S01.3 · Technology & Digital Infrastructure
The ~$480B hyperscale cloud market, growing ~25% annually as AI workload demand redraws capacity economics.
Cloud Infrastructure & Platform Services is the compute, platform and delivery layer under enterprise and AI workloads. Synergy Research puts the market at an annualized $470-500B in 2025 (off $119B of Q4 quarterly revenue), growing ~25% — and nearly all of that growth is AI buildout, not the cloud migration that carried the prior decade. The constraint has changed accordingly: this market is now gated by capacity — power, cooling, land — not by software.
Forward growth of 20-25% is a capacity forecast as much as a demand one. Demand is global but skews US and China; supply concentrates in the US and is spreading into Asia-Pacific and the Gulf states, where operators are chasing power availability. AWS, Azure and Google Cloud hold roughly 63% combined share (Synergy), with Oracle and the neocloud specialists — CoreWeave, Nebius — gaining share specifically on AI capacity rather than general-purpose workloads. Share gained on scarce capacity deserves a skeptical read: it lasts as long as the scarcity does.
The economics pair extreme capital intensity — multi-year, capex-heavy build cycles — with high margins once capacity utilizes. Revenue is predominantly consumption-based, so vendor revenue tracks workload volume directly; there is no seat cushion here. Upstream sit the chipmakers (Nvidia principally, alongside AMD), data-center REITs and power providers; downstream, SaaS vendors, enterprises and the AI labs themselves. Regulation increasingly shapes the map rather than the margin: data sovereignty and export controls now determine where capacity can be built and who may use it.
Hyperscalers are extending into AI model hosting, data platforms and edge infrastructure. Hosting is where the new dollars are — the fastest-growing pool of incremental spend in the stack. Edge is insurance, protecting against latency-driven bypass as workloads that cannot tolerate a round trip to a centralized region move toward the network edge.
Agentic AI is automating the operations layer itself — provisioning, autoscaling, incident response — which compresses the site-reliability and operations labor that running infrastructure at scale has always carried. It also works against the incumbents' favorite moat. Much of cloud lock-in has lived in provider-specific operational knowledge; as agents encode that knowledge and make it portable, switching gets cheaper and the lock-in earns less.
The structural fact is the new bottleneck. For a decade this segment competed on software features and price; it now competes on megawatts. AI capex already dominates 2025-2026 capacity decisions at every major hyperscaler, which makes power procurement — not product roadmap — the variable that decides who grows.