Microsoft and OpenAI are reportedly embarking on a transformative journey in the realm of artificial intelligence (AI), marked by a monumental data center project set to redefine the industry landscape. At its core lies the ambitious development of an AI supercomputer dubbed “Stargate”, representing a strategic move aimed at reducing dependency on Nvidia, a dominant player in AI hardware.
The genesis of this groundbreaking initiative stems from a report by Anissa Gardizy and Amir Efrati in The Information, revealing Microsoft’s commitment to invest over $100 billion in the project, with a targeted launch scheduled for 2028. This strategic collaboration seeks to propel AI capabilities to new heights while mitigating reliance on external hardware providers.
The project’s magnitude is underscored by its comprehensive approach, delineated into five distinct phases, with Stargate positioned as the pinnacle of technological innovation. While details regarding the project’s cost and specifications originate from insider sources, official confirmation from Microsoft and OpenAI is still pending.
Projected expenses for the endeavor soar beyond $115 billion, eclipsing Microsoft’s prior investments in server infrastructure by a significant margin. In tandem with the Stargate initiative, Microsoft is concurrently developing a smaller-scale supercomputer tailored for OpenAI, slated for deployment around 2026.
The envisaged Stargate system represents a paradigm shift in AI infrastructure, veering away from standalone supercomputers towards an ecosystem centered on data center capabilities. As elucidated by The Next Platform, discussions surrounding Stargate hint at a vast network of data centers outfitted with state-of-the-art computational devices, potentially accommodating up to one million accelerator processing units (XPUs) per machine.
In terms of hardware architecture, Stargate is poised to leverage forthcoming iterations of Cobalt Arm server processors and Maia XPUs, complemented by Ethernet scaling to facilitate seamless integration of hundreds of thousands to a million XPUs within a single machine. Notably, the project deviates from conventional AI hardware configurations, eschewing Nvidia GPUs and interconnects in favor of novel approaches.
While details remain speculative, the prospective impact of Stargate on AI research and development is profound. If realized, this ambitious undertaking promises to redefine the contours of AI infrastructure, heralding an era of unparalleled advancements in machine learning and computational prowess.