Reports say Microsoft is reshaping Azure leadership, widening Arun Ulag’s remit

Reports published October 13 say Microsoft cloud chief Scott Guthrie told Azure employees on October 12 that the platform’s leadership was changing, including a broader remit for Arun Ulag. Benzinga and The Next Web both attribute the account to The Information, rather than presenting separate reporting on the change (Benzinga; The Next Web).

What was reported about Azure leadership

Ulag previously led Azure database products. Benzinga reported that his expanded remit would cover core Azure and its main engineering teams. The reports describe a leadership change, not a move that Microsoft has separately announced in its Maia infrastructure update.

Maia 200’s production role and Microsoft’s 2027 plans

In a separate October 12 announcement, Microsoft said Maia 200 was serving production workloads, including Microsoft Copilot, and supporting model development by its MAI team. Microsoft said Maia 200 runs OpenAI models, Microsoft MAI models and leading open-source model families. The company also expected GitHub Copilot and Microsoft Foundry to follow by year-end 2026. Microsoft’s announcement says Maia 200 delivers up to 1.7 times more tokens per watt than leading GPU infrastructure for built-in AI features in Microsoft Teams, Copilot and Cowork experiences.

Microsoft plans to begin multi-gigawatt deployments of its Maia systems in 2027, with Maia intended to become the primary inference system for Microsoft products and services. Production deployments of Maia 300 are also planned to begin in 2027. Those milestones are future plans; Maia 200’s stated production use is a separate current-status claim.

What Microsoft’s Maia 300 figures measure

Microsoft’s up-to-seven-times figure refers to rack-scale token throughput—the amount of text-processing output across a rack—compared with Maia 200 in a full configuration of 96 accelerators. Its separate efficiency comparison is 40% more tokens per megawatt than leading GPU-based systems. One figure concerns rack-level throughput; the other compares tokens produced per unit of power.