Microsoft Restructures Azure Leadership Amid AI Focus

Microsoft plans to expand its Azure Maia AI chip family to multi-gigawatt deployments starting 2027, aiming to make it the primary system for AI models. The company introduced Maia 300, expected to deliver 40% more tokens per watt than leading GPUs. Maia 200, already deployed, offers 1.7x more tokens per watt in Microsoft Teams and Copilot. The chips are designed for efficiency and demanding AI workloads, with Maia 300 expected to have seven times the token throughput of Maia 200.

Microsoft CEO Satya Nadella urged companies to treat advanced AI models as potential threats, assuming compromise and implementing 'emergency brake' systems. He advised against relying solely on AI model makers' assurances and recommended independent audits, failure disclosures, and shared incident details. This follows recent AI incidents involving Anthropic and OpenAI. Microsoft released AI development guidelines on Sept. 14, emphasizing human control and safety.

Microsoft is reorganizing its Azure cloud unit leadership, with Arun Ulag expanding his role. Scott Guthrie announced the changes, which follow Azure's 42% revenue growth to $29.4B in Q2. Microsoft also unveiled its Maia 300 AI chip, set to power AI services like Copilot from 2027.

The US government suspended eight tech companies, including Tata Consultancy Services, Infosys, Wipro, HCLTech, Cognizant, Microsoft, Adobe, and Capgemini, from the PERM Green Card program. The move aims to address alleged misuse of foreign-worker programs. Existing H-1B visas remain valid, but the suspension may delay permanent residency for affected employees. Companies report limited PERM applications, suggesting a potential minimal impact on their operations.

NVIDIA (NVDA) shares fell 1.6% after Microsoft announced scaling its Maia 300 AI chips. NVIDIA's AI GPU shipments to China dropped 44% YoY in H1 2026. TSMC's results are crucial for NVIDIA's production outlook. Hyperscalers' AI capex is a key demand driver for NVIDIA. The company is exploring investment in open-source startup Reflection AI.

Microsoft introduced Maia 300, its third-gen AI chip, aiming to reduce reliance on NVIDIA GPUs. Deployment starts in 2027, potentially using over 1M chips. Maia 300 offers 40% more work per megawatt than leading GPU systems, with a rack delivering 7x throughput of Maia 200. It uses TSMC's 2nm process and 288GB HBM4 memory, targeting inference workloads. Microsoft will continue using NVIDIA and AMD hardware for training.