To enhance its AI capabilities, Meta Platforms, Facebook's parent company, plans to deploy custom chips, including the 'Artemis' chip, in its data centres.

The company's decision to develop and deploy its advanced chips is a significant move after having previously opted to buy billions of dollars worth of Nvidia's graphics processing units (GPUs). The new in-house chip, known internally as 'Artemis,' is intended to supplement Meta's existing infrastructure with hundreds of thousands of off-the-shelf GPUs.

Why does it matter?

This move aims to mitigate Meta's dependence on Nvidia's chips and control costs. Deploying its own chips should result in significant savings, like substantially reducing annual energy expenses and cutting down on other costs. Meta's transition to using custom chips could potentially shave off hundreds of millions of dollars at the scale at which the company operates.

The new chip is part of Meta's broader efforts to boost its computing capacity for AI applications, particularly in the areas of generative AI and inference workloads. Meta spokespersons confirmed the company's plan to put the new chip into production in 2024, stating that it would work in coordination with the existing commercial GPUs in operation.