According to foreign media reports, Meta Platforms plans to begin deploying its new self-developed ARKE chips in data centers in the first half of next year. The company said this move will save costs and energy when running artificial intelligence models. The company first announced plans to independently develop AI chips as early as 2023, and is currently testing the third-generation product in this series, codenamed MTIA 450, or Arke. The next generation product — codenamed 500, or Astrid — will be designed in about a month and put into use in data centers by the end of 2027. Meta expects more widespread adoption of this product in the future. Jiun Song, vice president of engineering at Meta, said in an interview: “Each generation of chips carries higher technical risks, but performance has also improved significantly.” According to the report, the Meta Super Smart Lab helps fine-tune the chip by providing insight into future AI models and their operating requirements. Song said the end result is that when running AI models, these chips will perform more efficiently than “any product currently launched by Nvidia,” “simply because we have done a lot of engineering work ourselves.”

According to foreign media reports, Meta Platforms plans to begin deploying its new self-developed ARKE chips in data centers in the first half of next year. The company said this move will save costs and energy when running artificial intelligence models. The company first announced plans to independently develop AI chips as early as 2023, and is currently testing the third-generation product in this series, codenamed MTIA 450, or Arke. The next generation product — codenamed 500, or Astrid — will be designed in about a month and put into use in data centers by the end of 2027. Meta expects more widespread adoption of this product in the future. Jiun Song, vice president of engineering at Meta, said in an interview: “Each generation of chips carries higher technical risks, but performance has also improved significantly.” According to the report, the Meta Super Smart Lab helps fine-tune the chip by providing insight into future AI models and their operating requirements. Song said the end result is that when running AI models, these chips will perform more efficiently than “any product currently launched by Nvidia,” “simply because we have done a lot of engineering work ourselves.”