Microsoft’s ambitious push to develop its own artificial intelligence (AI) hardware has hit a significant roadblock. The company’s next-generation AI chip, codenamed Braga and set to be released as the Maia 200, will not enter mass production until 2026, marking at least a six-month delay from the original 2025 target. This setback is attributed to a combination of unexpected design modifications, staffing constraints, and high employee turnover, which have collectively derailed Microsoft’s timeline for deploying the chip in its data centers.
Microsoft unveiled its Maia chip series in November 2023 as part of a broader strategy to reduce its reliance on Nvidia’s dominant AI hardware, particularly as the cost and demand for Nvidia’s GPUs continue to soar. The Maia chips, including the upcoming Braga (Maia 200), were designed to power Microsoft’s cloud infrastructure and AI services, positioning the company to compete with other tech giants like Google and Amazon, both of which have made significant strides with their own custom AI silicon.
Microsoft made unanticipated design revisions, some at the request of OpenAI, which introduced instability during simulations and set the project back by several months. The refusal of Microsoft executives to adjust project deadlines despite mounting challenges led to high stress among teams, resulting in the departure of up to one-fifth of staff on some chip design teams. The cumulative effect of these pressures has been a notable loss of experienced personnel, further slowing progress.
The delay is particularly consequential given the rapid progress of competitors. Nvidia’s Blackwell chip, launched in late 2024, has already set new performance benchmarks in AI computing, and industry insiders suggest that even when Microsoft’s Braga chip launches, it will lag behind Nvidia’s Blackwell in performance. This performance gap raises questions about Microsoft’s ability to compete effectively in the AI hardware market, especially as Google and Amazon continue to ramp up production of their own advanced AI chips.
Microsoft’s setback comes at a time when major technology firms are racing to develop proprietary processors for AI and general-purpose computing, aiming to optimize costs and performance while reducing dependency on third-party vendors like Nvidia. Google’s Tensor Processing Units and Amazon’s Trainium chips have already gained traction, with both companies unveiling new generations of their AI hardware in 2024 and 2025.
