The newly announced $500 billion Stargate Project should help address industry demands for more compute power needed for rapidly evolving AI workloads and the desire to keep the United States at the forefront of a highly competitive global AI market that includes China as its chief rival and pushes by the European Union (EU) and the UK to be significant players.
Announced at a White House event on President Trump’s second day of his return to office, the initiative creates a new company tasked with building new AI infrastructure for generative AI pioneer OpenAI in the United States, starting with a $100 billion project already underway in Texas and more sites around the country being evaluated.
OpenAI and Japanese multinational SoftBank are the top partners in the project, with OpenAI taking the lead in operations and SoftBank on the financial side, according to an announcement from OpenAI. Other equity partners are Oracle and MGX, a tech investment company that was created by the United Arab Emirates government last year, with AI infrastructure being a key focus.
Meanwhile, AI giant Nvidia, chip designer Arm, and Microsoft — which has invested billions of dollars in OpenAI over the past couple of years — are joining OpenAI and Oracle as initial technology partners.
A ‘Most Important Project’
OpenAI CEO Sam Altman, speaking at the White House event alongside Oracle founder and CTO Larry Ellison and SoftBank CEO Masayoshi Son — who will be the new company’s chairman — said the initiative “will be the most important project of this era.”
Altman, in the two-plus years since his company released ChatGPT and fueled the rapid acceleration in generative AI innovation and adoption, has been vocal about the need for more compute capacity for AI workloads, which data center solutions firm TierPoint noted “require complex computations, massive datasets, and a need to store and scale that greatly surpasses the needs of traditional workloads.”
Stargate is a good step in that direction, according to Bob O’Donnell, principal analyst with TECHnalysis Research, adding that the new company’s announcement at the White House “was a whole combination of things.”
“It obviously was intended to be a big announcement by the new administration, focused on a big topic of the day and focused on investing in infrastructure,” O’Donnell told The New Stack, noting the project in Texas was first written about by new site The Information. “But regardless of all of the politicking and PR, it addresses a real issue, which is there is a need for more data centers because AI needs a lot of compute power.”
AI’s Insatiable Demand for Compute
In a report last year, global consultancy McKinsey & Co. said the rise of AI is the key driver behind a global demand for more data center capacity that could grow annually by 19% to 22% between 2023 and 2030, reaching an annual demand of 171 to 219 gigawatts (GW).
“Our analysis suggests that demand for AI-ready data center capacity will rise at an average rate of 33% a year between 2023 and 2030 in a midrange scenario,” the authors wrote. “This means that around 70% of total demand for data center capacity will be for data centers equipped to host advanced-AI workloads by 2030.”
The major AI players have been investing billions of dollars to increase their AI compute capacity. Microsoft Vice Chair and President Brad Smith wrote this month that the company will spend about $80 billion to build AI data centers, with more than half the investment being in the United States. That’s apart from an investment partnership by Microsoft, MGX, and investment firms BlackRock and Global Infrastructure Partners to spend $100 billion to expand data center capacity and supporting energy infrastructure.
For its part, Amazon Web Services (AWS), the largest cloud services provider, said this month it was investing $11 billion to expand its infrastructure in Georgia to support cloud and AI workloads.
A Lot of Energy Needed
Such efforts will deliver more compute capacity for AI workloads, but they also bring concerns about the amount of power and water they consume. Global investment bank Goldman Sachs, in a report last year, noted that a ChatGPT query needs almost 10 times the electricity that a Google search consumes, adding that AI will drive a 160% increase in data center power demand by 2030.
TECHnalysis’ O’Donnell said that Stargate and other AI efforts no doubt cause power demand to spike but added that such initiatives also will likely spur industry efforts to develop less power-hungry technologies, such as new ways to train AI models or more efficient chip designs.
Some, including Microsoft, AWS, and Google, are investing in nuclear reactors to create clean, carbon-free electricity to be used for AI data centers. Constellation Energy, which owns the infamous Three Mile Island nuclear power plant, said it will reopen the site to help power Microsoft’s data centers.
“This agreement is a major milestone in Microsoft’s efforts to help decarbonize the grid in support of our commitment to become carbon negative,” Bobby Hollis, Microsoft’s vice president of energy, said in a statement.
Rob Enderle, principal analyst with The Enderle Group, said Stargate will have to build generating capacity along with the data centers, adding that the “best would be nuclear [power], where there have been several advancements in terms of relatively small and easy-to-maintain nuclear reactors.”
“Currently, AI implementations using Nvidia technology use warm-water closed-loop cooling, which isn’t as water-intensive as older water-cooling methods that used evaporative chillers,” Enderle told The New Stack. “For a data center project of this size, water use will be far lower than you’d typically expect.”
Jack Gold, principal analyst with J. Gold Associates, told The New Stack that power will be a significant issue with these data centers. There already is an issue finding power to existing and proposed data centers, with many municipalities refusing to allow them in their territory due to power constraints. In addition, Gold’s research indicates that a modest AI data center with 100,000 GPUs use enough electricity to power a small city of more than 50,000 residents.
And don’t talk to him about nuclear power.
“I don’t buy into the whole nuclear thing, even if it is a good solution,” Gold said. “It is too long to get approved and too expensive to build, and the new micro-reactors are still not production-ready or proven in real-world situations.”
Keeping the US in the Lead
OpenAI and Trump talked about the need to ensure the United States continues to be the AI leader in the world. According to the EU, private investments in AI in the United States were far ahead of other countries, spending more than $64 billion in 2023, compared with more than $7 billion in China. However, many companies in China and other countries benefit from huge government investments.
The effort got a nod from the Information Technology and Innovation Foundation (ITIF), with its vice president, Daniel Castro, saying in a statement that “by committing up to $500 billion to build state-of-the-art AI infrastructure, these companies are demonstrating their confidence in the U.S. as a hub for innovation. This partnership reflects a shared goal: creating the foundation for the U.S. to thrive in the AI economy.”
Elon Musk, who’s been at Trump’s side since before the election and owns his own AI firm, xAI, but is not mentioned in the Stargate news, is casting doubt on the initiative, saying in postings on X (formerly Twitter) that he doubts SoftBank has the money, writing that “SoftBank has well under $10B secured. I have that on good authority.”
Musk isn’t the only one wondering about the money. Gold pointed to the $500 billion for the company, noting that there have been other big announcements about major projects that never materialized, using Trump’s promise in his first term of the Foxconn manufacturing facility in Wisconsin that was never built.
He also said that such massive AI data centers can cost $5 billion to $10 billion to build and another $4 million to $8 million a month in power costs to run them. That adds up quickly, the analyst said.
On the Outside
With Stargate getting a lot of headlines, it’s interesting to note a lot of the IT and AI companies that aren’t part of project, such as cloud giants AWS and Google Cloud Platform. O’Donnell noted Oracle’s inclusion as a cloud provider through its partnership with OpenAI. He also pointed out that Nvidia and Arm are named as tech collaborators, while Intel and AMD are not.
In the announcement, OpenAI also noted changes in its partnership with Microsoft, saying it will continue to grow its use of Azure infrastructure even while working with Oracle. For its part, Microsoft wrote in a blog post that the “key elements of our partnership remain in place” through their contract, which run until 2030. However, there are changes regarding exclusivity for new compute capacity, with a new model that gives Microsoft the right of first refusal.
“To further support OpenAI, Microsoft has approved OpenAI’s ability to build additional capacity, primarily for research and training of models,” Microsoft wrote.
The inclusion of top-tier AI and IT companies in massive projects isn’t surprising but is an indication of the new administration’s view that such vendors will push the country forward in the rapidly expanding global market. It’s a change from the view his predecessor, Joe Biden, put forth in his sweeping AI executive order issued in 2023 (and revoked by Trump on his first day in office), which, among other points, said that the U.S. government should ensure that smaller businesses and entrepreneurs could get footholds in AI.
Enderle said, “This concentration of power [among so few vendors] is problematic. AI is very fluid at the moment, so putting everything in one AI basket could be problematic. However, Nvidia is part of this effort, and they clearly know of this problem and how to mitigate it.”
Gold agreed, adding that having three IT companies driving Stargate may dilute what others are doing.
“There may indeed be new algorithms and technology that’s not from Nvidia and Arm that could be revolutionary, but this concentration of power in this group may essentially lock them out,” he said. “Don’t get me wrong. I think it’s critical that the U.S. invest in AI and lead the rest of the world. I’m just not sure this is the best way to do so given so many questions. I hope this wasn’t just a hollow stunt, but I’m looking for real proof points before I become a cheerleader.”
For his part, O’Donnell was more optimistic. This is only the beginning, he said, adding that he expects other initiatives like Stargate to launch among major vendors and that smaller companies would get access to the compute power in some of these data centers.
“This won’t be the only one,” he said of Stargate. “This is one big spike in the ground. We’re going to see others.”
The post AI Power Needs May Strain US Grid as Stargate Project Grows appeared first on The New Stack.
OpenAI, Oracle, and SoftBank join forces in an unprecedented $500 billion initiative to dominate global AI infrastructure, with strong backing from the Trump administration. But is there enough power?