ai PM market analysis — 2026-09-23

The AI sector’s centre of gravity is moving from model training to distribution and infrastructure economics. Foundation-model providers are pushing further into agent-building and developer software, a shift that puts them in direct competition with some of the customers and independent builders who rely on their platforms, according to AgentMarketCap. Bruegel frames this as a competition problem, pointing to risks around vertical integration, bundling and preferential treatment by dominant platforms.

Underneath the platform contest sits a harder constraint: inference capacity. Rising inference demand is running up against limits in compute, power and deployment economics, which will shape which products can scale commercially rather than remain in pilot form. Whether that demand proves durable depends on enterprises converting experimentation into recurring production workloads, a transition the market-sizing research treats as the key swing factor for inference-related infrastructure spending.

Taken together, the picture is mixed rather than one-directional. Platform expansion by foundation-model providers creates both opportunity and competitive risk depending on where a firm sits in the value chain, while infrastructure economics remain constrained by capacity and cost rather than settled. Regulatory attention to bundling and self-preferencing adds a further variable that could alter how platform power is exercised over the coming year.