provider. But some executives inside its Redmond, Wash., headquarters were growing hesitant about funding the massive data-center build-out that Altman wanted, which seemed to increase with no end in sight.

In February, Nadella was asked in a podcast interview why he wasn’t investing $800 billion into data centers as opposed to $80 billion, given AI’s potential.

“At some point, the supply and demand have to map,” he responded. “You can go off the rails completely when you are hyping yourself with the supply-side, versus really understanding how to translate that into real value to customers.”

Two days later, analysts at the brokerage firm TD Cowen published a report stating that Microsoft had canceled some of its data-center leases in the U.S., citing its move away from supporting OpenAI workloads. It sparked fresh debate about whether spending on AI infrastructure had grown too extreme.

Microsoft gave OpenAI permission to find additional computing capacity from other cloud providers, and focused its efforts on courting other customers.

Then, Oracle swooped in to sign its own $300-billion contract with the startup. Its stock shot up by almost 40% after disclosing a massive uptick in future revenue thanks to the deal, putting it within striking distance of a $1-trillion valuation and briefly making its chairman Larry Ellison the world’s richest man.

It was a giant win for the 48-year-old legacy database provider, which had long struggled to establish itself in the cloud computing industry. Some Microsoft leaders privately criticized the deal, arguing that it wasn’t certain that Oracle would be able to pull it off. (The giant data-center build-out puts OpenAI on the hook for an average of $60 billion in yearly payments, more than 4 times its current revenue).

A week after Oracle’s stock spike, Nadella announced what he called “the world’s most powerful AI data center,” located in Wisconsin, that will be used in part to train OpenAI models, sharing a video online that showed how quickly Microsoft built the site.

Microsoft is now discussing providing additional computing capacity to OpenAI, according to people familiar with the matter.

A mysterious report

By the summer, the talks between OpenAI and Nvidia on their own AI infrastructure project had stalled.

Then, on June 27, the tech site The Information published a report stating that OpenAI had begun renting Google’s TPU chips to power ChatGPT, causing something of a stir inside Nvidia’s Santa Clara headquarters.

At the time, it was well-known that OpenAI was looking for ways to reduce its reliance on Nvidia. The startup was busy designing its own custom chip with Broadcom, and was also taking a close look at the chips offered by Nvidia rival AMD.

That spring, OpenAI also signed its first cloud contract with Google, raising the specter that it might begin using its chips as well.

After the story was published, Huang called Altman asking him if it was true, and signaled that he was open to getting the talks back on track, according to people familiar with the matter. Huang realized that Nvidia could help OpenAI by making a direct investment in the company, a person who spoke to him said.

Nvidia’s X account then posted a screenshot of an article wherein OpenAI denied having plans to use Google’s in-house chip.

Nvidia ended up signing an agreement to lease up to 5 million of its chips to OpenAI, costing $350 billion by today’s standards. It also has the right to invest up to $100 billion to help the startup pay for the deal.

As part of the deal, Nvidia is also discussing guaranteeing some of the loans that OpenAI plans to take out to build its own data centers, people familiar with the matter said—a move that could saddle the chip giant with billions of dollars in debt obligations if the startup can’t pay for them. The arrangement hasn’t been previously reported.

‘An AI icon’

Altman was the closing act at AMD’s splashy “Advancing AI” keynote, held at a giant convention center in downtown San Jose in June. After walking onstage to rapturous applause, he shared a quick hug with the chip-designer’s chief executive Lisa Su , before being hit with a somewhat awkward question.

“Can I call you an AI icon?” Su asked.

That day, Altman was dealing with a growing list of seemingly intractable problems, including a political fight over its for-profit restructure in California, that threatened to sink the company. But the rest of the tech world was certainly treating him like one—including Su.

“OpenAI has truly been at the center of the universe,” she went on to tell the crowd. “Everyone listens to what Sam Altman has to say.”

AMD had launched its first line of GPUs specifically designed for AI workloads, called Instinct, in 2018, but struggled to keep up with Nvidia. Su wanted a big AI moment, and felt she was getting close with Altman. OpenAI had worked closely with AMD on the design of its latest chip, MI450, and was optimistic about its potential capabilities.

When Su asked him if there would ever be enough GPUs, Altman mused that a significant fraction of the Earth’s power should be used to run AI.

The two companies began talks for a deal for OpenAI to spend tens of billions of dollars buying up to 6 gigawatts of capacity from AMD. Su was willing to give them up to 10% of AMD’s future stock as a reward for taking the chance on her unproven chips, effectively giving OpenAI a massive subsidy.

In a joint interview with Su, Altman, and OpenAI President Greg Brockman , the Journal asked Su whether she had any anxiety about how much of the tech world’s market cap was riding on OpenAI.

“No,” Su responded. “What keeps me up at night is getting as much compute as I can to Sam and Greg so that they can really unlock the power of AI.”

After AMD announced its deal with OpenAI on Oct. 6, its stock popped 24%, one of its largest-ever single day gains.

“Considering they were so excited about their next-generation product, I’m surprised that they would give away 10% of the company before they even built it,” Huang said about the deal on TV.

Soon after came another announcement: Broadcom said it agreed to build a new chip and computing system with OpenAI engineers. The two sides had been collaborating since early 2024 on the chip, but discussions for a deal accelerated after the Nvidia announcement.

Broadcom ended up matching the size of the Nvidia deal, offering 10 gigawatts of computing capacity for OpenAI.

Write to Berber Jin at berber.jin@wsj.com