Run the Workload

Capital starts here. Every dollar flowing through the AI infrastructure supply chain originates from this layer — a hyperscaler or cloud provider writing a check. Amazon, Microsoft, Google, and Meta together spent over $200B on capex in 2025, with 2026 projections even higher. When Microsoft announces $80B in data center spending, every company from L01 to L25 feels the demand signal. This is the layer that pulls the entire supply chain forward.

The structural question is whether this capex cycle is sustainable or a bubble. The Inference Architecture Shift trend provides the bull case: if inference demand grows 10x as AI products reach billions of users, the current capex is not enough — it's just the beginning. The bear case is that capex is front-loaded for training infrastructure that will have excess capacity once foundational models converge. The truth is probably somewhere in between, but the direction of the bet matters enormously for every upstream layer.

CoreWeave represents the neo-cloud tier — purpose-built GPU cloud providers that compete with hyperscalers on GPU availability and price rather than breadth of cloud services. Their $7.5B IPO in early 2025 validated the market's appetite for pure-play AI compute infrastructure. DigitalOcean occupies the developer-focused cloud tier, serving smaller AI companies and inference workloads that don't need hyperscaler scale.

Oracle has emerged as an unexpected AI infrastructure player through aggressive data center buildout, signing multi-billion-dollar deals with OpenAI and others. Their cloud strategy leapfrogged the traditional enterprise cloud market by going directly to AI workloads.

The unique feature of this layer: these companies are both buyers and sellers of AI infrastructure. Amazon builds data centers to sell AWS capacity. Microsoft builds them to sell Azure. Their capex is someone else's revenue — NVIDIA, Vertiv, Eaton, Quanta, TSMC. This creates a reflexive dynamic: if AI product revenue disappoints, capex gets cut, and the entire supply chain from foundries to construction crews feels it simultaneously. That reflexivity — the demand signal that can reverse — is the systemic risk embedded in L26 and the reason this layer matters more than its low bottleneck score suggests.