Microsoft´s in-house artificial intelligence chip delayed until 2026, falling behind Nvidia

Microsoft´s custom artificial intelligence chip has been delayed by six months and is now expected to launch in 2026, amid design setbacks and lagging well behind Nvidia´s Blackwell series.

Microsoft´s ambitious project to develop its own artificial intelligence chip, internally codenamed Braga, has encountered a significant setback, with new reports indicating a six-month delay that pushes mass production to 2026. Despite years of expensive research and development, insider sources claim the chip will not compete with Nvidia´s industry-leading Blackwell chips, which have already established a performance benchmark in the sector.

The delay stems from unexpected design changes, persistent staffing issues, and high turnover within Microsoft´s silicon engineering teams. A crucial factor was the late-stage addition of features requested by OpenAI, which resulted in instability during chip simulations and ultimately set the project back several months. Despite recognizing these setbacks, Microsoft reportedly did not revise its internal deadlines, putting heightened pressure on teams and causing up to a fifth of team members to leave certain units.

Microsoft has been working on multiple in-house silicon products since 2019, including the Maia 100—a 128-core Arm CPU that, while introduced in 2023, remains limited to internal testing rather than deployment in operational artificial intelligence services. Braga, along with future variants Braga-R and Clea, targets data center inference tasks, with staggered projected rollouts between 2025 and 2027. However, the Braga delay casts doubt across the entire timeline. There has also been a cancellation of a dedicated training chip, signaling further challenges facing Microsoft´s artificial intelligence hardware roadmap. As rivals Google and Amazon accelerate their own chip efforts, Nvidia CEO Jensen Huang has publicly questioned the value of custom silicon if it cannot surpass what is commercially available—a sentiment that recent Microsoft developments may prove accurate.

67

Impact Score

YouTube to automatically label Artificial Intelligence-generated videos

YouTube is shifting from voluntary disclosure to automated detection for significant photorealistic Artificial Intelligence-generated video content. Labels will become more visible across long-form videos and Shorts, with permanent markers for content made with YouTube tools or verified through provenance systems.

Axiom Math says its proofs reached peer reviewed journals

Axiom Math says proofs generated by its system have been accepted by several peer-reviewed journals, pairing machine-checkable formal proofs with human-authored papers. The development adds evidence that Artificial Intelligence tools are beginning to contribute to publishable mathematical research.

Google expands Gemini for Science

Google is rolling out Gemini for Science, a set of experimental tools aimed at compressing scientific work that would typically take months or years into days. The effort combines multi-agent research systems, computational discovery tools, literature analysis, and database-connected life science assistants.

Europe weighs technology sovereignty push amid internal debate

Europe is preparing a new policy push to reduce reliance on major technology platforms, but internal disagreements are shaping the scope and pace of the effort. The Artificial Intelligence Development Act is due to be unveiled on June 3 after repeated delays.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.