According to a report from Digitimes (source), Microsoft is shifting the cost burden of AI computation toward devices...

According to a report from Digitimes (source), Microsoft is shifting the cost burden of AI computation toward devices as its cloud bills continue to climb. This strategic pivot suggests that Microsoft expects to run more AI inference workloads locally on client devices rather than in the cloud, potentially reducing demand for cloud-based AI inference infrastructure. If this trend is adopted broadly, it could dampen demand for AI accelerators (e.g., GPUs from NVDA, AMD) used in hyperscale data centers, as well as impact foundry capacity allocations at TSM. The report is based on a single source with limited detail, so confidence is low. However, if confirmed, this would represent a meaningful shift in AI workload deployment patterns among hyperscalers, with potential spillover effects on semiconductor demand and data center capex plans.