OPENAI Stargate Data Center Buildout Infrastructure Lead Keith Heyde

This was not how Keith Heyde was planning to celebrate the holidays. Rather than spending time with his wife at home in Oregon, Heyde spent late December

Two months earlier, Heyde left Meta To join Openai as head of the infrastructure. His work consisted in transforming ambitious dreams of calculating the CEO Sam Altman in reality, looking for large expanses of land adapted to large installations which will ultimately be filled with powerful graphic treatment units to build large language models.

“My Christmas between the two and the New Year, the last year has been mainly spent watching sites,” said Heyde, 36, in CNBC in an interview. “So my family loved it, believe me.”

His life in 2025 has no longer become intense.

Since January, Openai has discreetly requested and examined the proposals of around 800 candidates hoping to welcome the next wave of its Stargate data centers, AI super-ordering poles designed to form increasingly powerful models.

About 20 sites are now at advanced diligence stages, with massive leaflets of land during examination in the Southwest, Midwest and Southeast. Heyde said that tax incentives are “a relatively low part of the decision matrix”.

The most important factors are access to energy, scale capacity and adhesion of local communities.

“Can we build quickly, is the power ramp rapid, and is it something where it makes sense from the point of view of the community?” He said.

Heyde leads the development of the site within the OpenAi industrial calculation team, a division which has quickly become one of the most important groups of the company. The infrastructure, once a support function, has now been raised to a strategic pillar on a par with the development of products and models.

With traditional data centers almost at maximum capacity, Openai is betting that possession of the next generation of physical infrastructure is at the heart of the control of the future of AI.

Energy needs are difficult to understand. A Gigawatt data center requires the amount of energy necessary for certain entire cities. At the end of last month, Openai announced plans for a construction of 17 gigawatts in partnership with Oracle, Nvidia and Softbank.

The new sites will have to include all kinds of energy options, including battery back installations, inherited gas turbine renovations and even small modular nuclear reactors, Heyde said. Each site seems different, but together, they form the industrial skeleton that the Openai must evolve.

“We have done this wonderful strangulation analysis to see what types of energy sources really allow us to unlock the journey on which we want to make,” said Heyde.

A good part of the capital comes from Nvidia. The flea manufacturer has agreed to invest up to $ 100 billion to feed the expansion of Openai, which will buy millions of NVIDIA GPUs.

Heyde, a former IA chief calculated at Metahelped supervise the construction of the first GPU cluster of 100,000 GPUs from Meta.

In addition to power, Openai assesses the speed with which he can rely on a site, the availability of work and the proximity of local support governments, according to the request for a Stargate proposal.

Heyde said the team had made around 100 visits to the site and has a short list of late review sites. Some will be of any new construction, and others will need conversions and renovations of existing facilities. Flexibility will be the key.

“Perfect plots are largely taken,” said Heyde. “But we knew that Perfect was not the objective – the objective for us was, number one, a convincing power ramp.”

Competition is fierce.

Meta builds which can be the largest data center in the Western hemisphere – a project of $ 10 billion in northeast Louisiana, fueled by billions of state incentives. CEO Mark Zuckerberg increased the upper end of the company’s annual capital expenditure range to $ 72 billion in July.

Amazon And Anthropic joins forces on an AI campus of 1,200 acres in Indiana. And across the country, the states deploy tax alleviation, electricity guarantees and accelerated zoning approvals to attract the next Grand Cluster of AI.

Openai has been a relative upstart, having been there for a decade and has known only the general public since the launch of Chatgpt less than three years ago. But he raised money mounds Microsoft And SoftBank, in addition to Nvidia, on the way for an evaluation of $ 500 billion.

And Openai shows that he is not afraid to show the way in AI. An self-constructed solar campus in Abiline, Texas, is already live.

While Optai is still based on partners such as OracleSarah Friar, OPENAI financial director, told CNBC last week in Abilene that possession of first part infrastructure provided a differentiated approach. It limits supplier increases, safeguarding key intellectual property and follows the same strategic logic which has once led Amazon to create Amazon web services rather than relying on the existing infrastructure.

However, Heyde indicated that there was no real game book with regard to AI, especially since companies pursue artificial general intelligence (AG) or AI which can potentially respond or exceed human capacities.

“It is a very different order of magnitude when we think of the type of delivery that must occur in these places,” he said.

Some candidates, including the former bitcoin The mining operators offered an existing electrical infrastructure, such as stations and modular buildings, but Heyde said that they were not always suitable.

“Sometimes we found that it is almost pleasant to be the first interaction in a community,” he said. “It is a very beautiful story that we bring the data center and the infrastructure there in the name of OpenAi.”

The 20 finalist sites represent phase one of the much larger construction. OPENAI ultimately plans to go through projects at Gigawatt unique to massive campuses.

“Any place or any site with which we are going ahead, we have really considered the viability and our own conviction that we can deliver the history of power and the history of the infrastructure associated with these sites,” said Heyde.

He understands why many people are skeptical.

“It’s difficult. There is no doubt about it,” said Heyde. “The figures we are talking about are very difficult, but it is certainly possible.”

WATCH: The construction of $ 850 billion in Openai complies with the limits of the grid