ai AM market analysis — 2026-09-18

The AI market’s am session centres on two intertwined contests: how open foundation models should be released, and how enterprises will pay for the inference infrastructure that production use demands. Stanford HAI frames open weights as a driver of broader adoption that simultaneously raises governance and accountability questions for providers. Bruegel’s analysis of foundation model competition points to a market shaped less by a single dominant approach and more by differing degrees of openness, proprietary resources and business models.

On the infrastructure side, reporting from Digitimes and market analysis of the inference sector both describe a shift as generative workloads move from experimentation toward routine production, pushing demand toward cloud and clustered accelerator environments. That shift keeps inference capacity, rather than training capacity alone, as a focal point for enterprise AI spending.

Neither thread points to a settled outcome. The economics of open-model commercialisation remain unresolved, and the balance between specialised accelerators and cloud-scale compute is still being worked out. The evidence supports a cautious, qualitative read on the sector rather than a directional call.