Quick Take on Microsoft: Capacity constraints are causing strange bedfellows

研报英文原文证据摘录

Quick Take on Microsoft: Capacity constraints are causing strange bedfellows

cross AI and non-AI workloads Microsoft is continuing the approach of moving 1st party apps to 3rd

party datacenters while leveraging Azure for Azure customers which makes sense. You want to get your customers used to a

fully Azure experience and lock them in for the long term, while you can move your own workloads more easily and not place

future customer demand at risk to go elsewhere.

What is interesting is the ramp in demand that Microsoft is seeing that we are not necessarily seeing elsewhere. We understand

from our checks that customers are being told to wait for Azure capacity due to limitations in Azure capacity (likely where the

workloads need to be located). Office 365 Copilot paid usage grew 33% QoQ (likely faster than most if not all other SaaS AI

workload at that scale) and now we are finding that GitHub usage is more than doubling year-over-year.

WHAT DOES THIS MEAN FOR ORACLE?

From an Oracle perspective the articles ( Times of India article) imply that Microsoft required Fedramp certification which we

understand Oracle Cloud for Government and Oracle Defense Cloud offer but not all other Oracle datacenters and datacenter

offerings which we believe is not atypical. In other words, Oracle did not have the capacity where Microsoft wanted it and the

cost and time of adding certification may have been prohibited, so Microsoft went to the proverbial plan B (Amazon AWS - which

had capacity with certification).

WHAT ABOUT AZURE USAGE TRIAGE?

We have been meaning to write for a while about the issue of where Microsoft is allocating additional Cloud resources. At a high-

level Microsoft has determined, and we would agree, that they must develop their own models.