OpenAI data center pivot underscores Wall Street IPO concerns

When OpenAI CEO Sam Altman took the stage at BlackRock’s U.S. Infrastructure Summit earlier this month, he acknowledged his company is facing a harsh reality: data centers are hard.

“Anything at this scale, it’s just like so much stuff goes wrong,” Altman said, in a fireside chat at the conference in Washington, D.C.

Altman gave an example of a severe weather event at a data center campus in Abilene, Texas, that temporarily “brought things down.” The facility serves as the flagship site of OpenAI, Oracle and SoftBank’s $500 billion Stargate project. Altman said his company has also been navigating supply chain challenges and pressure to meet tight deadlines. 

The stakes for Altman are growing as he aims to turn OpenAI, which was valued at $730 billion in a record fundraising round last month, from a private market darling into an investable asset for a more discerning class of public market fund managers. That’s meant retreating from some hefty spending plans, shelving certain ambitious projects and accepting OpenAI’s role as a purchaser of massive amounts of cloud capacity rather than as a builder of mammoth data centers.

“OpenAI has come to the realization that the market doesn’t necessarily appreciate the reckless approach to growth and spending,” Daniel Newman, CEO of Futurum Group, told CNBC in an interview. “The market wants to see OpenAI’s revenues rolling at a pace in which the spending can be justified. The pivot, in my opinion, has been to try to show a little bit more fiscal responsibility.”

The strategic shift means OpenAI may have to settle for doing less while simultaneously trying to compete with Anthropic, Google and a host of other companies developing AI models, apps and features. OpenAI trains and runs AI models that require enormous amounts of computational resources, including chips, processing power, memory and energy. Altman and other OpenAI executives have for years stressed that compute is a major bottleneck for the company, which has proceeded to raise astronomical sums of cash, including $110 billion earlier this year, with $50 billion coming from Amazon.

In a post on X in November, Altman wrote that OpenAI and other companies “have to rate limit our products and not offer new features and models because we face such a severe compute constraint.” 

Up to that point, the big story for OpenAI last year was the extreme lengths Altman went to secure capacity. The company inked a flurry of multibillion-dollar infrastructure deals with companies including Nvidia, Advanced Micro Devices and Broadcom. Altman said in his November post that OpenAI was looking at commitments of roughly $1.4 trillion over the next eight years.  

The deals rattled public markets, sparked fears about a potential AI bubble and caused many investors to question how OpenAI could afford to make such eye-popping commitments with $13.1 billion in revenue for the year. 

OpenAI’s most notable announcement was with Nvidia. The chipmaker, which is also the world’s most valuable company, agreed in September to invest up to $100 billion in the startup over a number of years, with capital distribution tied to OpenAI’s buildout and use of Nvidia’s technology. OpenAI said it planned to deploy at least 10 gigawatts of Nvidia systems, with the first $10 billion of investment arriving alongside completion of the first gigawatt, a unit of power that’s roughly comparable to the electricity consumption of a mid-sized city.

The press release said the partnership “enables OpenAI to build and deploy at least 10 gigawatts of AI data centers.”

Analysts told CNBC at the time that the deal was reminiscent of the vendor financing that fueled the dot-com bubble in the late 1990s. Altman repeatedly brushed off concerns about OpenAI’s ambitious infrastructure plans, suggesting that revenue would balloon into the hundreds of billions by 2030.

But in recent months, as the company has been gearing up for a potential IPO later this year, OpenAI has tempered expectations and outlined a more measured strategy. OpenAI told investors in February that it’s now targeting roughly $600 billion in total compute spend by 2030, a figure that’s meant to more directly tie to its expected revenue growth. 

The company is emphasizing discipline across other corners of its business as well. In December, OpenAI declared a “code red” to focus on improving its ChatGPT chatbot in the face of growing competition from Google and Anthropic. 

Fidji Simo, OpenAI’s CEO of applications, held an all-hands meeting with staffers earlier this month about the enterprise business, and said the company is “orienting aggressively” towards high-productivity use cases.

“What really matters for us right now is staying focused and executing extremely well,” Simo said, according to a partial transcript of the meeting reviewed by CNBC.

‘This is the race’

OpenAI doesn’t currently own any data centers, and may not for the foreseeable future, according to people familiar with the matter who asked not to be named because they weren’t authorized to speak publicly.

Instead, it’s opted to lean heavily on partners like Oracle, Microsoft and Amazon, trying to piece together as much capacity as possible.

A year ago, things looked very different for OpenAI. In January 2025, President Donald Trump unveiled the Stargate project alongside Altman, SoftBank CEO Masayoshi Son and Oracle Chairman Larry Ellison during an event at the White House. The companies pledged to deploy $500 billion over four years to build out new AI infrastructure in the U.S.

OpenAI would be responsible for project operations, while SoftBank would be in charge of the finances, according to a blog post at the time. Oracle and Nvidia were named as key initial technology partners.

“Oracle, Nvidia, and OpenAI will closely collaborate to build and operate this computing system,” the release said. 

As Stargate got underway, OpenAI was prepared to develop large portions of the project itself, and it aimed to directly lease or own some data center campuses, according to a report from The Information. But after the company came face to face with practical construction issues and struggled to secure backing from lenders, it pivoted. 

Oracle is leasing Stargate’s data center campus in Abilene, and has been funding the buildout by taking on tens of billions of dollars in debt.

OpenAI and Nvidia said in their September release that the first gigawatt of Nvidia systems will be deployed in the second half of 2026. Experts said that timeline would be tough in the best of circumstances.

Walid Saad, an engineering professor at Virginia Tech, said building a 1-gigawatt data center from start to finish could take anywhere from three to 10 years. Challenges can crop up every step of the way– from finding a site, securing proper permissions and permitting, accessing power, constructing the physical structure, delivering the hardware to finally bringing it online.

“There’s regulations, there’s permits, different locations have different processes,” Saad said. “There are processes they cannot control. You never know what pops up.”

Those obstacles have become very real for OpenAI, Arun Chandrasekaran, an AI analyst at Gartner, told CNBC in an interview.  

“They’re starting to say, ‘You know what, let’s try to secure the capacity that we can from the providers that are willing to give us that capacity now,'” Chandrasekaran said.

OpenAI didn’t provide a comment for this story.

As part of OpenAI’s $110 billion financing announcement last month, the company agreed to consume roughly 2 gigawatts of Trainium capacity through Amazon Web Services infrastructure. Trainium is AWS’ custom AI chip. Amazon announced the latest version, Trainium3, in December. 

Nvidia also contributed to OpenAI’s funding round, investing $30 billion. OpenAI said it expanded its collaboration with Nvidia as part of the deal, and agreed to use 3 gigawatts of dedicated inference capacity and 2 gigawatts of training capacity on Nvidia’s forthcoming Vera Rubin systems.

“OpenAI is doing what it must do, which is gain access to compute at scale,” Futurum Group’s Newman said, adding that Meta, Anthropic and Google are doing the same. “This is the race.”

Nvidia’s investment landed after months of speculation about the status of the major infrastructure deal that the companies announced in September. The chipmaker disclosed in a quarterly filing in November that the $100 billion deal may not come to fruition, and The Wall Street Journal reported in January that the agreement was “on ice.”

Nvidia noted in a February filing that there was “no assurance” that the company will enter into an “investment and partnership agreement with OpenAI or that a transaction will be completed.”

At a conference earlier this month, Nvidia CEO Jensen Huang reined in expectations even further, and said that the opportunity to invest $100 billion in OpenAI is probably “not in the cards.”

The latest investment is not tied to any deployment milestones, and is distinct from the deal structure the companies touted six months ago. Huang said it “might be the last time” Nvidia invests in OpenAI ahead of its IPO.  

“To their credit, they built an incredible growth story. It’s just – the rest of the ride won’t be a free one,” Newman said of OpenAI. “And because their cost structure is so high, their route to profitability will be scrutinized every step of the way.”

–CNBC’s Kate Rooney contributed to this report

WATCH: OpenAI renews focus on enterprise in all-hands meeting amid IPO push