Microsoft (MSFT, Financial) has pushed back its most ambitious in-house AI chips—codenamed Braga-R and Clea—to 2028 or later, while deferring Maia 200 to 2026 and focusing on interim designs after development delays.
Microsoft released its first in-house AI chip, Maia 100, last year and originally slated Maia 200 for 2025. Facing design and tooling hurdles, the company has now moved Maia 200 to 2026. Its follow-up “Braga” chip only wrapped up design last month—six months late—and the successor Braga-R won't reach mass production until 2028.
Clea, the third-generation AI ASIC, has slipped even further into a post-2028 timeline. These delays ripple through Microsoft's supplier Marvell Technology (MRVL, Financial), which provides key chip components and saw its shares dip on the news.
Custom AI silicon is a strategic lever for hyperscalers to optimize cost and performance, but extended timelines risk keeping Microsoft tethered to Nvidia's (NVDA, Financial) GPUs. Slower in-house rollouts could impact Microsoft's large-scale AI deployments and cede momentum to rivals like Amazon Web Services (AMZN, Financial), which itself partners with Marvell on its Trainium chips.
By reprioritizing interim designs through 2026 and beyond, Microsoft aims to maintain progress in custom silicon, but it now faces a critical balancing act between in-house innovation and third-party dependencies.
