Meta's Chief Financial Officer Susan Li announced the company's ambition to expand its own silicon production, focusing on creating processors for training complex AI models. Currently, Meta already successfully uses its own developments for ranking and recommendation algorithms in social networks. This was reported by Bloomberg, writes UNN.
Details
The company emphasizes that many of Meta's workflows are unique, which makes the use of standard market solutions not always optimal. Its own silicon will allow better adaptation of equipment to the architecture of the Llama family models and other internal services.
What is generative AI and how does it work?04.03.26, 13:48 • [views_44788]
According to Susan Li, custom chip manufacturing is an important part of a strategy that allows choosing the best tool for each specific use case – from recommendation systems to large-scale AI training.
Some of our workloads are truly very custom to us. We started with ranking and recommendation workloads, and that's where we deployed custom silicon technologies at the largest scale. But we expect and hope to expand that over time, including eventually to training AI models.
Balance between purchases from market leaders and in-house developments
Although Meta remains one of the largest buyers of Nvidia products, the development of its own production should reduce energy consumption costs and improve infrastructure performance in the long run.
The company continues to analyze its current needs, selecting optimal chip types for various tasks – from running ready-made models to training new AI iterations. Custom silicon is considered a critically important element that will allow Meta to maintain a competitive advantage in the generative artificial intelligence market.
OpenAI considers NATO contract for AI deployment - Reuters04.03.26, 17:49 • [views_5420]