Social media giant Meta Platforms (formerly Facebook) is set to deploy a new version of its custom chip in its data centers this year, according to an internal company document seen by Reuters. The chip is part of Meta’s ongoing efforts to support its artificial intelligence (AI) push, reduce dependence on Nvidia chips, and control the rising costs associated with AI workloads.
As Meta continues to develop power-hungry generative AI products for its platforms such as Facebook, Instagram, and WhatsApp, as well as hardware devices like its Ray-Ban smart glasses, the need for increased computing capacity has become crucial. To meet this demand, Meta has been investing billions of dollars in specialized chips and reconfiguring data centers.
The deployment of its own custom chip could potentially result in significant cost savings for Meta, as it aims to reduce both annual energy costs and billions spent on chip purchasing. This move comes at a time when the chips, infrastructure, and energy required to run AI applications have become a major investment sinkhole for tech companies.
A Meta spokesperson confirmed that the updated chip, which is the second generation of its in-house silicon line, will be put into production in 2024. The chip will work alongside the commercially available graphics processing units (GPUs) that Meta currently purchases. The spokesperson stated that Meta’s internally developed accelerators are highly complementary to GPUs and provide an optimal mix of performance and efficiency for Meta-specific workloads.
Meta CEO Mark Zuckerberg recently announced plans to acquire approximately 350,000 flagship GPUs from Nvidia by the end of the year. In combination with other suppliers, Meta aims to accumulate the equivalent compute capacity of 600,000 GPUs in total. This deployment of Meta’s own chip is a positive development for the company’s in-house AI silicon project, following its decision in 2022 to discontinue the chip’s first iteration.
The new chip, internally referred to as Artemis, shares similarities with its predecessor by focusing on the inference process. Inference involves using algorithms to make ranking judgments and generate responses to user prompts. Additionally, Meta is said to be working on a more ambitious chip capable of both training and inference, similar to GPUs.
Industry experts believe that an inference chip developed by Meta could be more efficient in processing recommendation models compared to energy-intensive Nvidia processors. This could potentially result in substantial cost savings for the company.
The development and deployment of Meta’s own custom chip mark significant milestones in its AI ambitions. By reducing dependency on external suppliers and maximizing the performance and efficiency of AI workloads, Meta aims to continue its innovation in AI products and make strides in advancing the capabilities of its platforms and hardware devices.
In conclusion, Meta Platforms’ plans to deploy an updated custom chip in its data centers this year demonstrate its commitment to driving AI advancements and reducing costs associated with AI workloads. By leveraging its own chip technology, Meta aims to optimize performance and efficiency, potentially saving billions of dollars while pushing the boundaries of AI on its platforms and devices.
