Get rid of external chip dependence! Microsoft MAIA 300 will be unveiled as soon as September and is expected to be delivered in 2027

Kuai Technology reported on August 11 that according to The Information, Microsoft is accelerating its self-developed AI chip strategy and plans to significantly increase the production of next-generation chips next year, targeting leading cloud customers such as Anthropic and promoting them from internal verification to external commercial use. Although the current adoption rate of the first-generation Maia 200 chip is less than expected, Microsoft has not slowed down. According to people familiar with the matter, the new Maia 300 chip may be officially released as early as September. At the same time, Microsoft has negotiated with TSMC to lock in the production capacity of more than 300,000 chips, which is expected to be delivered in 2027. It is worth noting that Maia 300 will be open to major cloud customers other than Azure for the first time, marking Microsoft's self-developed ASIC officially entering a new stage of large-scale commercial use. The chip is natively equipped with an Ethernet Scale-up architecture and has ultra-high inter-chip interconnection bandwidth. With the large-scale networking of Maia 300, the demand for bidirectional high-speed interconnection (Scale-up) and inter-cluster interconnection (Scale-out) will surge. This trend will directly drive the penetration rate of 800G/1.6T optical modules and high-density optical connection devices to increase - optical interconnection is no longer exclusive to GPUs, but will become the infrastructure required for all types of AI computing power clusters, further strengthening the certainty of industry demand.