Microsoft Unveils Maia 200 Chip Delivering 10 Petaflops and 3× AWS Throughput

MSFTMSFT

Microsoft unveiled its Maia 200 AI inference chip featuring over 100 billion transistors delivering 10 petaflops at 4-bit and 5 petaflops at 8-bit precision—offering 3× FP4 throughput versus AWS Trainium3 and outpacing Google’s TPU-v7. The company will deploy Maia 200 nodes in US Central and West data centers, enhancing Copilot and Superintelligence workloads.

1. Q2 Fiscal 2026 Earnings Preview Highlights Resilient Revenue Outlook

Visible Alpha consensus estimates show Microsoft’s fiscal Q2 2026 total revenue forecasts have held steady since late July 2025, reflecting confidence in the company’s core segments—Productivity and Business Processes, Intelligent Cloud and More Personal Computing. Analysts expect growth to be driven by continued strength in Office 365 Commercial seat adds (projected to exceed 20% year-over-year) and Azure consumption revenue, which grew 29% in Q1 FY26. While AI-related services are still ramping, management has guided to a modest sequential slowdown in Intelligent Cloud growth and slight margin contraction due to elevated infrastructure spend.

2. Wisconsin Data Center Expansion Secures Long-Term Capacity

Mount Pleasant, Wisconsin’s village board unanimously approved site plans for 15 additional data centers on land Microsoft acquired in 2023 and 2024. The two parcels will accommodate nearly 9 million square feet of building area and three new substations. This expansion is expected to support booked revenue from OpenAI and other AI customers, and create construction and operations roles sustained over the next 10 years. The taxable value of the proposed development exceeds $13 billion, reinforcing Microsoft’s strategy to scale capacity for generative AI workloads.

3. Introduction of Maia 200 Chip Targets AI Inference Efficiency

Microsoft unveiled its second-generation in-house AI inference processor, the Maia 200, featuring over 100 billion transistors and delivering more than 10 petaflops at 4-bit precision and approximately 5 petaflops at 8-bit. Built on TSMC’s 3 nm process, each server node links four chips via Ethernet, and clusters up to 6,144 units to optimize energy use and total cost of ownership. The Maia 200 is already deployed in U.S. Central data centers and is integrated into Copilot workloads and internal superintelligence model serving, offering a reported 3× FP4 performance advantage over AWS’s third-generation AI accelerators.

4. Cloud Adoption Remains Primary Growth Catalyst Despite AI Headwinds

Equity research firms maintain a strong buy stance on Microsoft, citing robust cloud momentum even as enterprise AI uptake progresses more slowly than expected. Cloud infrastructure spend represented over 60% of capital expenditure in FY25, with capex-to–operating cash flow rising toward a projected 50–55% range in FY26 as Microsoft scales Azure regions and AI hardware investments. Analysts note that while Copilot seat penetration is below initial targets, the broader Microsoft 365 installed base and Azure consumption continue to deliver high-teens revenue growth and sustain free cash flow margins above 30%.

Sources

B2ZFF
+15 more