E-Paper
How physical AI is reshaping the intelligent world
This article is authored by Balaji Sivakumar, senior director, technical product marketing, Sandisk.
Artificial Intelligence (AI) is no longer confined to answering questions on screens; it is steadily stepping into the physical world, transforming how machines perceive, reason and act within real environments. This transition, often described as the rise of physical AI, marks a decisive shift in the evolution of intelligent systems. It signals a movement away from passive, cloud-dependent computation towards embodied intelligence capable of interacting dynamically with the world around it.
The foundations of this transformation lie in decades of technological progress. Early robots, developed in the latter half of the 20th century, were designed for highly specific industrial tasks such as welding and material handling. These systems operated within rigid parameters, relying on centralised control and possessing minimal local storage. Their intelligence was limited, their adaptability constrained, and their ability to respond to changing environments virtually non-existent. Yet, they laid the groundwork for what was to come.
In recent years, the rapid advancement of AI has catalysed a fundamental architectural shift. Intelligence is no longer solely housed in distant data centres but is increasingly distributed to the edge, where decisions must be made in real time. This paradigm, known as edge AI, enables machines to process data locally, reducing latency and enhancing responsiveness. In environments where milliseconds can determine outcomes, such as autonomous driving or robotic navigation in warehouses, this shift is not merely beneficial but essential.
Collaborative edge AI represents the next stage of this evolution, where systems integrate local processing with cloud-based learning. Autonomous mobile robots, for instance, can now navigate complex environments, avoid obstacles, and optimise routes without constant external guidance. This ability to act independently while still contributing to a larger data ecosystem reflects a more sophisticated and resilient model of intelligence.
As this technological trajectory advances, the emergence of humanoid robots stands out as a particularly compelling frontier. Unlike their industrial predecessors, these machines are designed to operate in unstructured environments, interacting with objects and humans in ways that require nuanced understanding and continuous learning. Their intelligence is not static but evolves through experience, as they learn from each interaction—whether it is adjusting grip pressure to handle delicate objects or maintaining balance in unpredictable conditions.
This capacity for continuous learning generates vast volumes of data, placing unprecedented demands on storage infrastructure. High-performance, local storage has become a critical component of physical AI systems, enabling real-time decision-making and immediate access to relevant data.
However, the storage challenge is far from uniform. Different categories of humanoid robots require varying levels of computational power and storage capacity. Industrial robots, equipped with multiple sensors and cameras, must manage complex datasets including vision-language-action models and telemetry logs. General-purpose humanoids, designed for broader interaction, often rely on multimodal AI systems that integrate speech, vision and control, thereby increasing their storage requirements even further. In more demanding environments, such as disaster zones, the need for robust, high-capacity storage becomes even more pronounced.
The interplay between edge and cloud computing further complicates this landscape. While immediate processing occurs locally, significant volumes of data are transmitted to centralised data centres for long-term storage and model training. This dual architecture ensures both speed and scalability, but it also necessitates a rethinking of data centre design. AI workloads require faster access to larger datasets, driving the development of high-capacity, high-performance storage solutions capable of supporting complex data ecosystems.
This transformation extends beyond technology into the broader realm of governance and infrastructure. As physical AI systems become more prevalent, their integration into society will require careful consideration of ethical, regulatory and operational frameworks. The ability of machines to act autonomously in real-world environments raises questions about accountability, safety and trust, all of which must be addressed alongside technological innovation.
Yet, at its core, the evolution of physical AI is a story of convergence—between computation and cognition, between digital intelligence and physical action. It represents a reimagining of what machines can do, not as tools confined to specific tasks, but as adaptive systems capable of learning, evolving and contributing meaningfully to complex environments.
Ultimately, the success of this new era will depend not only on advances in artificial intelligence but also on the invisible infrastructure that supports it. Storage, often overlooked, emerges as a foundational pillar in this ecosystem. Without the ability to process, store and retrieve vast amounts of data efficiently, the promise of physical AI cannot be fully realised.
As intelligence continues to move off the screen and into the world, it is clear that we are not merely witnessing an incremental upgrade in technology. We are entering a phase where machines are beginning to understand and engage with reality itself, reshaping industries, redefining capabilities, and challenging our very notion of intelligence.
This article is authored by Balaji Sivakumar, senior director, technical product marketing, Sandisk.
