OpenAI data center pivot underscores Wall Street IPO concerns

When OpenAI CEO Sam Altman took the stage at BlackRock’s US Infrastructure Summit earlier this month, he acknowledged that his company faces a harsh reality: data centers are tough.

“Anything on this scale is like a lot of things going wrong,” Altman said during a fireside chat at the conference in Washington, D.C.

Altman gave an example of a severe weather event that temporarily “knocked things up” at a data center campus in Abilene, Texas. The facility serves as OpenAI’s flagship site. Seer and SoftBank’s $500 billion Stargate project. Altman said his company is also struggling with supply chain challenges and pressure to meet tight deadlines.

Shares are rising for Altman, which aims to transform OpenAI, which valued it at $730 billion in a record fundraising round last month, from a darling of the private market into an investable asset for a more discerning class of public market fund managers. That meant backing away from some heavy spending plans, shelving some ambitious projects, and accepting OpenAI’s role as a buyer of huge amounts of cloud capacity rather than a builder of giant data centers.

“OpenAI realized that the market didn’t appreciate the reckless approach to growth and spending,” Futurum Group CEO Daniel Newman told CNBC in an interview. “The market wants to see OpenAI’s revenues grow at a pace where spending can be justified. To me, the key has been to try to show a little more fiscal responsibility.”

The strategic shift means OpenAI will have to be willing to do less while trying to compete with Anthropic. Google and a number of other companies developing AI models, applications and features. OpenAI; The chips train and run AI models, which require huge amounts of computing resources, including processing power, memory, and energy. Altman and other OpenAI executives have emphasized for years that computing is a major bottleneck for the company; The company has continued to raise astronomical amounts of cash, including $110 billion earlier this year, with $50 billion coming from revenue. Amazon.

One Publish on X In November, Altman wrote that OpenAI and other companies “have to downgrade our products and not introduce new features and models because we face such a severe computing constraint.”

The big story of the past year for OpenAI up to this point has been Altman’s extreme efforts to secure capacity. The company has signed multibillion-dollar infrastructure deals with many companies, including: Nvidia, Advanced Micro Devices And broadcom. Altman said in his November post that OpenAI was looking at rough commits. $1.4 trillion over the next eight years.

The deals have rattled public markets, sparked fears of a potential AI bubble and left many investors questioning how OpenAI can afford to deliver on such eye-popping commitments with $13.1 billion in revenue for the year.

OpenAI’s most notable announcement was this: 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 several years; The capital allocation was tied to the formation of OpenAI and the use of Nvidia technology. OpenAI has said it plans to build at least 10 gigawatts of Nvidia systems, with the initial $10 billion investment coming with the completion of the first gigawatt, a unit of power roughly comparable to the electricity consumption of a mid-sized city.

Press release He 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 seller financing that fueled the dot-com bubble in the late 1990s. Altman has repeatedly ignored concerns about OpenAI’s ambitious infrastructure plans, suggesting revenue will reach hundreds of billions by 2030.

But in recent months, OpenAI has tempered expectations and outlined a more measured strategy as the company prepares for a possible IPO later this year. OpenAI told investors in February that it was targeting about $600 billion in total computing spending by 2030; This means the figure is more directly linked to expected revenue growth.

The company also emphasizes discipline in other areas of its business. In December, OpenAI declared a “code red” to focus on developing its ChatGPT chatbot in the face of growing competition from Google and Anthropic.

OpenAI’s CEO of applications, Fidji Simo, held a wide-ranging meeting with employees about enterprise business earlier this month and said the company was “aggressively pushing” toward high-throughput use cases.

“What’s really important for us right now is to focus and execute extremely well,” Simo said, according to a partial transcript of the meeting reviewed by CNBC.

‘This is a race’

OpenAI does not currently have any data centers and may not in the foreseeable future, according to people familiar with the matter who asked not to be named because they were not authorized to speak publicly.

Instead, it prefers to rely heavily on partners such as Oracle. Microsoft and Amazon is trying to bundle as much capacity as possible.

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

According to the statement, OpenAI will be responsible for project operations and SoftBank will be responsible for financing. a blog post In that case. Oracle and Nvidia were selected as key initial technology partners.

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

As Stargate got underway, OpenAI prepared to develop much of the project itself and aimed to lease or own some data center campuses outright. Information. But the company changed course after facing practical construction problems and struggling to get support from lenders.

Oracle is leasing Stargate’s data center campus in Abilene and financing the construction with tens of billions of dollars. debt.

OpenAI and Nvidia said in their September publication that the first gigawatts of Nvidia systems would be deployed in the second half of 2026. Experts said the timeline would be challenging at best.

Walid Saad, an engineering professor at Virginia Tech, said it could take three to 10 years to build a 1-gigawatt data center from start to finish. Challenges can arise every step of the way, from finding a site, obtaining the appropriate permits and permits, accessing power, building the physical structure, delivering the hardware, and finally getting it online.

“There are regulations, there are permits, different places have different processes,” Saad said. “There are processes they can’t control. You never know what might come up.”

These obstacles have now become very real for Arun Chandrasekaran, an artificial intelligence analyst for OpenAI. Gartner’she told CNBC in an interview.

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

OpenAI had no comment for this story.

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

Nvidia also contributed to OpenAI’s funding round by investing $30 billion. OpenAI said it is expanding its collaboration with Nvidia as part of the deal, agreeing to use 3 gigawatts of dedicated inference capacity and 2 gigawatts of training capacity in Nvidia’s upcoming Vera Rubin systems.

“OpenAI is doing what it’s supposed to do, which is gaining access to computing at scale,” Futurum Group’s Newman said. MetaAnthropic and Google do the same. “This is a race.”

Nvidia’s investment follows months of speculation about the status of the major infrastructure deal the companies announced in September. The chip maker announced in a statement quarterly filing It reported in November that the $100 billion deal might not happen, and The Wall Street Journal reported in January that the deal “on ice“

Nvidia stated in a February filing that there was “no assurance” that the company would “enter into an investment and partnership agreement with OpenAI or complete a transaction.”

At a conference earlier this month, Nvidia CEO Jensen Huang tempered expectations further, saying the opportunity to invest $100 billion in OpenAI was probably “not in the cards.”

The latest investment is not tied to any distribution phase and differs from the deal structure the companies put forward six months ago. Huang said it “might be the last time” Nvidia invests in OpenAI before its IPO.

Newman said of OpenAI: “They’ve built an incredible growth story of their own. It’s just that the rest of the journey won’t be free.” “And because cost structures are so high, paths to profitability will be examined at every step.”

–CNBC’s Kate Rooney contributed to this report

WRISTWATCH: OpenAI renews focus on enterprise in all-hands meeting amid IPO pressure