Microsoft's ambition to gain independence in AI hardware is facing a setback with the delay of its next-generation AI chip, codenamed Braga, to 2026. This delay impacts Microsoft's plans to reduce its reliance on Nvidia's GPUs and gain more control over its cloud-based AI infrastructure.
The mass production of the Braga chip has been pushed back by at least six months, moving the rollout to 2026 instead of the anticipated 2025. Microsoft had intended to integrate the Maia chip, which includes Braga, into its data centers as early as this year. Several factors contributed to this delay, including design changes requested by OpenAI, staffing constraints, and high employee turnover. Last-minute design changes requested by OpenAI caused instability during simulations, setting the project back. The intense pressure resulting from these unadjusted deadlines reportedly led to approximately one-fifth of the team members leaving some development teams.
Sources suggest that the Braga chip's performance may fall short of Nvidia's Blackwell chip, released late last year. This could pose strategic challenges for Microsoft as it seeks to optimize its AI infrastructure. The delay highlights the complexities of the Microsoft-OpenAI partnership. OpenAI is also developing its own custom AI chip design and plans to send it to TSMC for production using 3-nanometer technology. This parallel development suggests OpenAI is hedging its bets while potentially influencing Microsoft’s chip roadmap.
The delay has broader implications for Microsoft's AI ecosystem. Custom AI processors like Braga are crucial to the future of AI infrastructure. Companies like Microsoft, Amazon, and Google are developing in-house silicon to optimize their platforms for cost-efficiency and performance. Google has already launched its TPU v7, and Amazon is preparing to ship its Trainium3 chip later this year. Microsoft's Maia line, including the delayed Braga and the upcoming Clea variant, represents its answer to this trend. This delay means Microsoft is trailing its top cloud competitors in hardware readiness, a critical factor in controlling long-term AI compute costs.
Microsoft is reportedly working on two new versions of the chip: Braga-R and Clea, expected in 2026 and 2027, respectively. The company still aims to establish a strong position in AI chip innovation, albeit on a slightly longer schedule. The delay is a reminder that AI hardware development is as complex as the algorithms it powers. Microsoft's ambition to control its AI destiny requires both innovation and perfect timing.
The financial stakes are high, as Microsoft's stock dipped after the delay was reported. This reflects investor concerns about its ability to reduce reliance on Nvidia's costly GPUs. Nvidia's CEO has questioned the viability of custom AI chips unless they outperform existing solutions, a bar Braga may fail to meet by 2026.
