The conversation about AI and its influence on industry often centers on software. At the recent IFS.ai Industrial X Unleashed event, two sessions broke that pattern.
Dr. Merry Frayne, Director of Product at Boston Dynamics, and Dr. Sabine Erlinghagen, CEO of Siemens Grid Software, each examined a different corner of industrial transformation, but both landed on the same point – AI is becoming a force in the physical world, not just the digital one. Boston Dynamics' robots are learning to move, perceive, and act in environments built for people. The energy grids that Siemens' software analyzes are being asked to carry demand that rises faster than infrastructure can be built.
Taken together, the two sessions offer a snapshot of how the next era of industrial systems will depend on intelligent physical platforms that gather data, support human decision-making, and prevent failures long before they occur – as well as highlighting the growing gap between AI ambition and the physical energy infrastructure needed to power it.
The service era of robotics arrives
Frayne describes a field that has shifted from what she calls the "industrial era of robotics" – fixed, fenced-off systems designed for repetitive tasks – into what she terms the "service era," where robots operate in human environments and perform useful work with far more autonomy than before. Advances in AI tools, she explains, are the reason this transition is accelerating.
Building a general-purpose robot now requires a blend of mobility, perception, and manipulation. Mobility covers the basics: the robot must move safely through real buildings and rough terrain and reach places that humans cannot. Spot, Boston Dynamics' quadruped robot, for instance, already operates in hazardous locations such as the irradiated zone of the Fukushima Daiichi nuclear power plant.
Perception is about understanding what the robot sees. Spot can recognize a ladder and treat it as a potential hazard, rather than simply detecting its shape. It can identify cords, obstacles, and movement in ways that show an emerging ability to interpret context, not just objects. The robot's sensors extend beyond human capability: thermal cameras track machine health over time, while acoustic microphones listen in ultrasonic ranges for air leaks, steam leaks, and partial electrical discharge.
Boston Dynamics identifies three primary customer segments for Spot – industrial use cases focused on asset management and predictive maintenance, incident response involving high-stakes, remote-controlled operations such as explosive ordnance disposal, and research applications where the robot serves as a platform for adding new capabilities. Frayne notes that one customer, a seed oil manufacturer, uses Spot to detect spilled seeds on the factory floor – a deliberately mundane example that illustrates how the platform adapts to specific operational needs rather than requiring bespoke engineering for each use case.
The acceleration enabled by AI is substantial. Frayne explains that behaviors which would have taken months or years to develop using traditional physics-based models can now be built in a single day. Boston Dynamics is applying this to Atlas, its bipedal humanoid robot, with initial applications in automotive manufacturing. The company's partnership with IFS focuses on integrating robot-collected data into broader work streams – turning sensor readings into actionable maintenance tasks rather than standalone data points.
Frayne concludes:
An investment in physical AI is an investment in the future of humanity.
Grid infrastructure struggles to keep pace
Erlinghagen opens with a warning:
The grids are risking to become the bottleneck of the AI revolution.
Before the current wave of generative AI, electricity demand was already set to triple by 2050, driven by electrification of transport and heating. AI data centres have added a new layer of pressure. Erlinghagen notes that AI factories are predicted to require electricity equivalent to the entire economy of Japan – a country of 125 million people.
The numbers are stark. The connection request queue for new generation in the US currently stands at 2.6 terawatts, with 95% of that renewable. The same capacity needs to be added on the demand side. Average waiting times for grid connection have risen from two to three years previously to five years now. Erlinghagen observes that telling Microsoft or any AI company to wait five years for a data centre connection means the AI revolution will move much slower than anticipated.
The required investment is significant – $1.4 trillion in US grid infrastructure alone by 2030. Erlinghagen argues that even a fraction of a percentage point of misallocated investment represents a large sum, making data-driven decision-making essential.
She offers a concrete example. A Canadian utility with $5.2 billion in annual capital expenditure faced two competing priorities: deploying capital effectively and maintaining grid resilience. The utility used Siemens grid simulation technology for technical optimization and IFS Copperleaf for financial planning. When run separately, each system identified a different optimal scenario. When Siemens and IFS co-simulated the problem, a third scenario emerged that outperformed both the pure technical and pure financial choices. Without that integration, Erlinghagen explains, the utility would have either misallocated capital or compromised on grid stability.
Grid resilience is already under pressure. Outage costs in the US have risen 175% in the past 12 months, totalling $150 billion in economic impact. Siemens is addressing this through equipment monitoring that uses AI and data-driven approaches for both predictive and preventive maintenance, with automated handoff to work management systems.
A demonstration showed how this works in practice: Siemens electrification software monitors grid assets and flags early wear and tear, passing that data to IFS for anomaly detection. When an anomaly is detected, a work order is created automatically. An AI agent then helps bundle that intervention with other planned work at the same site to reduce operational costs, before optimizing crew schedules to ensure critical assets are serviced first. The goal is what Erlinghagen describes as the "autonomous grid" – a system where anomaly detection triggers work orders without human intervention until a decision point requires it.
My take
These two sessions, presented independently, tell a connected story. The constraint on AI's next phase is not compute or algorithms – it is physical infrastructure. For Boston Dynamics, the breakthrough is robots that can finally manipulate the physical world reliably and learn new behaviors quickly. For Siemens, the bottleneck is a grid that cannot keep pace with the demand that AI itself is creating.
The implication for enterprise buyers is clear. Investment in physical AI and grid infrastructure is becoming a prerequisite for realizing AI's value, not an afterthought. The Canadian utility case study is a strong example that siloed optimization - whether technical or financial – produced suboptimal results. Only by integrating both perspectives did the best answer emerge. That lesson extends beyond utilities. As AI moves from the digital realm into physical systems, the organizations that treat infrastructure investment as a strategic priority, rather than a cost centre, will be better positioned to capture value from what comes next.