Microsoft and OpenAI are contemplating an AI supercomputer data center that may cost as much as $100 billion. Called Stargate, the aim would be to have it operational by 2008 to drive OpenAI’s next generation of artificial intelligence. According to reports, the Stargate complex would span hundreds of U.S. acres and use up to 5 gigawatts of power, which is massive (the equivalent of a substantial metropolitan power grid). In light of those power needs, a nuclear power source is said to be under consideration. The project is not yet green-lit, and no U.S. location has been selected.

“Of the series of installations the companies plan to build over the next six years, Stargate has the largest scope, and it’s considered crucial for OpenAI to train and operate more advanced AI models than ChatGPT-4,” writes Tom’s Guide.

“OpenAI CEO Sam Altman and Microsoft employees have reportedly broken down the plan to build several supercomputers throughout the United States into five stages, with Stargate being the fifth and final stage.”

The Stargate data center is estimated to take five to six years to complete. Microsoft is also expected to launch what Tom’s calls “a smaller scope, phase 4 supercomputer for OpenAI around 2026.”

The Information, which broke the story, says Stargate will contain “millions of specialized server chips” and be “100 times more costly than some of today’s data centers,” adding that Microsoft “would likely be responsible for financing the project.”

Tom’s says the project is largely dependent “on whether OpenAI can deliver its next-gen large language model, rumored to be GPT-5,” noting that in 2023 the company missed its initial deadline to present Microsoft with an AI project called Arrakis “citing the limitations of current supercomputers as the bottleneck holding up development.”

OpenAI’s naming protocol for these projects is inspired by pop culture, taking their respective names from Roland Emmerich’s 1994 sci-fi film and the central planet in Frank Herbert’s 1965 novel “Dune,” source material for several movies.

The Next Platform underscores the “data center” aspect of the project, “because the data center — and perhaps multiple data centers within a region with perhaps as many as 1 million XPU computational devices — will be the supercomputer.”

Next writes that “the scalability goal of 1 million interconnected endpoints” is consistent with that of the Ultra Ethernet Consortium (Microsoft as a founding member), set for next-gen Ethernet networks.