Meenakshi Benjwal, Global Industry Marketing Head for TMT Industries at HCLTech, spoke with Kanika Atri, Senior Director at NVIDIA, about how telecom networks are evolving into AI-native platforms that power industries, enterprises and consumers. Their discussion highlighted a fundamental shift: networks are no longer simply moving data. They are becoming platforms for generating and distributing intelligence.

AI embedded across the telecom stack

AI is no longer being layered on top of telecom infrastructure. Instead, it is being integrated into every layer of the network architecture.

Atri described how telecom networks are evolving into fully AI-native systems, from infrastructure to applications. “In telecom infrastructure, AI is redefining every layer of the stack,” she explained. “It starts with the chips, infrastructure, models and applications.” At the infrastructure level, accelerated computing is transforming the underlying architecture. “The underlying substrate is becoming AI native using accelerated computing,” said Atri. “Telecom networks [are] becoming a machine for generating tokens, not just bits.” This transformation extends into network operations as well. AI models trained specifically on telecom data are enabling new forms of automation and intelligence. These large telecom models understand the language of logs, architecture diagrams and network operations.  “Agents are helping improve everyday operations and customer experience,” she added. 

From connectivity provider to AI service platform

The impact of AI extends beyond network operations. Telecom operators are increasingly positioning themselves as AI service providers for enterprises.

At the top layer of the stack, operators are enabling companies to build and deploy AI applications using their own data. “More and more telcos are now becoming AI service providers,” said Atri. “They’re helping enterprises embrace AI, adopt AI, bring in their own applications, their own data, train their own data, and turn it into real ROI for those enterprises.” The results are already becoming measurable. According to NVIDIA’s recent research on AI adoption in telecom, organizations are already seeing financial benefits. The report found that 90% of respondents said AI is already helping them increase revenues and decrease costs. AI has become much more than a buzzword. It’s integrated throughout the stack and delivering business outcomes. 

Telecom networks as the “great equalizer” for AI

As AI becomes more powerful, distributing that intelligence becomes just as important as creating it.

According to Atri, telecom networks are uniquely positioned to play that role. “First, you create AI, you create all these models, then you need to distribute and scale AI and make it accessible,” she explained. “Networks are in the middle. They are the great equalizer.” Networks connect AI infrastructure to end users — whether they are individuals, enterprises or intelligent machines.  In the future, those users will increasingly include autonomous systems such as sensors, robots and connected devices. “At the consumption layer, it’s users like you and me and enterprises, and in the future a lot of autonomous objects like robots and sensors,” said Atri.  Networks are what connect the creation of AI to its consumption. 

Real-world applications emerging at the edge

Many of the most compelling AI applications rely on real-time intelligence delivered close to the user, often at the network edge.

Atri shared several examples of how distributed AI infrastructure is already transforming industries and small businesses. For small businesses, voice AI agents are becoming digital assistants that can handle customer interactions autonomously. “AI agents are becoming the personal concierge for small businesses,” she explained. They can handle reservations, answer questions and run conversations in real time: “When a user is calling that agent to make a reservation, this agent is fully trained in the data of that particular restaurant.” As many small businesses cannot run their own AI infrastructure, telecom networks play a critical role. In other industries, AI-powered connectivity is enabling new capabilities across healthcare, agriculture and remote industrial environments. Vision language models can run analytics on the edge, and connected cameras and sensors can now deliver intelligence across remote locations.  “These are hard to reach places, and they are already ubiquitously connected through networks,” she added. “And that’s where now we bring an intelligence to that edge.” Healthcare is another emerging area. In Indonesia, for example, telecom operators are using AI to support preventative health services. “Every time somebody is turning 40, there is an automated AI agent that calls and asks, have you done these tests?” she said. 

From Agentic AI to Physical AI

Looking ahead, Atri sees rapid growth in autonomous AI agents and the emergence of Physical AI.

“I am thrilled to see how fast Agentic AI adoption has happened,” she said. Compared with previous years, the pace of innovation across telecom networks has accelerated significantly. “I’m seeing such a huge delta from Barcelona last year versus this year [when it comes to Agentic AI in action,” she continued. But the next phase will extend AI beyond software systems into the physical world. “We believe Physical AI is going to be really huge,” said Atri. “Robots and drones and sensors.” These systems will operate across highly connected environments. “These networks will be operating at machine scale,” she added. 

Building AI-native network architecture

Supporting this transformation requires telecom networks to become fully software-defined and AI-native.

Historically, telecom infrastructure evolved slowly, with major technology upgrades occurring roughly once per decade. Until now, telecom network stacks have largely been vertically integrated, with each new generation of mobile technology typically taking close to a decade to introduce major new capabilities. That model is beginning to shift. Telecom networks are moving toward software-defined architectures that separate hardware from software innovation. This transition allows operators to innovate much faster, introducing new capabilities at the pace of software development. As a result, the foundational infrastructure of future networks is expected to become fully software-defined and AI-native. 

Trust, sovereignty and inclusive intelligence

As AI adoption accelerates, questions around trust, sovereignty and governance are becoming increasingly important.

Atri emphasized that countries and enterprises must maintain control over their own AI systems and data. “Every country, every enterprise needs to build their own intelligence,” she said. That means adapting foundation models using local data and context so they reflect the language, culture and regulatory environment of each market. Telecom networks are well positioned to support this approach because they already operate within national infrastructure frameworks and are widely trusted institutions. At the same time, transparency and explainability must be built into AI systems from the start. Organizations need to understand how AI systems reach decisions and what data and safeguards shape their outputs. Ultimately, Atri believes telecom networks will play a critical role in making AI accessible to everyone. “Bringing intelligence into the hands of billions and making sure that that AI diffusion reaches that last person, no one left behind,” she said.