Meta Platforms, the parent company of Facebook, is set to implement its proprietary artificial intelligence chips, known as Artemis, in its data centers this year, as revealed in an internal document obtained by Reuters. This strategic move aims to potentially diminish Meta’s reliance on Nvidia’s predominant H100 chips, thereby managing the expenses associated with running AI workloads.
Dylan Patel, founder of the silicon research group SemiAnalysis, suggests that the successful integration of Meta’s custom chip, Artemis, could result in substantial savings, potentially amounting to hundreds of millions in annual energy costs and billions in chip procurement expenses.
Despite this shift towards self-sufficiency, Meta will still use Nvidia’s H100 GPUs in its data centers for the foreseeable future. CEO Mark Zuckerberg has outlined plans to have approximately 350,000 H100 processors in operation by the end of the year.
The deployment of Meta’s own chip marks a positive development for its in-house AI silicon project, following a decision in 2022 to discontinue the initial iteration of the chip in favor of Nvidia’s GPUs. Similar to its predecessor, the new chip, Artemis, is tailored for AI inference tasks, involving algorithmic decision-making and response generation to user prompts.
A Meta spokesperson emphasized the complementary nature of their internally developed accelerators and commercially available GPUs, delivering an optimal balance of performance and efficiency for Meta-specific workloads. While Meta’s efforts to decrease dependence on Nvidia’s processors could be perceived as a potential crack in Nvidia’s AI dominance, it is evident that Nvidia’s GPUs will continue to play a significant role in Meta’s AI infrastructure for the time being.