When most analysts discuss Physical AI, they picture humanoid robots — Tesla Optimus, Figure 01, Boston Dynamics Atlas. This framing is understandable but misleading. It focuses on the most visible application and misses the structural transformation underneath it.
Physical AI is not primarily about robots. It is about the infrastructure that makes the physical world legible to machine intelligence — and the construction of that infrastructure is the largest hardware buildout in human history.
THE CORRECT DEFINITION
Physical AI rests on three interconnected concepts: autonomy, digitalisation and communication. Together they create the infrastructure and space within which AI can perceive, reason about and act in the physical world.
Digitalisation: converting physical reality into machine-readable data
Communication: transmitting that data at the speed and scale AI requires
Together: the infrastructure layer between the physical world and AI models
The automobile is the most instructive example. A modern autonomous vehicle is not primarily a car — it is a robot, the first autonomous robot deployed at mass scale in the real world. Understanding the automotive industry as the first Physical AI market reveals the entire ecosystem: chips for perception, networks for communication, software for safety, sensors for environmental mapping, and the power infrastructure to charge and operate it all.
THE SIX LAYERS OF PHYSICAL AI INFRASTRUCTURE
Layer 1 — Compute chips
The intelligence layer. ARM architecture dominates embedded compute across Physical AI applications — from automotive SoCs to IoT endpoints — because its power efficiency at the edge is unmatched. Qualcomm brings together compute, modem and AI acceleration in single chips that power the connected, autonomous devices at the heart of Physical AI. MRAM (magnetoresistive RAM) addresses the data persistence requirements of edge AI systems that cannot afford power interruptions.
Layer 2 — Connectivity and network equipment
Physical AI generates and requires the transmission of enormous data volumes in real time. Cisco provides the enterprise and industrial networking infrastructure. Nokia is positioning itself at the intersection of private 5G networks and industrial automation — the connectivity backbone for smart factories, autonomous logistics and connected infrastructure. The transition to 6G will define the upper bound of what Physical AI can accomplish by 2030.
Layer 3 — Testing and measurement
This is the most underappreciated layer. Every Physical AI device, network component and safety-critical system must be tested before deployment. Keysight Technologies provides the electronic test and measurement equipment that validates chips, networks and systems. Teradyne tests the semiconductors. Without test infrastructure, Physical AI cannot scale — yet this layer receives almost no attention in mainstream analysis.
Layer 4 — Safety-critical software
BlackBerry QNX is the operating system running in approximately 235 million vehicles. It is the software layer that makes autonomous vehicles safe — certified to the highest functional safety standards (ISO 26262 ASIL D). As Physical AI expands from automotive to industrial automation, medical devices and critical infrastructure, safety-certified software becomes the non-negotiable foundation.
Layer 5 — Vision and spatial mapping
Physical AI must perceive and navigate the real world. TomTom provides the high-definition mapping data that autonomous systems use to understand their environment. Aeva's FMCW LiDAR technology delivers 4D perception — measuring not just position but velocity — enabling more reliable real-world interpretation. This layer converts physical space into the digital representation that AI models can reason about.
Layer 6 — Power infrastructure
Every layer above depends on this one. Chips need power. Network equipment needs power. Test facilities need power. Charging infrastructure for autonomous vehicles needs power. Smart factories running Physical AI systems need power. The buildout of Physical AI infrastructure is, at its foundation, a massive electrical infrastructure project — and it faces the same transformer shortage, grid connection constraints and power equipment lead times that are already blocking AI data center deployment.
THE DIGITALISATION OF THE WORLD
The profound implication of Physical AI infrastructure is not the individual applications it enables. It is the cumulative effect: the conversion of the physical world into machine-readable data at a scale and resolution that has never existed before.
Every connected sensor, every autonomous vehicle, every smart industrial system is simultaneously a consumer of Physical AI infrastructure and a producer of the data that trains future AI models. The infrastructure buildout and the AI capability improvement are mutually reinforcing — each makes the other more valuable.
Autonomous vehicles by 2030: projected 8–12 million
Industrial robots by 2030: projected 20+ million units
Aggregate power demand: hundreds of GW of distributed new load
Grid infrastructure requirement: distributed upgrades at millions of locations
WHY THIS MATTERS FOR POWER INFRASTRUCTURE
The distributed nature of Physical AI infrastructure creates a power challenge that is different from — and complementary to — the centralised data center challenge. While data centers require single large grid connections of 100–500 MW, Physical AI deployment requires upgrading millions of medium-voltage connections at factories, logistics facilities, transport depots and urban infrastructure nodes.
The same transformer shortage that is blocking data center construction is also constraining the electrical upgrades that Physical AI deployment requires at industrial sites. The medium-voltage switchgear, distribution transformers and grid connection capacity that were sized for pre-AI industrial loads are insufficient for the Physical AI economy.
THE INVESTMENT FRAMEWORK
For investors seeking exposure to Physical AI, the robot-centric view leads to concentrated bets on uncertain application outcomes. The infrastructure view — spanning all six layers — provides a more diversified, more defensible and ultimately larger opportunity.
The power infrastructure layer is the most defensible of all: it is required regardless of which Physical AI applications win, which robot designs prevail, or which connectivity standard dominates. Electricity is not optional.
"Do not think about Physical AI as robots. Think about it as the infrastructure and connectivity that digitalises the world — and the power infrastructure that makes all of it possible." — GridReadiness Intelligence