TechEx North America’s first day delivered a practical message beneath the excitement around agents and Physical AI: production intelligence depends on electricity, buildings, networks and security controls that develop more slowly than models.
Across tracks covering edge computing, industrial IoT, data centres and cybersecurity, speakers examined what must surround AI before it can operate reliably in factories, critical infrastructure and enterprise workflows.
The recurring problem was the distance between a successful demonstration and a system that survives old equipment, multiple sites, security threats and physical resource constraints.
Edge and industrial AI expose operational reality
Moving inference closer to machines can reduce latency and dependence on a central cloud. It can also create thousands of distributed systems that need consistent updates, observability and access control.
Sessions explored multi-site scaling, agentic network operations, hybrid inference, immutable infrastructure and zero-trust protection for industrial control environments. These concerns matter because an edge decision may affect equipment before a remote operator can intervene.
Industrial IoT and digital-twin discussions returned to “pilot purgatory”: projects that demonstrate value but never become part of daily operations. Modern intelligence often meets machinery and processes designed decades earlier.
A useful digital twin is not merely an attractive visual replica. It should help operators test decisions, predict maintenance or improve a live facility. Ownership and measurable outcomes determine whether it becomes an operational tool or another unattended dashboard.
AI strategy becomes physical inside the data centre
Data-centre sessions focused on construction, power procurement, cooling, water, land and connectivity. Dense AI compute requires all of them, yet infrastructure can take years to permit and build while model economics change quickly.
This mismatch places a material boundary around AI ambition. A company may forecast rapid inference growth without securing grid capacity, cooling systems or network routes. Local authorities and utilities become part of the deployment plan whether technology leaders acknowledge them or not.
Water and power constraints also affect location and public acceptance. Efficiency gains at the chip or model level do not eliminate demand when total usage grows faster. Boards need scenarios that connect AI adoption with electricity, construction timelines and operational resilience.
The data centre is therefore where an abstract software strategy turns into a long-lived capital decision. Poor forecasting can leave expensive capacity unused or prevent successful products from scaling when demand arrives.
Security must extend across data, agents and machines
Cybersecurity sessions covered ransomware, shadow AI, data exfiltration, open-source dependencies and the expanding relationship between the CISO and executive leadership.
Employees may place sensitive information into unsanctioned AI services without logging or review. Approved agents can create similar exposure if their permissions are too broad or their actions cannot be observed. Data governance and cyber governance increasingly describe the same control problem.
Legacy systems make the issue more acute at the edge. Connecting intelligent software to older industrial equipment can expose weaknesses that were previously isolated. Transport and energy environments cannot treat security as a later enhancement.
Zero-trust principles offer a foundation: verify human and machine identities, grant minimal access and assume that every connection needs explicit authority. Updates should be signed, recoverable and consistent across distributed sites.
TechEx’s cross-disciplinary format revealed why AI programmes struggle when responsibility is divided into separate conversations. The model team, facilities group, network engineers and security function are all designing parts of the same system.
Production AI is not switched on after a successful pilot. It is constructed around energy, cooling, data movement, operational processes and people who can respond when something fails.
The most valuable lesson from day one was not that enterprises should slow down. It was that speed comes from acknowledging the entire dependency chain early. Organisations that design infrastructure and security alongside intelligence can move from prototype to production without discovering physical reality only after the business has committed to scale.
