From BigGo Finance · BigGo Finance · · 1m 30s
Microsoft Unveils Maia 300 AI Chip
Microsoft's new AI chip could deploy over 1 million units next year.
In brief
Microsoft's new AI chip could deploy over 1 million units next year. Microsoft plans to deploy 1 million Maia 300 AI chips next year for internal services, cutting reliance on Nvidia GPUs. Originally reported by BigGo Finance.
Microsoft Unveils Maia 300
Microsoft has unveiled its third-generation in-house AI accelerator, the Maia 300, built on Taiwan Semiconductor Manufacturing's 2-nanometer process using a chiplet architecture.
288GB HBM4 Memory
The Maia 300 features 288GB of HBM4 memory with 18TB per second bandwidth, a 33% increase over the previous generation.
40% Higher Token Output
Microsoft claims the Maia 300 processes 40% more tokens per megawatt than leading GPU systems, with rack-level throughput seven times that of its predecessor.
Prioritizing Internal Services
Microsoft plans to use Maia chips for internal AI services like Copilot rather than renting out compute capacity externally.
1 Million Units Next Year
Single-year deployment could exceed 1 million units starting next year, prioritizing Microsoft's internal AI needs.
Reducing Nvidia Reliance
Microsoft aims to reduce reliance on Nvidia GPUs amid $175 billion in capital expenditure pressure.
Chiplet Design Strategy
The Maia 300 uses a hybrid 2-nanometer process for compute and 3-nanometer for data movement, optimizing cost and performance.
Co-opetition with Nvidia
Microsoft emphasizes expanding Maia deployment is not about cutting Nvidia products but increasing overall infrastructure capacity.
Hardware-Software Co-Design
Microsoft's strategy focuses on hardware-software co-design to improve AI business margins through in-house silicon.
What it means
Microsoft plans to deploy 1 million Maia 300 AI chips next year for internal services, cutting reliance on Nvidia GPUs.
