Keith Haide is on the scene in Abilene, Texas, with an open Stargate infrastructure build-out underway. Heyde, the former head of AI computing at Meta, is currently leading the push for Openai’s physical expansion.
Openai
It wasn’t something that Keith Hayed imagined celebrating the holiday. Rather than wandering around with his wife in Oregon, Haide visited potential data center sites across the US in late December
Two months ago, Heyde left Meta Join Openai as head of infrastructure. His work was looking for a vast lot suitable for a vast facility that would turn CEO Sam Altman’s ambitious computing dream into reality and be packed with powerful graphics processing units to ultimately build a large-scale language model.
“Between Christmas and last year’s New Year, it actually took most of the time to look at the site,” Hayd, 36, told CNBC in an interview. “So my family loved it, trust me.”
His life in 2025 has only become even more intense.
Since January, Openai has quietly solicited and reviewed suggestions from around 800 applicants who are hoping to host the next wave of AI supercomputing hubs designed to train more and more powerful models.
Currently, around 20 sites are in a sophisticated stage of hard work, with large land under examination in the Southwest, Midwest and Southeast. Heyde said the tax incentives are “a relatively small part of the decision matrix.”
The most important factors are access to electricity, capacity to expand, and buy-in from the local community.
“Can we build right away, will the power ramp be faster there? Is this making sense from a community perspective?” he said.
Heyde leads site development within Openai’s Industrial Compute team. This division will become one of the most important groups within the company. The once supported infrastructure has been promoted to a strategic pillar on par with product and model development.
With traditional data centers close to their maximum capacity, Openai bets that owning the next generation of physical infrastructure is the centre of controlling the future of AI.
Energy demand is difficult to estimate. Gigawatt data centers require the amount of energy needed for some cities. Later last month, Openai announced plans to work with Oracle, Nvidia and SoftBank to build out the 17 Gigawatt build-out.
Heyde said the new site should include all sorts of energy options, including battery-assisted solar power generation, legacy gas turbine renovations, and even small modular reactors. Each site looks different, but together the industrial backbone Openai needs to be expanded.
“We did this incredible bottleneck analysis to see what kind of energy sources actually can unleash the journey we want to take part in,” Hayd said.
A significant portion of the capital comes from Nvidia. Chipmaker has agreed to invest up to $100 billion in fuel supply in Openai’s expansion. This includes purchasing millions of Nvidia GPUs.
“Perfect wasn’t the goal.”
Heyde, former head of AI computing Metahelped oversee the build-out of Meta’s first 100,000 GPU cluster.
In addition to power, Powerai assesses how quickly it can build on a site, its workforce availability and proximity to supportive local governments, according to Stargate’s proposal request.
Heyde said the team has made about 100 site visits and has a short list of sites in late stage reviews. Some will be new builds, while others will need to be converted and renovated from existing facilities. Flexibility is important.
“The perfect parcel is mostly photographed,” Hayd said. “But we knew that perfect wasn’t the goal. Our goal was the number one, persuasive power ramp.”
The competition is fierce.
Meta is building what could be the largest data center in the Western Hemisphere. It is a $10 billion project in northeast Louisiana that drives billions of dollars in state incentives. CEO Mark Zuckerberg raised the company’s top annual capital expenditure range to $72 billion in July.
Steel frames from the data center under construction during a tour of the Openai data center in Abilene, Texas, USA on September 23, 2025.
Shelby Tauber | Reuters
Amazon Humanity is working together on Indiana’s 1,200-acre AI campus. And across the country, states are promoting tax cuts, power guarantees, and zoning approvals to attract the next big AI cluster.
Openai is a relative startup, which has been around for just 10 years and has been known only in the mainstream since it launched ChatGpt within three years. But it is grown from something like that Microsoft And, in addition to Nvidia, SoftBank is heading towards a $500 billion valuation.
And Openai shows that it is not afraid to lead the way with AI. The self-driving solar campus in Abilien, Texas is already live.
Openai is still leaning towards partners like OracleOpenai’s Chief Financial Officer Sarah Friar told CNBC last week that ownership of first-party infrastructure would provide a differentiated approach. It suppresses vendor markup, protects key intellectual property, and follows the same strategic logic that drives Amazon to build Amazon web services rather than relying on existing infrastructure.
However, Heyde showed that there is no real playbook when it comes to AI, especially since companies are pursuing artificial general information (AGIs), or because they pursue AI that can meet or exceed human capabilities.
“When you think about the types of delivery that have to happen in these places, it’s a very different size,” he said.
Some applicants including the former Bitcoin Mining operators provided existing power infrastructure, such as substations and modular build-outs, but Heyde said they are not always fit.
“I found it pretty much great that it was the first interaction in the community at times,” he said. “It’s a really great story bringing data centers and infrastructure on Openai’s behalf.”
The 20 finalist sites represent phase 1 of a much larger build-out. Openai will eventually expand from a single gigawatt project to a larger campus.
“We really thought about the viability and our own belief that the places and sites we are moving forward can provide the power and infrastructure stories associated with those sites,” says Heyde.
He understands why many people are skeptical.
“That’s difficult. There’s no doubt about it,” Hayd said. “The numbers we’re talking about are very challenging, but certainly possible.”
Surveillance: Openai’s $850 billion buildout is fighting against grid restrictions
