ET AIThe networks the world relies on today were built largely around people. According to the International Telecommunication Union (ITU), about 6 billion people were using the internet in 2025, and fixed and mobile broadband networks carried an estimated 8.8 zettabytes of traffic that year. In high-income countries, monthly traffic averaged about 18 GB per mobile broadband subscription and 505 GB per fixed broadband subscription.
People have only so many hours a day to spend online. AI agents and connected machines can keep generating traffic even when no one is online to prompt them. Their numbers depend on how many are deployed. IDC projects that the number of actively deployed AI agents worldwide will exceed 1 billion by 2029, or 40 times the number deployed in 2025.
If serving 6 billion people required the networks we have today, what will those networks need as they carry traffic from tens of billions of connected devices as well as AI agents? The question takes on particular importance when those systems act in the physical world.
Physical AI combines AI with machines, sensors and robotics so that systems can perceive their surroundings, make decisions and act. Applications range from machine vision to autonomous robots. In Kerala, Vizhinjam, India’s first semi-automated port, uses remotely operated ship-to-shore cranes to load and unload vessels, while automated cranes move containers in the yard. A vessel traffic management system developed with IIT Madras brings radar, ship identification, camera and weather data into one control room. It helps operators monitor ship movements and uses AI to flag navigational hazards.
Service providers see the commercial opportunity. A recent Ciena survey conducted with Censuswide covered 1,200 service provider experts across 12 countries, including 100 in India. Of the Indian respondents, 59% expect physical AI ecosystems to account for more than 15% of their total enterprise AI revenue within five years.
Much of the industrial capacity that will run physical AI in India is still to be built. That makes this the right time to plan the networks that will support it. Machine-vision cameras inspecting components, sensors tracking the condition of equipment and autonomous vehicles moving through warehouses all produce data as they work. AI models depend on that data to decide what happens next. Across the production lines, warehouses and ports that automate, these data flows will drive demand for network capacity.
Safety-critical control will largely stay within the plant, on industrial Ethernet, private wireless networks and on-site computing systems. The larger task for service providers is carrying the growing volume of data that moves between industrial sites, regional data centres and AI infrastructure for model updates, analytics, training and coordination across facilities.
The distinction here is important because the two carry different things: the control network inside the plant carries the instructions that keep machines running safely, while the wide-area links carry the data those machines generate to wherever models are trained, updated and analysed.
The first consequence is that traffic changes direction. Enterprise networks have traditionally carried data towards employees and end users, whereas physical AI sends large volumes of machine data the other way, from cameras, sensors and equipment back towards computing systems.
The second is that demand grows with the number of machines and agents rather than the number of people. Networks have long been planned around human patterns of use, with predictable daily peaks. Agentic AI adds a continuous layer of autonomous activity on top of that, and connected machines do the same. A camera on an inspection line does not log off at night, so every new robot, sensor or agent adds traffic that capacity plans have to absorb.
Indian service providers are preparing for this. In the same Ciena survey, 93% of Indian respondents agreed that next-generation optical technology, such as higher-capacity wavelengths and power-efficient coherent optics, is critical to making AI connectivity services more profitable and competitive. In addition, 88% said optical network upgrades are urgently needed, while 58% plan to generate revenue from consumption-based bandwidth services, which let capacity grow with a customer’s operations.
How much data will each site generate, and how fast will that grow? Networks should be sized for the machine population a facility will have in five years and for the share of its data that has to leave the site for training, analytics and coordination.
Can capacity be added without rebuilding? Links between industrial clusters, data centres and AI infrastructure should allow capacity to be added as more machines come online. They should also be operated with enough automation to adapt as traffic changes. In the survey, 94% of Indian respondents considered advanced network automation important or absolutely critical to capturing AI-driven opportunities.
Is the network being planned with the plant? Connectivity is far easier to get right when factories, AI infrastructure and the links between them are designed together than when networks are added to an operational site afterwards.
As manufacturing and automation expand together, networks built to serve billions of people will increasingly serve tens of billions of machines and AI agents as well. Treating those connections as industrial infrastructure, and planning them for that scale, will determine how well physical AI works once deployments move from individual use cases to whole industrial environments.
The author is Vice President, Solutions Engineering and Chief Technology Officer, Asia Pacific, India and Japan, Ciena.
People have only so many hours a day to spend online. AI agents and connected machines can keep generating traffic even when no one is online to prompt them. Their numbers depend on how many are deployed. IDC projects that the number of actively deployed AI agents worldwide will exceed 1 billion by 2029, or 40 times the number deployed in 2025.
If serving 6 billion people required the networks we have today, what will those networks need as they carry traffic from tens of billions of connected devices as well as AI agents? The question takes on particular importance when those systems act in the physical world.
Physical AI combines AI with machines, sensors and robotics so that systems can perceive their surroundings, make decisions and act. Applications range from machine vision to autonomous robots. In Kerala, Vizhinjam, India’s first semi-automated port, uses remotely operated ship-to-shore cranes to load and unload vessels, while automated cranes move containers in the yard. A vessel traffic management system developed with IIT Madras brings radar, ship identification, camera and weather data into one control room. It helps operators monitor ship movements and uses AI to flag navigational hazards.
Service providers see the commercial opportunity. A recent Ciena survey conducted with Censuswide covered 1,200 service provider experts across 12 countries, including 100 in India. Of the Indian respondents, 59% expect physical AI ecosystems to account for more than 15% of their total enterprise AI revenue within five years.
The industrial base is still being built
India’s operational stock of industrial robots stood at 52,570 units in 2024, placing it tenth worldwide, according to the International Federation of Robotics. Policy is pushing that number up. The National Manufacturing Mission, announced in the Union Budget 2025-26, includes the availability of technology among its five focal areas. NITI Aayog’s roadmap for advanced manufacturing identifies artificial intelligence, robotics and digital twins as high-impact enablers across 13 priority sectors, with the aim of manufacturing contributing more than 25% of GDP by 2035.Much of the industrial capacity that will run physical AI in India is still to be built. That makes this the right time to plan the networks that will support it. Machine-vision cameras inspecting components, sensors tracking the condition of equipment and autonomous vehicles moving through warehouses all produce data as they work. AI models depend on that data to decide what happens next. Across the production lines, warehouses and ports that automate, these data flows will drive demand for network capacity.
When AI acts, the network becomes part of the industrial system
For a user waiting a little longer for an AI-generated answer, the delay is an inconvenience. When an AI-enabled production line or port terminal loses the data it depends on, operations can stop. Once AI is operating machinery, the connection carrying its data becomes part of the industrial system. It has to be as reliable as the machinery itself.Safety-critical control will largely stay within the plant, on industrial Ethernet, private wireless networks and on-site computing systems. The larger task for service providers is carrying the growing volume of data that moves between industrial sites, regional data centres and AI infrastructure for model updates, analytics, training and coordination across facilities.
The distinction here is important because the two carry different things: the control network inside the plant carries the instructions that keep machines running safely, while the wide-area links carry the data those machines generate to wherever models are trained, updated and analysed.
The first consequence is that traffic changes direction. Enterprise networks have traditionally carried data towards employees and end users, whereas physical AI sends large volumes of machine data the other way, from cameras, sensors and equipment back towards computing systems.
The second is that demand grows with the number of machines and agents rather than the number of people. Networks have long been planned around human patterns of use, with predictable daily peaks. Agentic AI adds a continuous layer of autonomous activity on top of that, and connected machines do the same. A camera on an inspection line does not log off at night, so every new robot, sensor or agent adds traffic that capacity plans have to absorb.
Indian service providers are preparing for this. In the same Ciena survey, 93% of Indian respondents agreed that next-generation optical technology, such as higher-capacity wavelengths and power-efficient coherent optics, is critical to making AI connectivity services more profitable and competitive. In addition, 88% said optical network upgrades are urgently needed, while 58% plan to generate revenue from consumption-based bandwidth services, which let capacity grow with a customer’s operations.
Planning the network alongside the machines
For service providers and industrial enterprises, the practical work comes down to three questions.How much data will each site generate, and how fast will that grow? Networks should be sized for the machine population a facility will have in five years and for the share of its data that has to leave the site for training, analytics and coordination.
Can capacity be added without rebuilding? Links between industrial clusters, data centres and AI infrastructure should allow capacity to be added as more machines come online. They should also be operated with enough automation to adapt as traffic changes. In the survey, 94% of Indian respondents considered advanced network automation important or absolutely critical to capturing AI-driven opportunities.
Is the network being planned with the plant? Connectivity is far easier to get right when factories, AI infrastructure and the links between them are designed together than when networks are added to an operational site afterwards.
As manufacturing and automation expand together, networks built to serve billions of people will increasingly serve tens of billions of machines and AI agents as well. Treating those connections as industrial infrastructure, and planning them for that scale, will determine how well physical AI works once deployments move from individual use cases to whole industrial environments.
The author is Vice President, Solutions Engineering and Chief Technology Officer, Asia Pacific, India and Japan, Ciena.
Add 
as a Reliable and Trusted News Source
as a Reliable and Trusted News Source
Add Now!
(Disclaimer: The opinions expressed in this column are that of the writer. The facts and opinions expressed here do not reflect the views of www.economictimes.com.)
ETPrime Membership
