Microsoft has unveiled its Maia 200 AI chip, CNBC reported on Monday. Microsoft aims to use its second-generation AI chip to secure cloud customers and respond to demand from AI model developers.
Maia 200 provides 30 percent better performance than the existing model and is connected on an Ethernet basis instead of Nvidia InfiniBand.
Microsoft plans to deploy Maia 200 in data centres to reduce power consumption and maximise AI computing efficiency. Scott Guthrie (스콧 거스리), corporate vice president of Microsoft's Cloud and AI unit, said, "Maia 200 is the most efficient inference system Microsoft has developed," adding, "We will also provide an SDK for AI developers and researchers."
Microsoft will not sell the Maia 200 chip directly. It plans to use it for AI computing inside data centres. It appears to be pursuing a strategy to strengthen its position in the cloud market while avoiding direct competition with Nvidia by connecting 6,144 chips to maximise performance.
Chi-gyu Hwang
delight@d-today.co.kr