India’s AI Future Needs Open, Scalable, and Sovereign Infrastructure

As AI adoption accelerates across enterprises and cloud ecosystems, open software is emerging as a critical pillar for scalable and future-ready AI infrastructure. In this exclusive conversation with Vaishali Umredkar, Editor of Semiconductor For You, Manik Kapoor, Datacenter and AI Pursuit Lead at AMD India, shares insights on open AI ecosystems, hybrid cloud scalability, vendor lock-in challenges, and how software-defined infrastructure is shaping India’s AI-driven future.

As AI adoption accelerates across India, why is open software becoming central to the future of cloud infrastructure?

AI is now moving into production environments across enterprises, cloud, and edge infrastructure. In that environment, organizations need flexibility to work across different models, frameworks, and deployment environments without being tied to a single stack.

Open software helps reduce engineering complexity and gives developers access to widely adopted frameworks such as PyTorch® and TensorFlow™. It can also help organizations to scale AI more efficiently as inference workloads grow and infrastructure requirements evolve.

This is the focus behind platforms such as ROCm™, which are designed to give customers more choice, portability, and ecosystem support as they scale AI deployments.

Closed AI stacks are often seen as efficient in the short term—what risks do they create for enterprises as AI workloads evolve?

Closed AI stacks can simplify early deployments, but they may introduce long- term limitations around portability, cost, and infrastructure flexibility.

As AI workloads grow across cloud, enterprise, and edge environments, organizations benefit from the ability to adapt quickly without rebuilding around a single vendor ecosystem. Lock-in can make it harder to optimize performance, scale efficiently, or adopt newer models and frameworks over time.

AI infrastructure decisions today need to support long-term evolution, not just short term deployment speed.

How does an open AI software ecosystem help organizations avoid vendor lock-in while preserving scalability and performance?

An open AI software ecosystem gives organizations more freedom in how they deploy and scale workloads across on-premise, cloud, and edge environments.

The advantage is portability without sacrificing performance. Developers can continue using familiar frameworks such as PyTorch® and TensorFlow™ while optimizing workloads across different infrastructure environments. We are already seeing this approach scale across large cloud and enterprise deployments, including organizations such as Reliance Jio Platforms, where software portability and infrastructure flexibility become increasingly important as AI workloads expand.

Platforms such as AMD Helios extend this further by bringing together compute, networking, and open software into architectures designed to scale and evolve over time.

Many enterprises struggle to move AI from pilot projects to production. What are the biggest obstacles you see in this transition?

I think the biggest challenge is that production AI is far more complex than pilot environments.

Once organizations move beyond proof of concept, they need to manage orchestration, data pipelines, latency, governance, reliability, and infrastructure coordination across distributed environments.

Fragmented software environments add to the challenge because many teams operate across different frameworks, pipelines, and tools. In many cases, the issue is not only compute availability, but whether the full system is designed to scale effectively.

This is where solutions such as AMD Enterprise AI Suite can help by simplifying deployment through validated blueprints, optimized inference services, and more streamlined AI operations.

In practical terms, how does software act as the control plane for AI infrastructure in enterprise and cloud environments?

Software is what coordinates how AI infrastructure operates at scale.

It manages workload scheduling, data movement, and how compute resources are allocated across CPUs, GPUs, storage, and networking. Without effective orchestration, infrastructure can become inefficient and bottlenecks can start affecting performance.

This is why software layers increasingly matter in enterprise AI environments, particularly for deployment, monitoring, governance, and workload management.

Platforms such as ROCm™ and AMD Enterprise AI Suite are designed to help customers improve portability, resource utilization, and infrastructure coordination across environments.

What advantages do open software stacks offer when enterprises need to integrate AI across hybrid or multi-cloud environments?

The biggest advantage is portability.

Enterprises need the ability to move workloads across on-premise infrastructure, public cloud, and edge environments without constantly rebuilding or re-engineering applications.

Open software stacks help standardize development and deployment across environments, reducing integration complexity and operational fragmentation. They also give organizations flexibility in deciding where workloads should run based on performance, cost, latency, or compliance requirements.

Platforms such as ROCm™ and AMD Enterprise AI Suite support this through an open and modular approach designed for hybrid AI environments.

How can open AI software support India’s growing regional and sovereign cloud requirements, especially around localisation and compliance?

Organizations that use open AI software can gain greater control over how infrastructure is deployed, managed, and aligned with local compliance requirements.

That becomes increasingly important for sovereign and regional cloud environments where organizations need flexibility in data governance, localization, and infrastructure customization.

Open ecosystems also help avoid fragmentation and vendor lock-in while allowing organizations to continue evolving their infrastructure over time.

As India’s cloud ecosystem matures, this balance between compliance, control, and interoperability will become increasingly important.

Looking ahead, what should Indian enterprises prioritise today to build sustainable AI infrastructure for long-term growth?

The priority should be building infrastructure that can adapt as AI workloads evolve.

Models, deployment environments, and inference requirements are changing rapidly, so enterprises need architectures that can scale without locking organizations into rigid infrastructure choices.

The organizations scaling AI most effectively today are taking a system level approach by aligning compute, networking, and software around efficiency and interoperability.

That is also where open and modular software environments become increasingly important, particularly for deployment, inference, and long-term infrastructure management.

At AMD, our focus is enabling that through high performance computing and open software ecosystems designed to support AI infrastructure over time.