Microsoft slows down AI chip development pace: abandons aggressive approach, focuses on pragmatic design

Faced with the challenge of delays in the development of internal AI chips, Microsoft is adjusting its ambitious strategy, shifting towards a more pragmatic and iterative design approach to ensure it remains competitive with market leader NVIDIA in the coming years.

On July 2, according to media reports, Microsoft is adjusting its roadmap for internally developed AI server chips, focusing on releasing less aggressive designs before 2028 in an effort to overcome the issues causing development delays. The Maia 200 chip, originally scheduled for release in 2025, has been postponed to 2026.

The report cites a Microsoft spokesperson stating that the company did not directly comment on the development details of the Maia chip but indicated that it is "still committed" to developing internal hardware based on customer and its own computing needs while continuing to work with "close chip partners." Microsoft executives believe that the new Maia 280 chip can still offer a performance advantage of 20% to 30% per watt compared to NVIDIA's chips scheduled for release in 2027.

The report highlights that the core of this strategic adjustment is Microsoft's acknowledgment that designing a brand new high-performance chip from scratch every year is not feasible. By reducing the complexity of some designs and extending the development cycle of other chips, Microsoft hopes to advance projects more smoothly, with the ultimate goal of reducing reliance on NVIDIA's annual chip purchases worth billions of dollars. As of Wednesday's market close, Microsoft's stock price fell by 0.2%.

Development Delays Force Strategic Adjustments

Microsoft's setbacks stem from the development challenges of its second and third-generation AI chips.

The report notes that the company launched its first AI chip, Maia 100, in 2024 and immediately began developing three follow-up products, codenamed Braga, Braga-R, and Clea, which were originally planned for release in 2025, 2026, and 2027, respectively.

However, both Braga and Clea are based on entirely new designs, making development extremely challenging.

The design of the Braga chip was not completed until June, about six months behind the year-end deadline. The delays with Braga have raised concerns internally at Microsoft that the chips due in 2026 and 2027 may also be delayed, making them less competitive against NVIDIA's chips upon release.

As a result, Microsoft executives told engineers last week that the company is considering developing an intermediate chip for release in 2027, which would have performance between Braga and Braga-R. This chip may be called Maia 280, primarily based on Braga's design but composed of at least two connected Braga chips, allowing them to work together as a single, more powerful chip.

The chip originally codenamed Braga-R will now be named Maia 400, expected to enter mass production in 2028. This chip will utilize more advanced interconnect technology to integrate at the die level for faster performance. Microsoft plans to gradually increase production with each new generation of chips, with the ultimate goal of producing hundreds of thousands of self-developed AI chips annually The release of Microsoft's third-generation AI chip, Clea, has been postponed until after 2028, and its prospects remain unclear.

Chain Reaction on Partners

Microsoft's revised roadmap has negatively impacted professional chip design company Marvell. Microsoft hired Marvell to work on certain chiplets in Braga-R.

Marvell's stock price soared last year due to partnerships with major tech companies, including Amazon, as the company originally expected to receive revenue from Microsoft sooner.

However, due to delays in customer chip projects, a global economic slowdown, and trade tensions between the U.S. and other countries, its stock price has declined this year. On Wednesday, Marvell's stock closed down 2.61%.

Not all of Microsoft's chip projects are facing issues. Compared to AI chips (primarily based on GPUs), the design of central processing unit (CPU) projects, which are less complex, is progressing smoothly.

Microsoft released the CPU chip Cobalt for servers in 2024 as an alternative to Intel and AMD products.

Media reports citing multiple insiders indicate that Cobalt has been used internally to support services like Teams and has been made available to Azure cloud customers, generating revenue for the company. The design of its next-generation product (codenamed Kingsgate) was completed around March this year and will also adopt chiplet design and faster memory