Cisco’s Arun Shetty on why India’s AI moment hinges on infrastructure, not intent

As Indian enterprises move decisively from AI experimentation to real-world deployment, a clear pattern is emerging: ambition is outpacing readiness. While nearly half of organisations already have multiple AI use cases in production, scaling them reliably remains a structural challenge. In this conversation, Arun Shetty, Chief Technology Officer (CTO) & Senior Director, Solutions Engineering, Cisco India & South Asia, breaks down where companies are getting stuck—from infrastructure and networking gaps to security blind spots—and what it will take to build AI systems that are truly enterprise-grade. Edited excerpts.

There’s a shift from AI pilots to real deployments—how would you assess where Indian enterprises stand today on this journey from experimentation to scale?

The progress is real, but so is the gap. Nearly half of Indian enterprises now have multiple AI use cases in production, yet most are struggling to scale them. Starting is not the hard part—scaling is. That’s because organisations are dealing with three simultaneous pressures: rising traffic and risk as agents become part of the workforce, growing operational complexity across hybrid environments, and rising expectations from customers and employees.

There’s also a fundamental shift underway—from AI that answers questions to AI that acts. That changes what production readiness looks like. Networks designed for predictable, human-driven traffic aren’t built for always-on, agent-driven workloads. The difference between organisations that scale and those stuck in pilot mode comes down to whether they built the right foundation early.

What are the biggest infrastructure gaps you’re seeing in India as companies move toward production-grade AI use cases?

The gap is three-layered—power, compute, and networking—and these challenges compound each other. India is expected to need around 8GW of data centre capacity by 2030, with tens of billions in investment already anticipated. But capacity alone isn’t enough; architecture matters.

A second major gap is data infrastructure. Pilots run on clean datasets, but production AI must deal with fragmented, real-world data across systems. Without the ability to distil and correlate this data, outputs become unreliable.

The third—and often underestimated—gap is security and safety. AI introduces both model-level risks like bias and hallucinations, and external threats from attackers. Add to that the rise of “shadow AI,” and many organisations simply lack visibility into what’s being used internally. Security and safety need to be designed in, not added later.

What does “AI-ready infrastructure” mean in the Indian context—especially for enterprises balancing cost, scale, and performance?

It starts with clarity on use cases—what you’re running and where. Many organisations are now bringing workloads back on-premises due to data sovereignty concerns. Without that clarity, it’s impossible to size compute, GPUs, or network capacity correctly.

AI-ready infrastructure is ultimately about an integrated stack where security is embedded at every layer—from silicon to applications. It’s also about scalability. Enterprises need to scale within racks, across data centres, and even geographically as power constraints emerge. AI workloads are continuous, and infrastructure that isn’t designed for that will quickly show strain.

AI is significantly increasing data flows. How is this reshaping enterprise networking architectures in India, and where are companies underinvesting?

AI has fundamentally changed network behaviour. Agent-driven workloads create constant, 24/7 traffic, and data volumes are exploding. Traditional networks were built for “north-south” traffic—users accessing applications. AI shifts this to “east-west” traffic between servers, clusters, and processing environments.

This requires a complete rethink of architecture. Networks are no longer just about connectivity—they’re the integration layer linking users, data, and AI systems. Yet companies are underinvesting in intelligent capabilities like segmentation, visibility, and autonomous control. Without these, managing AI-driven complexity at scale becomes nearly impossible.

As secure networking becomes critical, what are the top security risks Indian enterprises are overlooking as they scale AI?

Most organisations treat AI security as a single problem, but it’s actually three. First, protecting the world from agents—because autonomous systems can act at machine speed and cause real damage if compromised. Second, protecting agents from external threats, including manipulation of model behaviour. Third, responding to threats at machine speed, which requires automation.
The biggest underlying issue is visibility. You cannot secure what you cannot see, and many organisations are deploying AI with significant blind spots.

With only a small proportion of organisations having mature AI governance frameworks in India, how should companies approach trust, safety, and regulatory readiness?

Governance has to start with separating security and safety. Security protects systems from external threats, while safety deals with how models behave—issues like hallucinations or unintended outputs. Both require different approaches. The real gap isn’t awareness but execution. Governance must be embedded into infrastructure—through Zero Trust, continuous monitoring, and architectures designed for future regulations. Treating governance as a compliance exercise will only lead to repeated rework as rules evolve.

What are the most common mistakes Indian enterprises make when building “secure-by-design” AI infrastructure?

The biggest mistake is sequencing—prioritising performance and scalability first, and treating security as an afterthought. By then, security becomes a source of friction rather than an enabler. Another issue is failing to treat AI agents as a new attack surface. They need identity, ownership, and strict permissions, just like employees. There’s also a growing gap between tools and talent. Many organisations deploy security tools without the expertise to validate or operate them effectively. Finally, tool sprawl is a real problem. More solutions don’t equal better security—especially without the skills to manage them. What’s needed is simplification through unified platforms.

Which sectors in India are leading in building AI-ready infrastructure, and where do you see the next wave of demand emerging?

BFSI and telecom are leading the way, driven by use cases like fraud detection, customer engagement, and real-time operations. Global capability centres are also investing heavily.
The next wave will be shaped by edge computing. By 2027, most enterprise data will be created and processed outside traditional data centres. Manufacturing is already seeing this shift through IT-OT convergence, with healthcare, retail, and energy close behind. What ties it all together is the scale of the opportunity. AI could contribute $1.7 trillion to India’s economy by 2035. The organisations that invest in the right foundations today will be the ones defining that future.