Microsoft has introduced the Maia 300, its latest third-generation artificial intelligence processor, alongside a target to deploy upwards of one million units by 2027. Revealed on Monday, October 12, the new hardware is designed to accelerate Microsoft’s AI operations while reducing expenses. The Maia 300 succeeds the Maia 200, which currently runs proprietary Microsoft AI models as well as OpenAI models hosted on the Azure cloud.
Inside Microsoft’s Maia 300 AI Chip Design
Rather than relying on a single large processor, Microsoft opted for a chiplet architecture. The primary processing components utilize TSMC’s 2-nanometer node, while the peripheral circuitry responsible for data transfer is manufactured using a more cost-effective 3-nanometer process.
According to Microsoft, the Maia 300 delivers approximately 40% greater useful computation per megawatt compared to leading GPU-based architectures. Every individual server rack accommodates 96 chips and draws roughly 167 kilowatts of electricity, contrasting with Nvidia’s GB300 racks, which house 72 GPUs and consume about 142 kilowatts.
Rani Borkar, who leads Azure hardware systems and infrastructure at Microsoft, shared with Reuters that the corporation is optimizing the entire infrastructure holistically. The ultimate objective is maximizing performance efficiency relative to every watt, dollar, and memory unit.
Also Read: Microsoft Decision-1: Faster, Cheaper AI Model Built for Structured Business Decisions
How Maia 300 Could Reshape Microsoft’s AI Costs
Major technology corporations are increasingly developing custom silicon to reduce their reliance on Nvidia, the dominant supplier of AI graphics processing units. Microsoft’s latest hardware push mirrors this strategy, granting the company greater oversight of its infrastructure and improved cost management.
Multi-gigawatt deployments of the Maia 300 are slated to begin in 2027 as part of a broader expansion to meet surging computational demands driven by AI adoption. Consequently, the enterprise projects its capital expenditures to exceed USD 175 billion by the year 2026.
Maia 300 highlights the seriousness of Microsoft’s custom silicon ambitions. Successfully reaching the one-million-chip deployment milestone could significantly lower the company’s AI operating expenses and decrease its dependence on Nvidia, fueling an intensifying race among major tech enterprises to design advanced proprietary chips.
