Oracle’s Palanivel Saravanan On Building AI-Ready Infrastructure For Startups to Scale

Over the past decade, India’s cloud ecosystem has moved from being an enterprise-led, cost-optimisation layer to becoming a foundational driver of how digital products are built and scaled.

What began as a shift away from on-premise infrastructure has evolved into a far deeper architectural transformation, shaped by the rise of digital public infrastructure such as Aadhaar and the Unified Payments Interface (UPI). With platforms like UPI now processing over 19 Bn transactions a month, the demand for infrastructure has fundamentally changed, pushing companies towards stateless, distributed, and highly resilient systems.

At the same time, cloud is no longer a backend decision. It is increasingly tied to product experience itself, with latency, cost per transaction and scalability becoming critical variables from day one. This shift is playing out even more sharply as AI adoption accelerates across sectors, leading to an increasing demand for high-performance infrastructure and data proximity.

India’s public cloud market, estimated to cross $30 Bn by 2029, also sees a more nuanced evolution. While the early adoption was driven by cost arbitrage and agility, enterprises and startups alike are now focussing on predictability of spend, security-first architecture, and the ability to operate at massive scale without repeated re-architecture.

This maturity is also bringing trade-offs. Multi-cloud strategies, once seen as a default hedge against vendor lock-in, are increasingly being reassessed because of their operational complexity. Regulatory expectations around data sovereignty and compliance, particularly from authorities such as the Reserve Bank of India (RBI), are forcing companies to rethink how and where data is stored and processed.

Few leaders have had a front-row seat to this shift as closely as Palanivel Saravanan, vice-president of cloud engineering at Oracle India. Having spent over two decades in the industry, he has seen India transition from monolithic systems to distributed architectures and now towards AI-first infrastructure.

In a conversation with Inc42, Saravanan spoke about the inflexion point that flagged off India’s cloud journey. He shared his views on what keeps early architecture decisions as critical as product-market fit, and how startups should think about cost, security and scalability in an AI-driven world. 

He also talked of Oracle’s approach to multi-cloud, its positioning in AI infrastructure, and the lessons from working on India’s DPI at a national scale.

Here are the edited excerpts from the interview…

Inc42: You’ve seen India’s cloud adoption story unfold over the last two decades. What has been the single biggest inflexion point in this journey, and how has it changed the way engineering leaders think about infrastructure from day one?

Palanivel Saravanan: I see the rise of digital public infrastructure, particularly systems like Aadhaar and UPI, as the inflection point.

These platforms fundamentally changed both the scale and expectations of technology in India. Today, with NPCI processing close to 18 Bn transactions a month on UPI alone, the underlying infrastructure requirements are vastly different from what enterprises dealt with a decade ago.

We have moved from monolithic architectures to completely stateless, distributed and resilient architectures. This means that if you want to build systems that can scale to 18 Bn transactions, or support issuing Aadhaar to 400-600 Mn citizens, you need a very different architecture from what we had earlier. That shift in mindset is very important.

Today, tech founders think differently. Even if they have only 10,000 or 20,000 users, they design infrastructure assuming it must scale to millions of workloads.

Aadhaar and UPI made us more of a consumption-based ecosystem. Then, we moved towards scale-oriented architecture. And finally, we started using the infrastructure to build systems that are resilient, stateless and distributed to handle this level of workload.

Inc42: For a non-technical founder, why is a cloud decision made in the first three months just as critical as product-market fit?

Palanivel Saravanan: In today’s environment, the decision comes down to three things: cost per transaction, latency and the ability to scale or pivot architecture. What this means is that cloud is no longer a backend decision. It is part of the core product experience. If you don’t address these three things, whether in UPI-like systems or otherwise, it becomes very difficult to succeed across industries.

A clear example is what happened during COVID. Many companies had adopted public or private cloud in the early 2010s and 2020s. But when the pandemic hit, they were forced to re-architect their systems. Many organisations, from startups to large banks, had to rework their architecture. That came at a high cost.

The point is simple: starting with a small architecture and trying to fix it later does not work. It is far more effective to design for scale from the beginning, taking into account latency, cost per transaction and the ability to scale or pivot.

Inc42: Treating cloud architecture as a second-order problem often leads to technical debt. What does this cost a founder in real terms, both in capital and engineering bandwidth? Also, how can they optimise these costs early on?

Palanivel Saravanan: If the cloud is treated as a second-order problem, you risk hitting a wall exactly when you should be hitting hypergrowth. From day one, you must account for regulatory compliance, such as RBI guidelines, and ensure your architecture is open enough to avoid vendor lock-in.

In reality, most founders prioritise features that drive valuation. If launching a new feature moves the needle, they’ll choose that over re-engineering. But without the right foundation, you end up applying temporary patches. When you finally hit scale, those patches accumulate, and the system can no longer hold. That is when latent architectural issues emerge, requiring extensive intervention across foundational systems. 

To optimise this without draining early capital, founders must focus on flattening their throughput economics. Many startups move across cloud providers to maximise free credits, but the real goal should be converting ‘unknown unknowns’ into ‘known knowns. For example, egress costs are often unpredictable. 

As a founder, you should choose a cloud partner that offers predictable egress and storage patterns so you can forecast costs with reasonable accuracy. By choosing a partner with a stable cost curve, you avoid the massive spikes that happen when you exceed initial commitments, ensuring your engineering bandwidth remains focused on innovation rather than fixing preventable architectural collapses.

Inc42: For a startup or a mid-market enterprise evaluating cloud providers in 2026, what is the honest case for Oracle Cloud Infrastructure over providers like Amazon Web Services or Microsoft Azure?

Palanivel Saravanan: Let me give a slightly different context. In my opinion, every cloud provider is doing its job well. Each is suited to specific use cases and requirements. However, OCI is a little different in a few key ways. We position it as a Generation 2 cloud for the following reasons:

First is price-performance. OCI offers flexible shapes. Instead of fixed instance sizes, you can configure compute and memory based on what your application actually needs. If your application requires uneven core and memory usage, you can customise it and pay only for what you use.

On top of this, there is a dynamic resource manager. For example, if you provision 100 compute cores but utilisation is below 20%, the system can scale it down to, say, 30 cores. This improves efficiency and reduces cost. When demand increases, it scales back up. So you get both higher efficiency and lower cost.

Second is security. OCI follows a security-first design. By default, everything is blocked. All access is denied unless explicitly allowed. Encryption is on by default, and traffic cannot enter without encryption. Customers can use OCI-managed keys, bring their own keys, or integrate external cryptography systems.

Third is enterprise-grade SLAs. Most cloud providers offer availability SLAs. OCI goes further by also offering performance and manageability SLAs. This means customers have assurance not just on uptime, but also on performance. If performance commitments are not met, credits are provided. More importantly, it ensures the platform is built to consistently deliver those performance levels.

Fourth is cost predictability and total cost of ownership (TCO). It is not just about being cheaper. Because of a more predictable, flatter cost curve, the overall TCO tends to be more favourable than that of other hyperscalers.

Inc42: Multi-cloud and hybrid cloud have been industry buzzwords for years. Do you believe multi-cloud is genuinely the right architecture for most enterprises?

Palanivel Saravanan: I think the industry has been asking the wrong question. A better question is whether a cloud provider should force customers to choose only one cloud.

Our approach is different. We are not asking customers to move entirely to OCI. If you are using Amazon Web Services, Google Cloud Platform or Microsoft Azure, you can continue to do so and extend your capabilities with OCI.

Instead of splitting a single workload across multiple clouds, which adds significant complexity, we bring OCI services into the environments customers already use. We have built direct interconnects between OCI and other hyperscalers. These are jointly operated and provide very low latency, often around a millisecond, along with high availability. From a workload perspective, it can function almost like a single cloud environment.

In addition, we have brought key Oracle services, especially database and related tooling, into these cloud ecosystems. This allows customers to use OCI capabilities without rebuilding everything from scratch.

So the goal is not to replace multi-cloud, but to simplify it. Reduce complexity, preserve flexibility and let customers choose how they want to evolve their architecture.

Inc42: AI and ML are now central to how products are being built. What is it about Oracle Cloud Infrastructure that makes it particularly suited for AI and ML workloads? And what is the value proposition for Indian startups choosing OCI?

Palanivel Saravanan: The core philosophy at OCI is simple: bring AI to the data, not the other way round. Oracle systems sit at the centre of critical data across industries. Banks, manufacturing companies, pharma firms and many other enterprises run their core systems on Oracle databases. This means the most valuable data already resides within Oracle environments.

For AI to be effective, accuracy depends on access to the right data. For example, if a bank wants to build an AI model, it needs to train it on its own customer and transaction data. Moving that data elsewhere creates friction, latency and risk. OCI addresses this by enabling AI directly where the data exists. There are three key elements to this approach.

First, dedicated AI clusters. OCI provides single-tenant GPU clusters that give customers full control. They can run any mode and train them on their own data. These clusters can be deployed in the public cloud, in a customer’s data centre or in sovereign environments. Second, AI-native databases. Oracle has embedded AI capabilities directly into its databases, including vector support. This allows users to interact with data using natural language instead of traditional SQL queries. 

Third, high-performance AI infrastructure. OCI has built its AI infrastructure with a strong focus on networking. Using RDMA over Converged Ethernet, it enables high-speed communication between GPUs. This allows customers to train models faster and operate at scale.

Inc42: The infrastructure is critical. Any lapse in security can have serious consequences. With AI expanding the attack surface, how should Indian startups think about it, and what’s Oracle’s solution for it? 

Palanivel Saravanan: This is a very important question, especially at this point in time. Attackers today do not manually target specific organisations. They use automated tools to continuously scan the internet for misconfigurations, whether it is an exposed storage bucket or an open interface, and exploit them quickly.

Because of this, security is no longer something you add on top of infrastructure. It has to be built into the infrastructure itself as a core property. There are a few key principles here.

First is a zero-trust architecture. By default, nothing is trusted. Every access is denied unless explicitly allowed. Network communication is tightly controlled, often limited to one-to-one connections, so only specific services can communicate with each other.

Second is encryption by default. All data and traffic are encrypted, and this cannot be turned off. This ensures that even if data is intercepted, it remains protected.

Third is continuous monitoring and automated response. Every open port or configuration is constantly scanned and evaluated. Tools like Cloud Guard monitor the environment 24/7, and if a vulnerability or threat is detected, the system can automatically shut it down.

On top of this, security operates across multiple layers of the infrastructure, making it significantly harder for attackers to find and exploit weaknesses. For startups, the takeaway is clear: security cannot be treated as an afterthought. It has to be designed into the system from day one, just like scalability or performance.

Inc42: Oracle has worked closely with India’s digital public infrastructure. How is this kind of partnership different from a typical commercial engagement, and if you could redesign one aspect of India’s cloud and tech ecosystem from scratch, what would you prioritise for the future?

Palanivel Saravanan: Working on India’s DPI is fundamentally different from a standard commercial engagement. The focus is not just on delivering a service, but on building infrastructure that operates at a national scale, with strict requirements around resilience, security, cost-efficiency and data sovereignty.

One key learning is that systems at this scale must be designed differently from the outset. They need to be highly distributed, resilient and built for massive concurrency, while also ensuring that data remains within geographical boundaries. Data sovereignty and regulatory alignment are not add-ons; they are core design principles.

Another insight is that infrastructure must continuously evolve with use cases. Whether it is enabling AI-driven services, supporting real-time transactions or scaling education platforms, the architecture must remain flexible without requiring constant rework.

If I were to redesign one aspect of India’s cloud and technology ecosystem, three priorities stand out. 

First is sovereign AI. India needs to build and scale its own foundational models relevant to its languages, industries, and use cases. 

Second is strengthening the developer ecosystem. India already has a strong base, but the next phase of growth will come from enabling developers in tier II and tier III cities to access AI tools, modern infrastructure, and security capabilities.

Third is building industry-specific, regulation-ready architectures from day one. In sectors like BFSI, healthcare and agriculture, systems should be designed to handle scale, comply with evolving regulations and maintain low transaction costs without requiring repeated re-architecture.

If these foundations are in place, organisations can focus on innovation and growth, rather than constantly reworking core systems.

The post Oracle’s Palanivel Saravanan On Building AI-Ready Infrastructure For Startups to Scale appeared first on Inc42 Media.