Meta unveils 4 new in-house AI chips to power data centers, boost efficiency

• Meta Platforms announced a roadmap for four new in-house AI chips under its Meta Training and Inference Accelerator (MTIA) program.
• The first chip, MTIA 300, is already deployed for ranking and recommendation systems, with others rolling out by 2027.
• The in-house designs aim for greater energy efficiency and cost-effectiveness compared to off-the-shelf products.
• Meta is rapidly expanding its data centers, planning new chip releases every six months to meet exploding inference demand.
• While successful with inference chips, Meta continues to work on generative AI training chips.