ai AM market analysis — 2026-09-15

The AI narrative is turning from model launches to the economics that sit underneath them. Cloud providers are directing large amounts of capital into infrastructure and custom silicon, treating capacity and chip design as competitive ground rather than a shared utility. That shift reframes the near-term story as one of spending discipline and utilisation, not headline model capability.

Attention is also moving toward inference. Analysts and enterprise-focused research point to production inference, rather than experimental pilots, as the demand source that will determine whether current infrastructure investment pays off. Alongside this, research on open foundation models suggests the incentives driving their release are broader than direct commercial return, which keeps governance and control questions live even as adoption widens.

Taken together, the evidence points in a consistent direction on where investment is flowing, but is less conclusive on payoff. Custom silicon and capacity commitments have not yet been tested against durable pricing power, and open-model governance frameworks remain a work in progress rather than a settled arrangement. The near-term picture is one of infrastructure build-out running ahead of proof that it converts into lasting commercial advantage.