Keith Heyde, a former Meta executive, has spent the past 10 months leading OpenAI’s nationwide search for sites to build its Stargate data centers—massive AI supercomputing facilities designed to power the next generation of large language models. His journey began in December, when instead of celebrating the holidays with his wife in Oregon, he was touring potential locations across the United States, assessing land for future AI infrastructure. Heyde, 36, joined OpenAI in late 2024 as head of infrastructure, tasked with turning CEO Sam Altman’s vision for unprecedented compute capacity into reality. His mission: secure vast tracts of land capable of hosting tens or even hundreds of thousands of GPUs, the backbone of modern AI training. Since January, OpenAI has reviewed proposals from roughly 800 applicants across the U.S., narrowing the field to about 20 sites in advanced stages of evaluation. These are spread across the Southwest, Midwest, and Southeast. Heyde emphasized that while tax incentives play a role, they’re not the deciding factor. The real priorities are access to reliable, scalable power, the speed at which construction can begin, and strong community support. “We’re not looking for perfection,” Heyde said. “The perfect parcels are largely taken. Our goal is a compelling power ramp—can we get the electricity we need, fast?” Heyde brings deep experience from his time at Meta, where he led AI compute initiatives, including the rollout of one of the company’s first 100,000-GPU clusters. Now, within OpenAI’s industrial compute team, infrastructure has become a core strategic pillar—on par with product and model development. With traditional data centers nearing capacity, OpenAI is betting that owning its own infrastructure is essential to maintaining control over AI progress. The company is building a self-sustaining solar-powered campus in Abilene, Texas, already operational, as a model for future sites. While OpenAI is still relatively young—only a decade old and mainstream only since ChatGPT’s launch—its $500 billion valuation, backed by major investors like Microsoft, SoftBank, and Nvidia, gives it the financial muscle to compete. But the competition is fierce. Meta is constructing a $10 billion data center in Northeast Louisiana, one of the largest in the Western Hemisphere, backed by significant state incentives. Amazon and Anthropic are collaborating on a 1,200-acre AI campus in Indiana. States across the country are offering tax breaks, power guarantees, and fast-tracked zoning to attract AI infrastructure. Heyde acknowledged the challenges. The scale of power and construction required is unprecedented. “It’s hard. There’s no doubt about it,” he admitted. “But the numbers are possible.” Some applicants, including former bitcoin miners, offered existing power infrastructure like substations and modular buildings. But Heyde said many don’t meet OpenAI’s needs. Instead, the company values being the first major tech presence in a community, creating a narrative of economic uplift and innovation. The 20 finalists represent Phase One of a much larger plan. OpenAI aims to scale from single-gigawatt facilities to sprawling campuses capable of supporting artificial general intelligence—systems that may one day match or exceed human cognitive abilities. “Any site we move forward with, we’ve thoroughly vetted for viability,” Heyde said. “We believe we can deliver the power and infrastructure story.”