Microsoft's upcoming AI chip "Braga" is now expected to be delayed by at least six months, with mass production pushed back to 2026, according to The Information.
The report points to design changes, staffing shortages, and high turnover as the main reasons for the setback. Braga is also projected to lag significantly behind Nvidia's Blackwell chip, which launched at the end of 2024. Microsoft had originally planned to deploy Braga in its data centers starting in 2025.
Microsoft's first AI chip, the Maia 100, is reportedly still only being tested internally and isn't used for services like ChatGPT or Copilot. Developed in 2019, Maia 100 was originally designed for image processing rather than generative AI. After Maia 100, Microsoft had planned three successors: Braga, Braga-R, and Clea, with launches targeted for 2025, 2026, and 2027. With the Braga delay, that timeline is now in question.
Design changes and staffing issues slow Microsoft's chip ambitions
The new chips are intended for inference workloads - running trained AI models in production. A dedicated training chip, once planned, was scrapped in early 2024. During Braga's development, Microsoft added extra features at OpenAI's request, which caused instability in simulations. Despite these issues, management insisted on sticking to the original design freeze, resulting in mounting pressure and a wave of resignations.
According to the report, Microsoft isn't expected to compete with Nvidia in AI hardware - especially on performance per watt - until at least Clea, also known as Maia 300.
