AI infrastructure race shifts towards power and execution

•AI inference workloads move into secondary markets where power is available

•Power access, fibre density and deployment speed replace capacity as top operator priorities

The fact

At ITW 2026, data centre and connectivity executives discussed how AI demand is reshaping infrastructure deployment. DartPoints CEO said AI inference workloads are commercially viable at scale and less latency-sensitive than traditional cloud, allowing deployments into secondary markets with available power. Zayo said interconnected AI agents could increase bandwidth demand sevenfold, while FarmGPU estimated many GPU clusters operate at only 30%–40% utilisation.

The assessment

The shift is from expansion narratives to execution-led competition. Power access, fibre densification and deployment speed are now operating requirements, not future planning items. AI inference's lower latency sensitivity weakens the industry's traditional hub-first model, increasing the relevance of secondary markets. For BTW readers, infrastructure value is increasingly judged by commercial flexibility and deployable capacity, not network footprint alone.

What to watch

Watch whether hyperscalers and AI inference operators begin committing larger deployments into secondary regional markets where power availability, metro fibre density and flexible deal structures can be secured together.

Also read: UK forum commissions lunar spectrum framework study

Also read: atNorth joins Nordic Compass AI infrastructure alliance