OpenAI and Microsoft are reportedly planning a $100B supercomputer

"Stargate" could be used to train the world’s most powerful AIs.
Subscribe to Freethink on Substack for free
Get our favorite new stories right to your inbox every week

Microsoft and OpenAI are reportedly planning to build a $100 billion data center and supercomputer that could lead to the creation of AIs far more capable than anything possible today.

Power hungry: Soon after investing its first $1 billion into OpenAI in 2019, Microsoft set out to solve one of the biggest challenges facing the AI firm: the need for a lot of processing power to train and run its generative AIs. 

That led to the creation of a custom supercomputer at Microsoft data centers in Iowa.

“These supercomputing systems are really the lifeblood of our research,” Katie Mayer, who manages the Microsoft partnership for OpenAI, said in September 2023. “To do the work that we’re doing at this scale and to develop really novel AI capabilities, you need these systems. They’ve really accelerated the rate of progress that we are all benefiting from now.”

Interior of a modern data center with rows of servers and yellow columns under bright lighting.
John Brecher / Microsoft
OpenAI’s custom supercomputing system in Iowa

The report: On March 29, the Information reported that OpenAI and Microsoft are planning to spend up to $100 billion on a new data center project containing a supercomputer with “millions of specialized server chips.”

The Information cited three anonymous sources it says have been involved in conversations about the project, including one who has spoken directly to OpenAI CEO Sam Altman about it and one who has seen Microsoft’s eye-popping cost estimates.

“These supercomputing systems are really the lifeblood of our research.”

Katie Mayer

According to the Information’s sources, the supercomputer is being called “Stargate,” and it could launch as soon as 2028. It might then be expanded over the course of two years, with the final version requiring as much as 5 gigawatts of power. (To put that in context, the largest nuclear power station in the US — Palo Verde in Arizona — has a total capacity of just 3.9 gigawatts.)

Microsoft is also considering building a different, smaller supercomputer for OpenAI in Wisconsin. It could launch as soon as 2026 and cost up to $10 billion.

OpenAI declined to comment on the report, while a Microsoft spokesperson issued a general statement that the company is “always planning for the next generation of infrastructure innovations needed to continue pushing the frontier of AI capability.”

Looking ahead: Seemingly little about the project has been decided, according to the Information — Microsoft and OpenAI are still contemplating where Stargate should be located, which computer chips and cables to use, and how to power the whole operation, with nuclear energy being one possibility.

Regardless of how the details shake out, if the project does come to fruition, the processing power it would provide OpenAI could lead to the development of AIs with capabilities we can only dream about today — and maybe even the first artificial general intelligence.

We’d love to hear from you! If you have a comment about this article or if you have a tip for a future Freethink story, please email us at [email protected].

Subscribe to Freethink on Substack for free
Get our favorite new stories right to your inbox every week
Related
A personal assistant for everyone: The promise of ambient AI
We’re leaving the app era and entering the age of ambient AI: intelligent help that’s always on, but never in the way.
Gen Z: We must resist the temptation to cheat on everything
Adopting the “cheat on everything” mentality — treating thinking as a burden AI can eliminate — is not only wrong, it’s dangerous.
There are no new ideas in AI — only new datasets
Our next AI breakthrough will come when we unlock a source of data we’ve either overlooked or never fully harnessed.
Albert Einstein said automation caused the Great Depression. It didn’t.
Einstein blamed automation for the widespread unemployment of the Great Depression, but his reasoning was based on a false premise.
Has AI made “learn to code” obsolete?
Freethink talks to the creator of the world’s most popular AI coding assistant to find out whether learning to code is still worthwhile.
Up Next
Black electric sedan in motion on a road with motion blur effect.
Subscribe to Freethink for more great stories