Microsoft and OpenAI are collaborating on a significant endeavor to construct a cutting-edge AI-focused supercomputer within a massive datacenter, as per a report by _The Information_. This ambitious project is estimated to cost over $115 billion and aims to house millions of GPUs. The supercomputer, known as “Stargate” internally at OpenAI, is slated to be based in the U.S.
Microsoft is expected to finance the datacenter build, which could surpass the costs of current large operating centers by a factor of 100. Stargate represents just one of several planned datacenter projects the two companies intend to undertake over the next six years, with a target operational date set for 2028.
According to the report, the development of these supercomputers is progressing in stages, with Stargate designated as a phase 5 system. A phase 4 system, with a lower cost, may launch as early as 2026, potentially starting in Mt. Pleasant, Wisconsin. Given the substantial power demands of a system like Stargate, reaching several gigawatts, Microsoft and OpenAI are exploring alternative power sources like nuclear energy.
Challenges lie ahead due to the need for innovative designs tailored for increased GPU density in racks, enhancing chip performance and efficiency, and implementing efficient cooling solutions. Notably, there are indications that the companies may be looking to reduce dependence on Nvidia’s technologies, favoring Ethernet cables over Nvidia’s InfiniBand cables for Stargate.
While details like the final price, location, and structure of Stargate remain uncertain, the evolving nature of the project suggests potential fluctuations in plans. Nonetheless, both Microsoft and OpenAI are heavily investing in advancing AI capabilities, with Microsoft notably unveiling specialized processing units, and OpenAI exploring chip production ventures.
HEALTH
WORLD AFFAIRS
TECHNOLOGY
CREATIVE
