Microsoft to Unveil Next-Generation AI Chips Maia 300 in September
Microsoft
Microsoft plans to present the next generation of its AI accelerator chips, Maia 300, in September to reduce reliance on costly Nvidia chips. The company is in talks with TSMC to produce over 300,000 units with delivery in 2027, potentially aiming to reserve capacity for over a million chips.
Microsoft is set to introduce the next generation of its AI accelerator chips, Maia 300, in September as part of its strategy to develop proprietary AI infrastructure and decrease dependence on expensive Nvidia chips. According to Reuters, Microsoft is discussing with TSMC the production capacity for over 300,000 Maia 300 units with delivery scheduled for 2027. The company plans to scale up production and attract major cloud clients, including Anthropic. The ultimate goal could involve reserving capacity for more than one million such chips, though this depends on component availability and ongoing negotiations with TSMC. Andrew Wall, General Manager of Azure Maia, confirmed long-term investments but noted that published figures do not reflect the full scope of the program. Microsoft's first Maia accelerator was introduced in November 2023, but the company has lagged behind Alphabet and Amazon in this area. Google has started recognizing revenue from direct sales of its Tensor Processing Units, while Amazon's solutions, including Trainium, have gained wider adoption. In January, Microsoft showcased the second-generation Maia 200, manufactured by TSMC using a 3nm process, featuring increased SRAM memory to accelerate handling of large numbers of user queries in AI systems.
- Abbreviations
- SRAM = Static Random-Access Memory — статическая память с произвольным доступом
Source: 3DNews —
original
