Opinion
The next AI revolution will not be decided in the cloud, but at the edge
"In the coming decade, competitive advantage will not be determined solely by the quality of the model, but by the ability to operate it in the real world, in real time, and at scale," writes Yaniv Sulkes, Head of the Physical AI Division at Hailo.
In recent years, artificial intelligence has been advancing at a staggering pace. Models are growing larger, generative AI capabilities are breaking new ground, and massive investments are being made in cloud infrastructure. The race appears to be focused primarily on who will build the largest and most powerful model, and who will control the computational infrastructure behind it.
But beneath the surface, a deeper shift is taking place, one that is not only about performance, but about where decisions are made. The transition to Physical AI, artificial intelligence that operates in the physical world, is changing not only the technology itself, but also the economic rules of the game, and the identity of its winners.
Until now, most of the value of AI has been created in the digital world. Even as models became more sophisticated, they produced text, images, or code. However, as systems move beyond the screen into the physical world, into robots, drones, industrial systems, autonomous vehicles, and medical systems, the requirements change dramatically, and the gap between what works in the cloud and what works in the field continues to widen.
In the physical world, there is no luxury of time. Systems must operate in a continuous loop of sensing, processing, and response, in real time and under changing conditions. A delay of even a fraction of a second is no longer a matter of user experience, but of safety, reliability, and cost. In this reality, a model that waits for the cloud to make decisions is not only slow, it is simply not relevant, and at times even dangerous.
This is where a gap emerges that many still overlook. Companies that continue to build cloud-dependent AI solutions risk finding that their products are not suited for the next generation of applications. In a world where value is created at the moment of decision, the advantage shifts to those who can bring intelligence to the edge, to the device itself, and operate it in a reliable, continuous, and efficient manner.
The cloud, of course, is not disappearing. It will remain critical for model training, data management, and large-scale deployment of updates. But the center of value is shifting. The decision is no longer made in the data center, but in the field, in the factory, on the street, in the hospital, or on the production line. Whoever controls the point of decision controls the value.
This shift is already beginning to reshape the map of players. Alongside cloud giants, a new layer of chip companies, computing infrastructure providers, and edge platforms is gaining strength. These are companies that bridge software and hardware, enabling advanced AI to run on small devices, with low power consumption and at a cost that supports broad deployment. In many ways, this marks a transition from a software-only world to a world of complete systems.
At the same time, the field of robotics is also sobering up from the hype. Despite the popular image of humanoid robots capable of performing any task, the economic reality points in a different direction. The challenge is no longer purely algorithmic, but physical and operational, precision, flexibility, reliability, and cost. As a result, in the foreseeable future, the market will be filled with specialized systems, each excelling at a specific task.
A robot that cleans, a drone that performs a defined mission, a system that manages a specific industrial process, not because they are less advanced, but because this is the model that works. Higher reliability, lower risk, and a cost structure that enables truly broad deployment. This is the difference between an impressive demo and a product that creates value.
And this is precisely the heart of the matter, scale in the world of Physical AI is not measured by complexity, but by deployment. Not by a single robot that can do everything, but by millions of systems that perform defined tasks well. In industry, healthcare, retail, and security, this is where the most significant economic value of the coming decade will be created.
To enable this scale, a new way of thinking at the infrastructure level is also required. Not only better models, but architectures designed for edge deployment, energy efficiency, hardware integration, and the ability to manage fleets of millions of smart devices over time. This is no longer just a question of algorithms, but of system engineering at scale.
The implication is clear, technology companies, investors, and decision-makers who do not adapt their strategies to a world of edge AI will be left behind. This means investing in edge capabilities, in deep integration between hardware and software, and in understanding real physical environments, not just data.
In the coming decade, competitive advantage will not be determined solely by the quality of the model, but by the ability to operate it in the real world, in real time, and at scale. Those who continue to think in cloud-only terms will quickly discover that it is no longer enough.
The next battle of AI has already begun. And it will not be decided in the cloud, but at the edge.
Yaniv Sulkes is Head of the Physical AI Division at Hailo.
