Microsoft — AI supply-chain exposure

Microsoft sits at the integrator end of the chain — a hyperscaler that buys compute in enormous volume and resells it as cloud capacity and finished AI services. It runs one of the largest fleets of GPUs in the world, wires them into inference-serving infrastructure, and packages the result as the cloud and assistant products businesses rent, along with the model access it distributes on behalf of its partners. It makes none of the silicon; it assembles power, chips and networking into a machine other people pay to use.

Its structural hook is demand aggregation and the two constraints that now bound it: access to accelerators and access to grid capacity. Microsoft is one of the handful of buyers large enough to shape what gets built upstream, and increasingly a builder of its own power arrangements and sovereign-compute programs for governments. The model places it near the centre because it is simultaneously the chain's biggest customer and the gatekeeper through which much of the world reaches frontier models — a distribution position more durable than any single model generation.