According to insiders, Microsoft and OpenAI are planning to build a $100 billion supercomputer called "Stargate" to massively accelerate the development of OpenAI's AI models, The Information reports.
Microsoft and OpenAI executives are forging plans for a data center with a supercomputer made up of millions of specialized server processors to accelerate OpenAI's AI development, according to three people who took part in confidential talks.
The project, code-named "Stargate," could cost as much as $100 billion, according to one person who has spoken with OpenAI CEO Sam Altman about it and another who has seen some of Microsoft's initial cost estimates.
Officials discussed launching Stargate in 2028, with an expansion by 2030 that could require up to five gigawatts of power toward the end. It would be built in the U.S. and would be the largest of a series of data centers to be built over the next six years. The system would require several gigawatts of power - the equivalent of at least several large data centers today, according to The Information.
A large part of the project's cost would be to procure the chips, but providing enough power sources to operate would also be a challenge. Funding would likely come primarily from Microsoft.
If the project comes to fruition, it would be one of the largest investments in the history of computer science. It shows how far ahead OpenAI and Microsoft are thinking.
Lately, Altman has repeatedly pointed to a shortage of chips for AI development. He is said to be in talks with numerous investors and project partners to address this by building a global infrastructure of chip factories.
Smaller supercomputer to be launched as early as 2026
OpenAI and Microsoft are said to be working on a smaller supercomputer, known as "Phase 4," which is expected to be operational in 2026. When completed, it could cost $10 billion - still many times the cost of today's data centers.
According to one source, Microsoft's decision to move forward with the Stargate project, Phase 5, will also depend on whether OpenAI can significantly improve the capabilities of its AI and truly achieve some sort of "superintelligence".
With more servers available, some OpenAI executives believe the company could use its existing AI to create the right synthetic training material for better models - a self-improving AI, according to The Information.
Echoing The Information's report are rumors from OpenAI insider "Jimmy Apples," who recently wrote on X that OpenAI CEO Sam Altman is "building a big special something" "somewhere in the desert." Energy companies Helion Energy and American Clean Power are allegedly involved. (Update: Jimmy Apples confirmed he was referring to Stargate.)
