Why AI Is Moving Off the Screen and Into the World

Until recently, AI largely lived in the cloud and on screens: recommendation engines, digital ads, chatbots, predictive analytics dashboards. That phase is ending. Fuelled by breakthroughs in generative AI, cheaper sensors, powerful edge chips and massive government investments, AI is now being embedded in things, places and national infrastructures.

Analysts at the World Economic Forum describe this as a new era of physical AI, where intelligent robots and machines leave controlled factory cages and begin working alongside people in dynamic environments. Advisory firms and research houses such as Gartner simultaneously highlight ambient intelligence and sovereign AI as top strategic technology and policy trends for the coming decade.

The result is a powerful three-way convergence:

• Physical Robots and autonomous systems become more capable, adaptable and affordable.
• Ambient Buildings, hospitals, factories and cities become instrumented and responsive through ubiquitous sensing and edge AI.
• Sovereign Governments treat AI models and semiconductor fabs as strategic assets, rewiring supply chains around national and allied interests.

For business and policy leaders, understanding this landscape is not a theoretical exercise. It shapes where capital will flow, how value will be created and captured, what risks must be managed and which geographies will lead or lag in the AI-powered economy.

Physical AI: Intelligent Machines in Motion

From industrial robots to embodied intelligence

Physical AI refers to AI systems that are embodied in physical machines — robots, drones, autonomous vehicles and smart equipment that can perceive, reason and act in the real world. Instead of AI limited to pixels and datasets, physical AI combines:

• Perception — cameras, lidar, radar, microphones and tactile sensors streaming rich data.
• Cognition — machine learning models interpreting that data, planning and making decisions.
• Action — motors, actuators and end-effectors that manipulate or move through physical space.

This is a step change from earlier generations of industrial robots that repeated pre-programmed motions inside caged work cells. Today’s embodied systems can navigate cluttered warehouses, collaborate with humans on assembly lines, handle variable objects and even learn new tasks through demonstration.

Key technology enablers

Several technology waves are converging to unlock physical AI at scale:

• Advanced sensors and mechatronics. High-resolution cameras, depth sensors, force-torque sensors and better actuators dramatically expand what robots can see and do.
• Powerful perception and control models. Vision, speech and reinforcement learning models allow robots to recognize objects, understand scenes, plan paths and adapt to uncertainty.
• Simulation and digital twins. Platforms from companies like NVIDIA make it possible to train robot policies in photo-realistic virtual environments, generating synthetic data at scale before deployment in the real world.
• Edge AI hardware. Compact AI modules such as NVIDIA Jetson and similar systems from Intel and others bring powerful inference to the robot itself, reducing latency and cloud dependency.

Real-world applications and early adopters

Physical AI is no longer experimental; it is quietly becoming a backbone technology in operations-intensive sectors:

• Logistics and e‑commerce. Amazon operates hundreds of thousands of robots in its fulfillment centers, combining mobile shelf movers, robotic arms and AI planning systems to compress delivery times and mitigate labor shortages.
• Electronics and automotive manufacturing. Contract manufacturers such as Foxconn and original equipment manufacturers are deploying AI-enabled cobots for complex assembly tasks, quality inspection and flexible production changeovers.
• Autonomous vehicles and drones. Self-driving programs at companies like Tesla and leading AV startups rely on large-scale neural networks to interpret sensor feeds and control vehicles in real time. AI‑powered drones are being trialed for inspection, agriculture and last‑mile delivery.
• Healthcare and services. Hospitals are experimenting with autonomous delivery robots, robotic surgery assistants and telepresence systems that combine computer vision and language models to interact more naturally with patients and staff.

Early adopters report gains in throughput, safety and resilience, with new human roles emerging around robot supervision, maintenance and orchestration.

The emerging Physical AI technology stack

A recognizable five‑layer technology stack is emerging around physical AI deployments:

1. Robotic hardware — the physical platform (mobile bases, arms, grippers, drones) and sensors.
2. Edge compute — on‑board CPUs, GPUs or accelerators handling real‑time inference and sensor fusion.
3. Robot operating system and middleware — software layers such as ROS and proprietary frameworks that coordinate components and expose APIs.
4. Simulation, training and data infrastructure — digital twins, synthetic data generation and MLOps for continuous learning.
5. Applications and integration — workflow logic, user interfaces and connectors into enterprise systems, often delivered as Robots‑as‑a‑Service.

New entrants and incumbents are positioning at different layers of this stack, with ecosystem partnerships increasingly critical. Chip designers, robot OEMs, cloud providers, systems integrators and consultancies must work together to deliver full solutions.

Opportunities, constraints and risks

The opportunity is substantial: frontline automation to offset demographic pressures, improve safety and unlock new business models. Yet several constraints remain material:

• Data and “common sense”. Robots require vast amounts of diverse, grounded experience to operate robustly in messy environments, and that data is hard and expensive to collect.
• Power and form‑factor limits. Many mobile robots are constrained by battery capacity and thermal envelopes, limiting how much compute can run on the edge.
• Safety, liability and regulation. When AI acts directly in the physical world, failures can injure people or damage infrastructure, raising the bar for testing, certification and oversight.
• Workforce transitions. Physical AI displaces some repetitive roles but also creates new ones; navigating this transition responsibly is a leadership challenge, not just a technology one.

Despite these challenges, research and investment momentum suggest physical AI will be one of the defining frontiers of AI adoption over the next decade.

Ambient Intelligence: Smart Environments That Adapt Around People

From connected things to context-aware environments

Ambient intelligence (AmI) describes environments where sensing, connectivity and AI are woven into the fabric of everyday life. Rather than interacting directly with a single device, people move through spaces that are continuously:

• Sensing — via IoT devices, cameras, microphones, wearables and environmental sensors.
• Interpreting context — using machine learning at the edge or in the cloud.
• Acting and adapting — tuning lighting, temperature, flows, alerts and services without explicit commands.

In the ambient paradigm, computing “fades into the background”. Interfaces become more natural, multimodal and often invisible. Instead of repeatedly instructing systems, people experience environments that appear to anticipate needs.

Core building blocks

Most ambient intelligence deployments share a common set of enablers:

• IoT and sensor networks. Billions of connected devices capture everything from motion and occupancy to air quality, machine vibrations and biometric signals.
• Edge computing. Local gateways and embedded AI chips process data close to its source, minimizing latency and bandwidth usage while improving privacy.
• Machine learning and analytics. Models detect patterns, predict demand, classify anomalies and make context‑aware decisions.
• Actuation and orchestration. Building management systems, industrial control systems and consumer devices adjust lighting, HVAC, doors, traffic lights and more.

Illustrative use cases across smart environments

Smart homes and living spaces

In a mature ambient home, the environment constantly self‑optimizes:

• Thermostats, blinds and lighting adapt to occupants’ routines, external weather and real‑time occupancy to minimize energy use while maximizing comfort.
• Security systems distinguish familiar faces from unknown visitors and flex between “welcome mode” and “alert mode” automatically.
• Appliances coordinate with user calendars and preferences — ovens preheat when a recipe is in progress, washing machines run when energy is cheapest.
• Entertainment systems curate content based on time of day, context and inferred mood.

Voice assistants from companies such as Amazon, Google and Meta are evolving from on‑demand utilities into ambient “companions” embedded in rooms, cars and wearables, responding proactively to cues like alarms, doorbells and safety events.

Healthcare, hospitals and assisted living

Ambient intelligence is particularly transformative in health and care settings:

• Hospital rooms equipped with sensors and microphones can continuously monitor patients’ vital signs, movements and distress signals, alerting staff in real time.
• Ambient clinical intelligence platforms, such as those pioneered by Nuance (a Microsoft company), automatically draft clinical notes by listening to doctor–patient conversations, reducing administrative burden.
• In assisted living, combinations of wearables and home sensors detect falls, medication adherence lapses and changes in daily routines, triggering early interventions.

Smart factories and workplaces

Industrial and corporate environments are becoming ambiently intelligent as well:

• Smart factories embed sensors in machines and production lines to enable predictive maintenance, adaptive scheduling and real‑time quality control.
• Indoor positioning and occupancy sensing let offices dynamically reassign meeting rooms, desks and HVAC settings based on actual usage.
• Safety systems track high‑risk zones, alerting workers and shutting down equipment preemptively when patterns suggest danger.

Smart cities and public infrastructure

At urban scale, ambient intelligence promises more efficient, livable and sustainable cities:

• Adaptive traffic management systems adjust signal timing and routing based on real‑time congestion and give priority to emergency vehicles.
• Street lighting responds to pedestrians and vehicles, improving safety while reducing energy consumption.
• Environmental sensors continuously monitor air quality, noise and water systems, supporting targeted interventions and better policy decisions.
• Public safety networks fuse video, audio and other signals to detect accidents, hazards or abnormal patterns quickly, augmenting human responders.

Privacy, trust and governance in ambient systems

Ambient intelligence raises profound questions about privacy, consent and control. Always‑on sensing risks turning workplaces, homes and cities into surveillance systems if governance is weak. Emerging regulations — notably the EU’s GDPR and AI Act — are already shaping deployment models:

• Greater use of on‑device and on‑premise processing to minimize raw data leaving local environments.
• Stronger requirements for transparency: users must understand what is being sensed and why.
• Explicit approaches to data minimization, retention and anonymization.
• Protections against algorithmic bias in high‑stakes contexts like employment, finance and policing.

Businesses that design ambient systems around privacy by design and user agency will be better positioned to earn trust and scale their deployments.

Sovereign AI: Nations Racing for Digital Autonomy

What sovereign AI really means

As AI becomes a general-purpose technology underpinning economic competitiveness, security and societal narratives, governments are unwilling to depend entirely on foreign platforms. Sovereign AI captures this desire for national (or regional) autonomy over:

• AI models — large language models and other systems developed or controlled domestically.
• Data — local data assets subject to domestic law and values.
• Compute — high‑performance computing and cloud infrastructure within national or allied jurisdictions.
• Semiconductors — the chips that power AI training and inference.

The rise of U.S. foundation models like GPT‑4 and Claude, combined with China’s rapid progress in its own ecosystem (e.g., Baidu, Alibaba, Tencent, iFlytek), made many countries acutely aware of the risk of becoming permanent “AI takers” rather than “AI makers”.

National drivers of sovereign AI

Sovereign AI initiatives are typically motivated by a combination of:

• National security. Ensuring that critical defense, intelligence and infrastructure systems do not depend on foreign AI components that might be restricted or compromised.
• Cultural and legal alignment. Training models that reflect local languages, norms and regulations, including strict privacy or content rules.
• Economic competitiveness. Capturing value from domestic AI IP, startups and talent rather than simply consuming imported platforms.
• Data sovereignty. Keeping sensitive citizen and industry data within national borders and under national law.

Examples of sovereign model initiatives

Around the world, governments and national research consortia are co‑funding or directly building domestic AI models. A non‑exhaustive set of illustrative examples includes:

Building sovereign AI infrastructure and cloud

Sovereign AI is not only about models. It also requires sustained investment in:

• National AI supercomputers. Canada, the UK, EU members, Japan, India and others are investing billions in GPU‑rich clusters accessible to domestic researchers and companies.
• Regional or national clouds. Local cloud providers and specialized “sovereign cloud” offerings from global hyperscalers (for example, Microsoft Azure sovereign cloud and regional offerings from Amazon Web Services) are positioned to comply with local data residency laws.
• Policy, standards and regulation. The EU’s AI Act, national AI strategies and export control regimes all shape how and where models and compute can be used.

Different strategic postures by major powers

While almost every country now has an AI strategy, their approaches to sovereignty differ:

• United States. Focuses on maintaining technological leadership and denying adversaries access to critical chips and models. The CHIPS and Science Act incentivizes domestic manufacturing; export controls limit China’s access to advanced GPUs and fabrication equipment.
• European Union. Pursues “digital and data sovereignty”, seeking to double its share of global chip production, regulate AI risks and ensure that key digital infrastructure can be hosted and governed within the bloc.
• China. Aims for comprehensive self‑reliance in AI and semiconductors, backed by large state funds, industrial policy and data localization. Chinese authorities simultaneously exert strong ideological and information control over how AI is trained and used.
• India. Emphasizes “AI for All”, with initiatives to build public digital infrastructure, expand compute capacity and encourage domestic innovation, including semiconductor investment incentives and national AI compute projects.
• Middle East players. Sovereign wealth funds in the UAE and Saudi Arabia are investing heavily in AI and chip companies worldwide. Abu Dhabi’s G42 and the UAE’s partnership with Microsoft are emblematic of hybrid strategies combining local platforms with global alliances.

For businesses operating across borders, this emerging sovereign AI landscape will shape data strategies, vendor choices, market entry decisions and compliance obligations.

The Global Chip Ecosystem and the Geopolitics of AI Hardware

Chips as the strategic foundation of AI

Every aspect of physical AI, ambient intelligence and sovereign AI rests on one critical foundation: semiconductors. AI training requires cutting‑edge GPUs and accelerators; edge devices and robots require increasingly capable but power‑efficient chips. The global semiconductor supply chain, however, is both highly concentrated and under growing geopolitical pressure.

Where the world’s most advanced chips come from

At advanced process nodes (7nm and below), the global landscape is starkly concentrated:

• Foundry dominance. TSMC in Taiwan and Samsung Electronics in South Korea manufacture the overwhelming majority of the world’s most advanced logic chips. For leading‑edge AI accelerators from NVIDIA, AMD and others, TSMC is often the only realistic fabrication option.
• Design leadership. U.S. firms such as Intel, NVIDIA, AMD, Qualcomm and Apple dominate high‑performance CPU, GPU and mobile SoC design. Arm (UK‑origin) provides the underlying architecture for most mobile and many edge devices.
• EDA and tools. Chip design software is effectively controlled by a small group of companies, notably Synopsys, Cadence and Siemens EDA.
• Manufacturing equipment. High‑end lithography and fab tools are dominated by firms such as ASML (extreme ultraviolet lithography), Applied Materials, Lam Research and Tokyo Electron.
• Memory. DRAM and NAND are largely supplied by Samsung, SK hynix, Micron, Kioxia and Western Digital.

This concentration makes advanced chips a natural focal point for geopolitical competition and industrial policy.

Reshoring, friend‑shoring and subsidy races

In response to pandemic‑era shortages and rising geopolitical risk, governments have launched aggressive programs to reshape semiconductor supply chains:

• United States. The CHIPS and Science Act and related state‑level incentives are driving tens of billions of dollars in new fabs and packaging plants from Intel, TSMC, Samsung, Micron, Texas Instruments and others.
• European Union. The EU Chips Act aims to double Europe’s global market share by 2030, supporting projects by Intel, Infineon, NXP and joint ventures with TSMC.
• Japan. Backed by substantial government subsidies, the Rapidus consortium intends to produce cutting‑edge logic chips domestically, while TSMC is building new fabs in Kumamoto.
• China. Major national funds support domestic players including SMIC, YMTC and Huawei’s chip design arm, with an emphasis on self‑reliance under export controls.
• Middle East investors. Mubadala Investment Company (UAE) owns GlobalFoundries, while Saudi Arabia and the UAE have announced multi‑billion‑dollar AI and semiconductor investment funds.

These efforts are capital‑intensive and will take years to fully bear fruit, but they signal that chips are now viewed as instruments of sovereignty and strategic leverage, not just industrial inputs.

Implications for physical, ambient and sovereign AI

For leaders planning AI roadmaps, chip geopolitics is no longer an abstract backdrop. It directly shapes:

• Access to AI compute. Export controls may restrict cutting‑edge GPUs in certain markets, affecting the feasibility and cost of training frontier models.
• Edge device availability and pricing. The cost and availability of AI‑capable edge chips influence the pace of ambient intelligence and physical AI deployment.
• Vendor and location choices. Companies must weigh where they host AI workloads and who supplies their chips in light of evolving sanctions, alliances and subsidy regimes.
• Risk management. Single‑point dependencies on a particular geography or supplier for critical chips now represent material operational and geopolitical risks.

Strategically, semiconductors are the connective tissue between the three frontiers: they determine which nations can realistically pursue sovereign AI, how fast physical AI can scale and how pervasive ambient intelligence can become.

Smart Environments at the Edge: Where These Frontiers Converge

The convergence of physical AI, ambient intelligence and sovereign AI is most visible in smart environments: smart cities, factories, hospitals, campuses, logistics hubs and even military bases. These spaces are:

• Dense with sensors, actuators and connectivity.
• Populated by physical AI systems such as robots, drones and autonomous vehicles.
• Coordinated by edge and cloud AI platforms that interpret context and orchestrate responses.
• Increasingly subject to national or regional sovereignty requirements for data, compute and vendors.

Smart cities: AI in the urban fabric

In a smart city deployment, ambient sensing and physical AI intertwine. Consider:

• Camera networks and traffic sensors feeding into edge AI that dynamically optimize traffic flows and enforce safety rules.
• Autonomous inspection drones checking bridges, power lines and railways, sharing data with digital twins of infrastructure.
• Responsive public lighting, waste collection and environmental controls adjusting in near real time.

Municipalities must decide whose platforms orchestrate these systems. In Europe, questions of data sovereignty, cybersecurity and vendor trustworthiness steer many projects towards European or allied providers rather than untrusted foreign vendors. “Who runs the city’s AI?” is becoming a strategic infrastructure decision.

Factories and warehouses: cyber‑physical production systems

Modern factories are effectively cyber‑physical organisms. Physical AI robots move materials and assemble products; ambient sensors constantly monitor machine health and process status; edge controllers and AI models coordinate everything:

• Autonomous mobile robots navigate based on both their own sensors and facility‑wide indoor positioning systems.
• Predictive maintenance models analyze vibration and temperature data streams to schedule interventions just‑in‑time.
• Energy management systems dynamically adjust power usage based on production schedules, real‑time prices and grid conditions.

For manufacturers, the sovereignty dimension appears in decisions about industrial cloud platforms, data lake locations and the origin of AI control systems, especially when factories are designated as critical national infrastructure.

Hospitals, campuses and critical facilities

Hospitals, research campuses and airports are becoming highly instrumented, with ambient monitoring for safety, security and efficiency. Physical AI appears as delivery robots, cleaning robots and autonomous guided vehicles. These deployments often require:

• Strict local processing of sensitive data (e.g. patient health information).
• Highly resilient edge infrastructure capable of operating even during network outages.
• Integration with building management systems, security operations and emergency response procedures.

Here, strategic choices about vendors, architectures and jurisdictions determine not just functionality, but also compliance and resilience.

Defense, Security and Dual‑Use Dynamics

Defense and security establishments are among the most aggressive adopters of physical AI and ambient intelligence, precisely because they recognize the strategic leverage and risks involved.

• Autonomous systems. Militaries are investing heavily in autonomous aerial, land and maritime vehicles, from small quadcopters to large unmanned systems, for surveillance, logistics and potentially offensive operations.
• Ambient sensing on the battlefield. Networks of sensors, satellites, drones and cyber‑intelligence feeds are fused by AI to create “persistent ISR” (intelligence, surveillance and reconnaissance) and to accelerate decision‑making.
• AI‑enabled cyber offense and defense. AI is used both to identify vulnerabilities and to detect sophisticated attacks, including against critical infrastructure.

In this domain, sovereign AI is non‑negotiable: no military will rely on an adversary’s or distant ally’s black‑box models for core decision‑support. This is one reason why investments in national compute and chip capacity are often justified on security grounds and why export controls on advanced AI hardware and software have tightened.

At the same time, many physical and ambient AI technologies are dual‑use. The same drone that inspects power lines can be weaponized; the same camera network that monitors traffic can enable pervasive surveillance. This dual‑use reality increases the responsibility of technology companies and investors: due diligence, export compliance and ethical frameworks are becoming core competencies, not side concerns.

Over the coming decade, international norms around autonomous weapons, military AI and cyber‑physical security will play a critical role in shaping how far and how fast these technologies are deployed.

Sources, References and Additional Reading

The analysis in this report synthesizes insights from a broad set of public sources, including the following selected references:

• World Economic Forum — “Physical AI: The Next Frontier in Industrial Automation” (white papers and articles). https://www.weforum.org
• Gartner — “Top Strategic Technology Trends” series, including coverage of Physical AI and Ambient Intelligence. https://www.gartner.com
• European Commission — EU Chips Act and AI Act documentation (digital and data sovereignty, semiconductor strategy). https://digital-strategy.ec.europa.eu
• U.S. Government — CHIPS and Science Act summaries and Department of Commerce export control announcements on advanced semiconductors. https://www.commerce.gov
• IDTechEx — Reports on semiconductor geopolitics and advanced packaging.
• Lawfare — Analyses on Sovereign AI, national AI models and public AI infrastructure.
• Reuters — Reporting on AI in the Chinese military, export controls and global AI competition.
• McKinsey & Company, Boston Consulting Group and Bain & Company — Industry perspectives on automation, AI readiness and smart manufacturing.
• Nuance Communications — Ambient clinical intelligence case studies in healthcare. https://www.nuance.com/healthcare.html
• NVIDIA, Intel, AMD — Technical blogs and developer resources on robotics, simulation, edge AI and data center GPUs. https://developer.nvidia.com | https://www.intel.com | https://www.amd.com
• TSMC, Samsung, GlobalFoundries — Investor and technology roadmaps on advanced process nodes and capacity expansion. https://www.tsmc.com | https://www.samsung.com | https://gf.com
• Technology Innovation Institute (UAE) — Falcon LLM documentation and open‑source releases. https://www.tii.ae