Facebook: Meta Redesigned its Data Center After Abandoning AI Chip Rollout And Switching to Nvidia GPUs

Reuters reports that Meta abandoned plans to release its own artificial intelligence processor and instead adopted Nvidia GPUs. Initial use of CPUs for AI training and chips for inference caused the company to fall behind in AI development. The switch to GPUs necessitated the redesign of its data centers and the cancellation of multiple initiatives, as reported by DCD a year ago. GPUs require more power and generate more heat as a result. Additionally, they must be located closer together for AI system networking. According to Reuters, the new facilities required between twenty-four and thirty-two times the networking capacity and new liquid cooling systems.

Simultaneously, Meta has begun developing a new AI processor capable of both training and inference. It is scheduled for 2025, per Reuters. The Information also reports that Microsoft's corporate vice president of silicon engineering, Jean Boufarhat, will join Meta. He will lead the Facebook Agile Silicon Team, or FAST, in an endeavor to revitalize the company's faltering chip efforts. FAST is also developing chips for its augmented reality glasses, whereas Meta used Qualcomm for its virtual reality headsets and Ray-Ban smart glasses following the failure of prior internal chip development efforts.