Meta plans to incorporate custom AI chips into servers

407

Meta Platforms, the parent company overseeing Facebook, has announced its intention to incorporate its proprietary artificial intelligence (AI) chips, codenamed Artemis, into its data centres later this year.

The strategic move signals Meta’s bid to diminish reliance on Nvidia’s dominant H100 chips and rein in the soaring expenses associated with AI workloads.

The tech giant has allocated significant resources, amounting to billions of dollars, to bolster its computing capabilities in support of resource-intensive generative AI endeavours integrated across its platforms, including Facebook, Instagram, and WhatsApp. Such efforts entail substantial investments in specialized chips and the reconfiguration of data centre infrastructure to accommodate them.

Dylan Patel, founder of the silicon research group SemiAnalysis, anticipates substantial cost savings for Meta, with the successful deployment of its in-house chip potentially slashing annual energy expenditures by hundreds of millions and chip procurement costs by billions.

Despite Meta’s strides towards self-sufficiency, it will retain Nvidia’s H100 GPUs within its data centers for the foreseeable future. CEO Mark Zuckerberg outlined plans to have approximately 350,000 H100 processors operational by year-end.

Also Read: Meta set to label AI-generated images from other companies

This development represents a positive trajectory for Meta’s internal AI silicon initiative, following the 2022 decision to discontinue the initial iteration of the chip in favor of Nvidia’s GPUs. Like its predecessor, Artemis is tailored for AI inference, facilitating algorithmic decision-making and response generation to user interactions.

A Meta spokesperson emphasized the complementary nature of internally developed accelerators with commercially available GPUs, striving to achieve optimal performance and efficiency for Meta-specific workloads.

While Meta’s shift to diminish dependence on Nvidia’s processors could signify a potential chink in Nvidia’s AI dominance, it’s evident that Nvidia’s GPUs will continue to wield significant influence over Meta’s AI infrastructure in the immediate future.