Physical AI: Moving from cloud-centric AI to intelligence where work happens

AI is entering its most consequential phase. Beyond generating text or analyzing patterns, it’s now moving, sensing and acting in the physical world — what the industry calls physical AI.

By 2030, there could be as many as 39 billion IoT devices in operation, all generating data at the edge — on factory floors, in mining operations and across warehouse networks. Yet, most AI strategies are still anchored in centralized cloud architectures that weren’t designed for a distributed, physical world.

The transformative impact of physical AI won't be seen in data centers that generate tokens and process text prompts. It will unfold where work happens — where systems can’t afford latency, downtime or blind spots.

The physical AI imperative: Moving beyond the cloud

In a data center, an error affects a line of code; in a mine or a factory, it has physical consequences. Traditional AI architectures were designed for centralized intelligence, but in the physical world, the “round trip to the cloud” model fails for three critical reasons:

1. Safety and latency: In a high-speed warehouse, a decision delayed by 500 milliseconds becomes a safety hazard. Decisions must happen in real time, with consequences measured in both safety and productivity.
2. Environmental complexity: Unlike controlled digital environments, physical systems operate in messy, unpredictable conditions — variable lighting, changing layouts, human interaction and equipment degradation. This means AI must be resilient enough to adapt continuously, not just perform well under the conditions it was originally trained for.
3. Data sovereignty: Operational environments generate vast amounts of sensor data — video feeds, vibration signatures and thermal readings, for example. This sensitive data is often too proprietary, too regulated or too large to transmit to the cloud.

Drivers for the adoption of physical AI

As a result, organizations are rethinking how and where intelligence should live. Several factors are accelerating the adoption of physical AI, including:

• Real-time decision-making: In physical environments, delay is risk. Industries need AI systems that can sense, decide and act instantly — whether in a warehouse aisle, along a hospital corridor or on a factory floor.
• Less dependence on network connectivity: Continuously transmitting high-volume sensor data, video streams and telemetry to the cloud places significant demands on network infrastructure. By processing data locally, edge architectures reduce bandwidth requirements, lower operating costs and ensure critical operations can continue even when connectivity is limited or disrupted.
• Hardware and sensor maturity: Advances in sensors, motors, connectivity and edge computing have moved physical AI from theory to deployment. Machines can now perceive and interact with the real world with the precision and endurance that complex operations demand.
• Safety and risk reduction: AI can replace humans in hazardous zones, reducing injuries and operational downtime.
• Sovereignty and compliance: Processing sensitive data at the edge allows organizations to retain control over where their data resides and how it is governed.

Amid the fundamental shift from reactive automation to adaptive intelligence, physical AI demands a different architecture than cloud-native AI. It succeeds only when intelligence is embedded directly into operations, not layered on top of them.

Why physical AI deserves attention now

Physical AI isn’t emerging in a vacuum. It’s the result of converging pressures that are changing enterprise operations.

Labor shortages and rising costs are forcing a rethink of automation, because static systems that simply execute are no longer enough. Organizations also need intelligence that adapts in real time, and the underlying technology has reached maturity: Edge computing, advanced connectivity such as 5G and 6G and increasingly effective AI models are moving physical AI from concept to deployment.

Already in early 2026, IDC* estimated that “global spending on edge computing solutions accounted for $265 billion in 2025 and is projected to grow at a compound annual growth rate (CAGR) of 12.5%, reaching nearly $450 billion by 2029.”

And while edge computing is the essential utility, a projected 33.3% CAGR for physical AI points to the additional enterprise capex and infrastructure transformation that’s underway.

Another force at play is competitive necessity. Early movers are already seeing measurable gains in safety, efficiency and operational agility. In this dynamic landscape, delays just widen the gap.

• ALSO READ → Digital infrastructure meets human design: How physical AI transforms cities

How NTT DATA operationalizes physical intelligence

In July 2024, NTT DATA introduced the industry’s first fully managed edge AI platform, specifically designed to meet the rigorous demands of physical AI. Rather than managing physical assets as “data sources,” we treat them as intelligent, software-defined systems embedded directly into operations.

The operational impact is measurable:

• Manufacturing: Machines detect microvibrations and thermal signals, autonomously adjusting their parameters to prevent failure and reduce unplanned downtime.
• Energy optimization: Systems dynamically balance consumption against live demand and operational load without compromising performance.
• Autonomous inspection: Quadruped robots operate in hazardous areas, identifying structural or thermal anomalies invisible to the human eye and reducing human exposure to risk.

Physical AI in action

Making the shop floor more productive

NTT DATA delivered a shop-floor task-verification solution for a leading discrete manufacturer. The solution uses real-time, on-premises analysis of live production processes to verify adherence to standard operating procedures (SOPs) and maintain product quality.

By monitoring tasks as they are performed, the edge-based system helps ensure consistent execution and generates alerts whenever processes deviate from established standards.

More visibility and resilience — ready for AI

For a global battery manufacturer, we designed a scalable industrial architecture capable of integrating equipment from multiple original equipment manufacturers, supporting high-speed production environments and meeting stringent cybersecurity and regulatory requirements. Our solution provided:

• Unified real-time visibility and centralized monitoring of all production operations.
• Robust, AI-ready data pipelines to improve overall equipment effectiveness, enable predictive maintenance and enable the expansion of physical AI initiatives.
• High resilience and data integrity through a disaster recovery architecture designed for redundancy and store-and-forward buffering, eliminating single points of failure.

Intelligence where it matters

The future of AI is no longer limited to dashboards or analytics on a screen. It is about embedding intelligence directly into the machines, facilities and supply chains that keep the world moving.

For organizations in manufacturing, mining, logistics and energy, the benefits are clear: Real-time decision-making without network dependency, operational data sovereignty, regulatory compliance and streamlined integration of legacy OT with modern IT systems.

At NTT DATA, we have the platforms, the partnerships and the expertise to take AI out of the cloud and put it to work on the floor. The infrastructure exists and the intelligence is ready.

We’re past the point of asking if physical AI will reshape your operations. It will — so how quickly can you lead the transition? Is your infrastructure ready?

• ALSO READ → Factory floor 2.0: How AI-ready networks are reshaping manufacturing

* IDC Press Release. Edge Computing Global Spending to Grow at 15%, Reaching $450 Billion by 2029. 25 February 2026.