Physical AI as a holistic, distributed system

The deployment of physical AI is not a question of if – it’s already here. Behind the spectacle of headline-grabbing humanoid demos, robotic vacuum cleaners, industrial robots, drones and autonomous guided vehicles have been operating reliably for some time. Humanoid robotics alone is forecast to expand to $5 trillion in the next decade, driven by ageing populations, labour shortages, reshoring of manufacturing and heightened security demands.

What has held the next wave of physical AI back is not ambition or demand, but economic and operational practicality.

As physical AI systems become more intelligent and autonomous, they must operate in complex, dynamic environments alongside people. This relies on continuous loops of sensing, perception, decision-making and action. Cameras, LiDAR, radar and other sensors generate vast amounts of data that must be processed in near real time, often within milliseconds. Any delay or loss of connectivity compromises performance and safety.

Trying to solve this by putting more compute on the robot makes them more expensive, power-hungry and difficult to scale or maintain. Equally, pushing all intelligence to a distant cloud introduces unacceptable latency and reliability risks.

We believe the practical solution lies in distributing intelligence across the system.

Telecoms as the quiet but powerful enabler of physical AI

So how would this work? One way to think about physical AI is as an intelligent pipeline. At one end sits the robot or machine interacting with the real world, at the other sit centralised data centres and AI factories that train and refine large models. Between them lies a continuum of intelligence: some on the device, some at the edge and some further away.

In most real-world physical AI deployments, the edge is where value is unlocked. Placing high-performance compute close to where robots operate, in factories, hospitals, campuses, logistics hubs, or secure facilities, allows heavy AI workloads to be offloaded while still meeting strict latency and reliability requirements.

This delivers multiple benefits at once. Robots become cheaper and lighter, battery life improves and systems become easier to upgrade as AI models evolve. Most importantly, safe coordination in human-occupied spaces becomes achievable because intelligence can react in real time.

But this only works if the network can guarantee performance. Vision-based robotics, for example, may offload perception or planning tasks to the edge – but even small amounts of lag mean the robot is reacting to the past, not the present.

This is why private 5G, designed as part of an intelligent edge architecture, is ideally placed to meet these requirements and support physical AI at scale. By re-architecting physical AI so that some workloads run on the robot, some execute on-site at the network edge over private 5G, and others execute in AI data centres, it becomes possible to balance performance, cost, power consumption and safety.

This shift transforms physical AI from a device-centric challenge into a distributed intelligence challenge to provide the compute, network, storage and associated software services to make it all work. This challenge is not one that can be solved with best-effort networking, instead requiring the guaranteed quality of service that 5G can offer.

Here lies the opportunity for telcos. It’s clear that physical AI is the direction we’re heading, so by enabling this intelligent connectivity, telecoms can become the essential enabler of physical AI systems across industries – as long as they innovate first.

Physical AI in action

Our physical AI demo at MWC showcases exactly this, illustrating how physical AI can move from concept to real-world impact. Viewers can interact directly with an autonomous humanoid robot, using simple gestures to instruct it to carry out tasks such as two-arm object picking and box handling.

At the core of the demo is a fine-tuned version of NVIDIA’s state-of-the-art Physical AI foundation model, trained using data collected in our labs. Combined with our unique human-robot interaction capability, whole body control and fine manipulation development, the system demonstrates robust, safe collaboration between humans and robots in real-world tasks.

But beyond these technical capabilities is one crucial point: much of the intelligence in this system does not sit on the robot itself. Advanced AI processing is offloaded to nearby edge compute, connected via private 5G. This enables high performance and safe operation while reducing robot cost, power consumption and system complexity.

What’s being demonstrated here is a practical blueprint for the industrial deployment of physical AI across healthcare, defence and security and automated industry – and how telecoms leaders can place themselves at the centre of this wave of AI innovation.

Moving up the value stack

Physical AI doesn’t just demand technical innovation; it also requires new business models. Traditional telco pricing based on SIMs or data volumes does not reflect the value delivered by intelligent physical systems, nor does it justify investment in edge AI compute and assured performance.

Instead, physical AI demands service-based and outcome-driven models. Customers are not just buying connectivity; they’re buying safe, reliable, intelligent systems that deliver measurable productivity gains. Physical AI systems span a complex stack: devices, sensors, radios, edge compute and storage, AI platforms and cloud infrastructure. Few enterprises want to integrate and operate this entire stack themselves.

Telecoms providers are uniquely positioned to orchestrate the core of this stack, hosting AI platforms, providing edge infrastructure, guaranteeing performance and ensuring compliance, resilience and sovereignty. Once enterprises depend on this guaranteed intelligence layer, the telco moves from a commodity provider to a strategic partner.

Achieving this vision requires deep expertise in both physical AI systems and 5G infrastructure. As the deep tech powerhouse of Capgemini, CC is uniquely placed to support telecoms players in seizing this opportunity by defining winning strategies for physical AI, identifying the right partners, designing end-to-end architectures and moving rapidly from concept to deployment. From strategic direction-setting to rapid implementation, we help telecoms providers position themselves at the heart of the physical AI revolution.

Telcos must seize this opportunity, or others will

If telecoms providers do not step into this role, others will. Hyperscalers are already offering integrated compute, private wireless and AI platforms that threaten to bypass traditional telco models entirely.

This makes physical AI both an opportunity and a warning. It is one of the few emerging domains where telecoms can be fundamental to making the system work across multiple industries. The technology is ready, the demand is real, the value is moving up the stack. The only question is whether telecoms will choose to enable physical AI, or watch others define the future without them.

Reach out to frank.long@cambridgeconsultants.com or michael.massey@cambridgeconsultants.com to continue the conversation.