Microsoft Launches New AI Chips Amidst Global Workforce Adjustments
Translated & summarized from Ice by baba
Microsoft has launched Maia 300, its third-generation AI chip, aiming to produce over a million units next year to support its AI services and reduce dependence on Nvidia. The new chip features a chiplet architecture and improved energy efficiency, offering an estimated 40% more effective processing per megawatt. In parallel, Microsoft is undergoing workforce adjustments in its Israeli R&D centers, with about 50 employees facing layoffs, while Ohad Yassin prepares to take over as CEO of Microsoft Israel R&D.
The story in 5 lines · by baba
- Microsoft unveiled Maia 300, its third-generation AI chip, with plans to produce over a million units next year.
- The new chiplet architecture aims to compete with Nvidia by offering improved processing and energy efficiency.
- Microsoft claims its new system provides about 40% more effective processing per megawatt compared to leading GPUs.
- The company is reducing its workforce in Israel, with around 50 employees from the Copilot Studio group facing layoffs.
- Ohad Yassin will become the new CEO of Microsoft Israel R&D in November, replacing the Outgoing CEO of Microsoft Israel R&D.
Microsoft is advancing its commitment to independent AI infrastructure with the unveiling of Maia 300, the third generation of its proprietary AI chips. The company plans to distribute over a million of these new chips next year to bolster its internal AI services and enhance processing capabilities. According to Reuters, the Maia 300 is designed for complex AI tasks, including code generation and business operations support.
Previous generations of these chips already power Microsoft's internal models and OpenAI's models offered via its Azure cloud service. A key innovation in Maia 300 is its chiplet architecture, which differs from Nvidia's approach. Instead of a large central processing chip and separate memory components, Microsoft's design integrates multiple specialized chiplets. The core processing units are manufactured using TSMC's 2-nanometer technology, while data interconnect components use the more economical 3-nanometer technology.
Energy efficiency is also a major focus. Maia 300 server racks are expected to house 96 chips and consume approximately 167 kilowatts, compared to Nvidia's GB300 racks, which contain 72 GPUs and use 142 kilowatts. Microsoft claims its new system offers about 40% more effective processing per megawatt than leading market GPUs. Rani Borker, President of Hardware Systems and Infrastructure at Azure, stated the goal is to provide "more intelligence per watt, per dollar, and per byte of memory."
This chip development aligns with a broader trend among cloud and AI companies to create their own silicon, reducing reliance on Nvidia and tailoring hardware to specific service needs. Microsoft intends to begin deploying several gigawatts of this infrastructure starting next year, amidst projections of capital expenditures exceeding $175 billion for the current calendar year.
Concurrently, Microsoft is implementing organizational changes within its Israeli R&D centers. Approximately 50 employees, primarily from the Copilot Studio group focused on AI agents, are facing potential layoffs as part of a global restructuring. The company estimates that some affected employees may remain within the R&D center or be offered alternative roles within Microsoft. Microsoft employs around 3,000 R&D personnel across six sites in Israel, with its largest facility located in Herzliya.
These workforce adjustments coincide with a leadership transition at the Israeli R&D center. Ohad Yassin is slated to assume the CEO role in early November, succeeding the Outgoing CEO of Microsoft Israel R&D, who is departing the position.