Home > News > Microsoft's self-developed AI chip Braga has been delayed for six months, and its performance is inferior to Nvidia's
RFQs/Order (0)
English

Microsoft's self-developed AI chip Braga has been delayed for six months, and its performance is inferior to Nvidia's


The latest report shows that Microsoft's self-developed AI chip project has encountered significant setbacks. The company's first internally developed AI chip product will be delayed for six months and is expected to be launched in 2026, when its performance will not be as good as Nvidia's currently available Blackwell chip.

The development cycle of this chip, codenamed Braga, far exceeded expectations, and large-scale production has been postponed until next year. According to internal sources, the performance of the chip is expected to be far inferior to Nvidia's flagship product Blackwell, which was launched last year. For Microsoft, which originally planned to deploy the chip in data centers this year, this is undoubtedly a major blow. The project personnel attributed the delay to unforeseen design changes, staffing restrictions, and high turnover rates. According to reports, due to immense pressure, one-fifth of the project team's employees have left.

Microsoft started developing chips in 2019 and launched the Maia 100 in 2023. This 128 core Arm CPU was originally planned to be deployed to data centers in early 2024, but it is mainly used for internal testing and has not yet been put into practical use. According to insiders at Microsoft, this chip does not provide computing power support for any of the company's AI services, mainly because it was already approved before OpenAI's ChatGPT triggered industry changes, and was originally designed for image processing, rather than generative AI and big language models.

Currently, Microsoft is secretly developing three chips, codenamed Braga, Braga-R, and Clea, with plans to deploy them to data centers in 2025, 2026, and 2027, respectively. The delay of the first chip has raised doubts about whether Microsoft can achieve this ambitious release goal. According to reports, all three chips are designed for inference scenarios, and one chip originally intended for AI model training has been discontinued in early 2024.

It is worth noting that a design change for the Braga chip was requested by OpenAI to add new features, which resulted in stability issues during simulation testing, causing the project progress to be delayed by several months. However, Microsoft did not adjust the final deadline, which put tremendous pressure on the project team.

According to analysis, Microsoft may not have a chip that can compete with Nvidia products until at least 2027 when it releases the Clea version of the Maia chip. During this period, Microsoft will fall significantly behind NVIDIA, without considering the significant technological breakthroughs that NVIDIA may achieve during this time.

Although Nvidia remains the industry leader in the field of AI chips, tech giants such as Microsoft, Google, and Amazon are working hard to develop internal chips to reduce their dependence on Nvidia. However, Nvidia CEO Huang Renxun seems to have no concerns about these competitive efforts. He even hinted that many competitors' chip projects will eventually be abandoned by large technology companies, and questioned "what is the point of manufacturing custom chips if they are not better than what can be bought on the market?" If the latest report on Microsoft chip delays is true, Huang Renxun's viewpoint may be validated.

Select Language

Click on the space to exit