Meta Platforms Announces Expanded Use of Proprietary AI Chips
Meta Platforms (META) is advancing its custom chip strategy with the deployment of next-generation proprietary AI processors, including MTIA 450 (Arke), into data centers by early 2027. The move aims to reduce compute costs and improve efficiency as part of Meta's Training and Inference Accelerator program. Meta has already begun testing its third-generation Arke chip with TSMC and is working on the fourth-gen Astrid, developed in collaboration with Broadcom (AVGO). This strategy shifts focus from dual-purpose designs like Olympus to general-purpose inference accelerators. The company plans to deploy over a gigawatt of proprietary capacity to reduce reliance on external GPUs. Market strategist Dan Niles remains bullish on META shares despite broader tech sector challenges and high U.S. 10-year yields, citing regulatory settlements and new AI monetization opportunities like the Spark 1.3 model and commercial API launch.
Tickers: $AMZN $AVGO $GOOG $GOOGL $META $MSFT $TSM
Topics: AI chips custom silicon data center efficiency Meta Platforms TSMC