Amit Sharma, CEO and Founder of CData – Interview Series

Amit Sharma, CEO and Founder of CData Software, is a tech executive who has guided CData from its early startup phase to become a global leader in data connectivity and integration. With a career spanning roles as a software engineer at Infosys and Elixar, a technical architect at /n Software, and later CTO at CData, he has built deep expertise in enterprise data access and infrastructure. Since becoming CEO in 2014, he has led CData’s mission to simplify how organizations connect, integrate, and utilize data across systems, helping position the company as a foundational layer of modern data movement.

CData Software is a leading provider of data access and connectivity solutions. Its self-service data products and connectivity platforms deliver universal access to live data across hundreds of widely used on-premises and cloud applications. Millions of users worldwide rely on CData to support advanced analytics, accelerate cloud adoption, and build more connected, data-driven organizations. Designed to be consumable by any user, accessible within any application, and scalable for enterprises of all sizes, CData is redefining how businesses access and use data.

You started your career in India at Infosys and later transitioned into U.S. enterprise software. What early lesson from that phase still shapes how you lead today?

My time at Infosys gave me early exposure to the demands of large-scale enterprise technology — the complexities, the need for reliability, and how big organizations approach technical problems. That shaped a deep respect for structure and enterprise-grade quality. But when I transitioned to a U.S. startup, I discovered I thrived on speed, agility, and the ability to make direct impact. Today, that dual background guides how I lead CData Software: I insist on enterprise-grade standards and robustness, while fostering a lean, fast-moving culture that values simplicity, real-world usability, and rapid execution.

After more than a decade as CEO of CData, what shift in mindset or approach was most important in scaling the company from its early days into a global enterprise?

The biggest shift for me was moving from thinking like a builder of technology to thinking like a builder of an organization. In the early days, my focus was almost entirely on the product; making sure it was elegant, reliable, and solved real problems. As CData scaled, I had to learn that great software alone isn’t enough; you need great people, strong leaders, and processes that scale without slowing you down. That meant investing earlier in hiring, empowering teams, and building repeatable systems across sales, support, and operations, while still protecting our engineering culture. The mindset shift was realizing my job wasn’t just to make great technology, but to create an environment where great technology could be built consistently by a growing global team.

CData has long focused on “simplifying access to any data, anywhere.” How has that mission evolved as the industry moves deeper into AI-native applications?

From the beginning, our mission at CData has been to make data universally accessible using familiar, standardized interfaces, because we believed the biggest bottleneck to innovation wasn’t storage or compute, rather it was access. That core idea hasn’t changed, but the context has. As organizations moved from analytics to cloud and now into AI, the cost of fragmented, inconsistent data access has only increased. What’s evolved is our responsibility: it’s no longer just about connecting applications to data, it’s about making sure data is trustworthy, real-time, and usable across increasingly complex and distributed environments. In the AI era, access alone isn’t enough. Data has to be immediately usable without weeks of custom engineering.

As AI-native applications become the norm, our mission has expanded to include making data AI-ready by default. That means enabling consistent semantics, high-performance connectivity, governance-aware access, and real-time integration across structured and SaaS data sources, so models and agents can work with fresh, reliable information, not brittle point integrations or stale copies. In practical terms, we’re focused on eliminating the friction between where data lives and where AI systems operate, so teams can move from experimentation to production faster. We see ourselves not just as a connectivity provider, but as a foundational data layer for AI-driven enterprises quietly powering the systems that make intelligent applications possible.

With generative AI accelerating, what does “AI-ready data” really mean to you, and where do you see organizations misinterpreting that idea the most?

To me, “AI-ready data” means data that is accessible, reliable, current, and understandable by both humans and machines without layers of custom plumbing. It’s not just about moving data into a lake or warehouses. It’s about ensuring that systems, models, and agents can consistently access the right data at the right time through standard, governed interfaces. AI readiness depends less on where data is stored and more on whether it can be discovered, queried, trusted, and integrated in real time. Without that foundation, even the most advanced models end up operating on incomplete or stale information.

Where I see organizations misinterpreting the concept is in assuming that centralization automatically equals readiness. Teams often believe that once data is consolidated into a single platform, they’re “AI-ready,” when in reality they’ve just created a new silo. Others overinvest in tooling without addressing data quality, semantics, and connectivity, the unglamorous problems that make or break real-world AI systems. AI doesn’t fail because of models; it fails because of messy, inaccessible, or outdated data. The organizations that will win are the ones that treat data readiness as operational discipline, not a one-time migration project.

Your new research, The State of AI Data Connectivity: 2026 Outlook, shows that only 6% of AI leaders believe their data infrastructure is fully ready for AI. Why do you think the readiness gap is so large, and what does this tell us about the industry’s current trajectory?

The gap is so large because most organizations invested in collecting and storing data long before they invested in making it usable for AI. Over the past decade, companies have built lakes, warehouses, and pipelines, but they rarely built a cohesive access layer that ensures data is consistent, real-time, and available across systems. As a result, leaders discover that once they begin deploying AI into real workflows, their underlying infrastructure can’t support the speed, scale, or reliability AI demands. The 6% figure doesn’t reflect a lack of ambition, but rather the reality that AI exposes weaknesses that were always there but didn’t matter as much in traditional analytics.

What the data tells us about the industry is that we’re early in the AI adoption curve, not late. Organizations are experimenting aggressively at the application layer, but they’re now realizing that success depends on modernizing their data foundation underneath. We’re entering a corrective phase where the focus is shifting from flashy pilots to operational readiness—standardized access, governed integration, and real-time connectivity. The winners won’t be the companies that build the most proofs of concept, but the ones that modernize their data infrastructure fast enough to move those experiments into production at scale.

The findings also show that 71% of AI teams spend more than a quarter of their time on data plumbing. In your view, what part of this work is actually strategic rather than just technical debt?

Some amount of “data plumbing” is absolutely strategic when it’s about creating durable access to data through standard interfaces and designing for scalability and governance from the start. Investing in consistent connectivity, shared semantics, and reliable integration patterns is foundational work that pays dividends across every application and model that comes later. The problem is that most teams aren’t doing that kind of plumbing. They’re rebuilding one-off pipelines, writing brittle connectors, and patching integrations that only solve a problem once. That’s technical debt disguised as progress.

What’s strategic is anything that reduces future friction: eliminating custom code in favor of standards, building reusable data services, and connecting systems in ways that scale across teams and use cases. When plumbing becomes invisible and repeatable, it stops being a tax on AI teams and becomes an enabler. The real goal isn’t to spend less time on data. It’s to stop spending time on the same data problems over and over again.

One striking data point from the report is that 46% of enterprises now require real-time access to six or more data sources for a single AI use case. Does that reflect what you’re seeing with customers, and what makes that level of connectivity so difficult?

Yes, that aligns closely with what we’re seeing with customers. Modern AI use cases, whether predictive analytics, recommendation engines, or autonomous workflows, rarely rely on a single system. Enterprises often need to combine ERP, CRM, SaaS apps, streaming platforms, and legacy databases to generate meaningful insights. The challenge isn’t just the number of sources; it’s the variety, different protocols, formats, and update frequencies, and the expectation that this data be available in real time for AI models to consume.

What makes this level of connectivity difficult is that traditional integration approaches were never designed for the scale, speed, and reliability AI requires. One-off connectors and batch pipelines simply can’t keep up. True real-time access demands standardized, managed interfaces, consistent semantics across systems, and monitoring to ensure data quality and availability. Without that foundation, teams spend more time firefighting pipelines than building AI solutions, which slows innovation and introduces risk. The organizations that succeed are the ones that treat connectivity as a strategic capability, not just a technical chore.

The report emphasizes semantic consistency, context, and connectivity as defining characteristics of mature AI data infrastructure. How should organizations think about sequencing these priorities?

When thinking about sequencing, organizations should start with connectivity. If data isn’t reliably accessible across systems, everything else becomes irrelevant. AI models can’t learn from what they can’t reach. Establishing standardized, governed connections across all critical data sources lays the foundation for everything that follows. Without that layer, teams end up building fragile, one-off pipelines that create more work down the line.

Once connectivity is in place, semantic consistency becomes the next priority. Data needs a common language so that information from multiple sources can be interpreted correctly and combined meaningfully. Context naturally follows: understanding not just the values but their meaning within the business process, timing, and relationships ensures that AI models can make accurate, actionable predictions. Treating these elements as a structured sequence—connectivity first, semantics second, context third—allows organizations to build an AI-ready data infrastructure that scales and supports reliable, production-ready intelligence.

AI-native software providers now require roughly three times more external integrations than traditional vendors. What’s driving this widening gap, and what does it reveal about where software is headed?

The widening gap is driven by the nature of AI itself: AI-native applications thrive on diverse, real-time data from multiple sources. Unlike traditional software, which often operates within a single system or suite, AI models need to ingest, correlate, and analyze information across ERP systems, CRM platforms, SaaS apps, streaming sources, and more. Each integration is essential to give the AI sufficient context and coverage to generate accurate predictions, recommendations, or automated actions.

This trend reveals that software is moving from isolated applications toward interconnected, intelligent ecosystems. The winners won’t be the products that work well on their own. They’ll be the platforms that can seamlessly access and integrate data wherever it lives. In practical terms, it means that connectivity, standardization, and real-time integration are no longer nice-to-haves. They’re foundational capabilities for AI-native software to deliver real value.

Looking ahead five years, what do you believe will become the most significant bottleneck for AI success—connectivity, real-time pipelines, semantic modeling, governance, or something else entirely?

Looking ahead, I believe governance and security will become the most significant bottleneck for AI success. While connectivity and real-time pipelines remain foundational—AI models can only be as effective as the data they can access—organizations are rapidly realizing that ungoverned AI is unsustainable and potentially dangerous. As AI moves from experimentation into production and begins influencing critical business decisions, the risks of bias, compliance violations, data leakage, and operational errors multiply exponentially.

The challenge isn’t just about moving data anymore—it’s about moving the right data, with the right controls, to the right systems, in a traceable and auditable way. Organizations that fail to embed strong governance frameworks and security protocols from the start will face mounting regulatory pressure, reputational risk, and ultimately, AI systems they can’t trust or scale. We’re already seeing early signs: businesses hesitant to deploy AI because they can’t ensure data lineage, access controls, or compliance with evolving regulations.

The most successful organizations five years from now will be those that treat governance and security not as afterthoughts, but as core enablers of AI. Yes, you need connectivity and real-time pipelines to get data flowing—but without governance and security in place, that data becomes a liability rather than an asset. The future of AI isn’t just about speed or scale; it’s about trust, accountability, and responsible deployment at every layer of the data stack.

Thank you for the great interview, readers who wish to learn more should visit CData Software.