For most people, the energy transition conjures images of solar panels, wind turbines, and EVs. While those are effective, they overlook the single largest energy-consuming sector on the planet: industry. The industrial sector accounted for nearly 40% of final global energy demand, the largest share of any end-use sector.¹ And though industry is electrifying, it’s not fast enough as global electrification rates have stagnated at around 20% (meaning around 80% of demand is still reliant on fossil fuels).²

Moreover, industry also accounts for the lion’s share of all growth in energy demand (around two-thirds) making it a prime target for efforts to improve energy efficiency.³ In this context, a new generation of technologies, notably physical AI, is emerging as a key enabler that should drive further electrification, energy efficiency and energy security (through reduced fossil-fuel reliance), across industries, particularly industrial manufacturing.

Physical AI marks a step-change for industrial technology, moving AI out of digital environments and into factories, buildings, and power grids. By combining sensing, connectivity, and real-time inference, machines now perceive, decide, and act autonomously in the physical world.

The adoption of physical AI is progressing in three stages. The first, which includes ‘pick-up-and-place' automation, predictive maintenance, and industrial sensing, is already widely deployed. The second, which is being actively piloted, centers on AI-assisted design, advanced process monitoring, and digital twins. The third, at the frontier, involves multi-robot and human collaboration and ‘world-model’ simulations where AI understands physical constraints well enough to manage entire plant environments autonomously (see Figure 2). A mid-single-digit to low-teens share of leading factory automation revenues already includes AI functionality, and AI is embedded in a significant proportion of new product launches.⁴

Conversations with leading industrial technology companies reveal a race to own the physical AI stack, with energy management as a key priority. For example, Siemens* is positioning itself as the backbone of ‘AI for the real world.’ Its wheeled humanoid robot in a live electronics factory environment achieved high autonomous task success rates in logistics operations such as choosing, transporting, and handling containers.⁵

The company’s forthcoming ‘Digital Twin Composer’ combines Nvidia’s* Omniverse simulation with real-time engineering data, enabling large-scale industrial metaverse environments where AI agents simulate process changes with physics-level accuracy before any physical modification is made.⁶ Early pilots have already achieved near-complete design validation and material reductions in capital expenditure.

French industrial giant, Schneider Electric* is also using Nvidia’s GPUs and Omniverse to create digital twins of advanced power systems. These simulated models show the power requirements from the utility grid down to individual chips, enabling precise planning of power distribution and resilience at the facility level.⁷ Japan’s industrial conglomerate, Hitachi*, already deploys AI agents to diagnose industrial equipment issues with high accuracy and very short response times. In power networks, AI-enabled inspection has significantly improved operational efficiency.⁸ Nexans*, a French based cable and power transmission specialist, is using AI in a similar way, embedding AI and sensors throughout grid infrastructure to improve failure prediction and real-time power-flow optimization.⁹

Though these are all different use cases, they all use ‘digital twins’ platforms to generate simulation using real factory data, edge AI embedded into machines and equipment for real-time adaption on the factory floor, and finally, autonomous AI agents that can rapidly synthesize production/processes based on enterprise-wide data. Each AI layer reduces energy and waste and moves manufacturing closer to a net-zero future (see Figure 3).

If AI enables rapid data processing and inference, the brains, then actuators within robots can be characterized as the brawn. Actuators, which turn energy into movement, are a less discussed but a rapidly emerging dimension of factory energy efficiency. Every industrial robot, every collaborative arm, every humanoid is ultimately an assembly of motion-control systems (e.g., actuators, motors, gearboxes, and drives). The energy performance of those components is rapidly becoming an investment focus.

Electric motors in robotic joints achieve roughly 80% efficiency in isolation, but that figure drops to approximately 40% once gearboxes and transmission systems are factored in.¹⁰ Across a global installed base of some 4.3 million industrial robots,¹¹ aggregate energy losses will be substantial. AI-optimized motion planning is a powerful efficiency lever as hardware upgrades. For example, with the help of digital twins linked to Fanuc* industrial robots, engineers reduced energy consumption by 74.8% in pick-and-place operations.¹²

Humanoids, which can contain dozens of actuators as well as other energy-consuming components, are full-bodied energy challenges. For example, Tesla’s Optimus, Agility Robotics’ Digit, and Apptronik’s Apollo at Mercedes-Benz* – run for only two to four hours per charge, constrained by actuator inefficiencies, heat drain, and batteries with low energy density.¹³

These reduce the productivity, efficiency and ultimately the economic viability of humanoids in live factory environments. However, the engineering response is accelerating. Next-generation quasi-direct drive (QDD) actuators offer three to five times greater torque-to-weight ratios versus conventional harmonic drives,¹⁴ while new magnetic gear technologies and high-efficiency brushless motors are pushing actuator efficiency beyond 90%.¹⁵

The companies solving the actuator energy problem – through advanced motor design, power electronics, and AI-driven motion optimization – are building a technology moat at the heart of the next industrial era.

Industry consumes 40% of final energy demand, making it an ideal target for technologies such as physical AI that promise to increase energy efficiency and reduce waste. But it’s not just about adopting AI, the next generation of industrial leaders will be defined by how efficiently they deploy physical AI – combining intelligent ‘brains’ with energy-intensive ‘brawn’.

At the factory level, the ‘brawn’ of physical AI – actuators, motors and robotics systems – are critical drivers of energy demand, reinforcing the need for more efficient industrial processes and motion technologies. As manufacturing electrifies and automation becomes more AI-driven, electricity use is emerging as a core point of competitiveness.

At the same time, the ‘brains’ – semiconductors, AI models and control software – determine how effectively that energy is used, underpinning demand for power-efficient chips, advanced control systems and optimization software. Physical AI systems require high-quality, real-world operational data to optimize performance under real constraints and many incumbents with data and manufacturing platforms are seizing the opportunity, partnering with AI technology leaders to gain a competitive edge.

But AI-driven productivity can also be applied to other areas supporting factory operations including energy distribution and management via smart grids, storage and power electronics that enable reliable, flexible electricity supply. As AI integration broadens, the Smart Energy strategy’s long-standing exposure across the entire energy value chain will help the team identify which companies are best positioned to integrate and deploy AI technologies to optimize energy flows and future growth.

*All companies referenced in this article are for illustrative purposes only in order to demonstrate the investment strategy on the date stated. The companies are not necessarily held by the strategy nor is future inclusion guaranteed. This is not a buy, sell or hold recommendation, nor should any inference be made on the future development of these companies.