Meta Platforms, the entity behind Facebook, has unveiled its ambitious strategy to integrate its proprietary artificial intelligence (AI) chips, known as Artemis, into its data centres in the upcoming months.
This strategic manoeuvre marks Meta’s endeavour to reduce dependence on Nvidia’s prevalent H100 chips and mitigate the escalating costs associated with AI workloads.
The tech behemoth has committed substantial resources, amounting to billions of pounds, to enhance its computational capabilities in support of demanding generative AI initiatives integrated across its array of platforms, encompassing Facebook, Instagram, and WhatsApp. Such initiatives necessitate significant investments in specialised chips and the overhaul of data centre infrastructure to accommodate them.
Dylan Patel, the founder of the silicon research group SemiAnalysis, foresees substantial cost efficiencies for Meta, with the successful implementation of its in-house chip potentially leading to a reduction in annual energy expenditures by hundreds of millions and chip procurement costs by billions.
Despite Meta’s strides towards self-sufficiency, it will continue to deploy Nvidia’s H100 GPUs within its data centres for the foreseeable future. CEO Mark Zuckerberg outlined plans to have approximately 350,000 H100 processors operational by year-end.