Amazon — AI supply-chain exposure

Amazon operates at both ends of the AI value chain at once. Through AWS it rents out compute at enormous scale, making it one of the largest buyers of accelerators, memory, and networking gear anywhere. But it also designs its own silicon — custom ASICs for training and inference — which the model marks it as producing, letting it serve models on chips it controls rather than relying entirely on merchant GPUs. It is a hyperscaler, integrator of inference serving, and chip maker in one.

The structural hook is vertical integration around demand it already owns. Because Amazon supplies the cloud that startups and enterprises build on, it can steer inference workloads onto its in-house accelerators and capture margin that would otherwise flow to a chip vendor. That optionality — merchant silicon when it needs the best, custom silicon when it wants cost and supply control — is why the model places it centrally: it is simultaneously customer, integrator, and maker, rather than a pure buyer at the mercy of the chip cycle.