Microsoft's custom AI chip faces delays, giving Nvidia more leverage

Microsoft's Braga AI chip faces production delays until 2026, boosting Nvidia's market leverage further.

: Microsoft's custom Maia AI chip, named Braga, has been delayed until 2026, extending six months past its initial schedule. This poses challenges in competing with Nvidia, as Microsoft's reliance on Nvidia's GPUs deepens, and highlights barriers in creating competitive AI silicon. Development faced numerous setbacks including design changes and staffing issues, partly due to partner OpenAI's late requests. Nvidia's Blackwell chips, boasting superior technology, are already leading the market, escalating competitive pressure on Microsoft.

Microsoft's ambitious project to develop a custom AI chip named Braga, as part of its Maia series, has encountered significant delays, now scheduled for mass production no earlier than 2026. Initially intended to reduce dependence on Nvidia's powerful GPUs widely used in AI data centers, this delay poses strategic challenges. Complications stem from unexpected design changes and significant staffing issues, including a high turnover rate reaching up to 20% in some teams. Furthermore, a late request from OpenAI, a key partner of Microsoft, introduced additional features, complicating the final design and introducing instability during simulations.

Nvidia, in contrast, has been advancing its Blackwell chips—highly optimized for both AI training and inference—which are built using over 200 billion transistors on a custom TSMC process offering remarkable speed and energy efficiency. Launched in late 2024, these chips have solidified Nvidia's dominance in the AI infrastructure market worldwide. Microsoft's Braga, lagging in development and solely optimized for inference, faces a substantial challenge in bridging the performance gap.

The delay in Braga's development exacerbates Microsoft's reliance on Nvidia GPUs. This reliance potentially inflates costs and limits Microsoft's capacity to differentiate its Azure cloud services from competitors. Compounding this is Amazon and Google's progress, with Amazon's Trainium 3 and Google's Tensor Processing Units further intensifying competition in data center AI.

In response to emerging players, Nvidia CEO Jensen Huang has expressed confidence, questioning the rationale of creating custom chips when Nvidia products already excel in performance and efficiency. This stance highlights Nvidia's strong market positioning and underscores the high stakes in the AI chip world.

This scenario illustrates the broader industry dynamics where stakes are high due to rapid advancements in AI technology and significant investments by major cloud service providers. The ongoing delay contributes to broader questions about Microsoft's ability to challenge established players like Nvidia in the high-performance AI chip space when faced with such organizational and technical hurdles.

Sources: TechSpot, The Information, Reuters

AIHardwaredata center