KubeCon India wrapped up its third edition in Mumbai last week (June 18-19), and the energy at the Jio Convention Centre was something else entirely. But this isn't a blog about how great our booth looked or how many stickers we gave away (though, if you picked up a Kloudfuse tote bag, you made a solid choice).

This is about the conversations.

Over two days, our team ran back-to-back demos, fielded questions from SREs, platform engineers, DevOps leads, VPs of engineering, and a fair number of people who opened with "I've heard of you, but I want to see how this actually works." What stood out wasn't any single question. It was the patterns. The same themes kept surfacing across companies of different sizes, different industries, and different stages of observability maturity.

Pic 1: Team Kloudfuse at KubeCon India 2026, Mumbai.

Here's what we heard.

1. "We're drowning in tools. How do we actually consolidate?"

This wasn't new. We heard it in 2024, we heard it in 2025, and we're hearing it louder in 2026. But the nature of the question has shifted.

A year ago, people were asking "is consolidation even possible?" Now they're asking "what does the migration path look like?" and "how long does it actually take?" The conversation has moved from theoretical to tactical.

One of the highlights of this year's event was having folks from Zscaler stop by the booth. When you're processing 400+ TB of telemetry data per day and you've consolidated 30+ monitoring tools into a single platform, that's not a marketing talking point. That's a lived experience. And having real customers share that experience on the booth floor carries more weight than any slide deck.

The tool consolidation conversation usually unfolded in one of two ways. Either the visitor was running Datadog or New Relic and feeling the cost pressure as data volumes scaled, or they had cobbled together an open-source stack (Prometheus, Grafana, ELK, Jaeger) and were spending more time maintaining their monitoring infrastructure than actually using it to resolve incidents. In both cases, the appeal of a unified platform where logs, metrics, traces, profiling data, and real user monitoring all live in a single queryable data lake landed hard.

Pic 2: Pankaj Thakkar (CEO) talking observability on the booth floor.

2. "Data sovereignty isn't a nice-to-have anymore. It's a requirement."

If there was one theme widely talked about it was this: Data residency and sovereignty aren't abstract compliance checkboxes for Indian engineering teams. They're operational realities driven by RBI guidelines, the DPDP Act, and increasingly strict internal security policies at large enterprises.

The Self-SaaS model, where you get the operational simplicity of a managed platform but everything runs in your VPC and your data never leaves your environment, wasn't just interesting to visitors. For several of them, it was the reason they walked up to the booth in the first place.

Fintech companies, healthcare organizations, and government-adjacent teams were particularly engaged here. The questions weren't "do you support VPC deployment?" They were "show me exactly how the control plane works" and "what happens during an upgrade?" and "if I need to pass an audit, what does that look like with your architecture?"

The combination of SOC 2 Type II certification, FIPS 140-3 compliance, and the ability to run everything on customer-managed encryption keys within their own cloud account. That stack of trust signals opened doors that pure-SaaS vendors simply can't walk through.

Pic 3: The Zscaler team stopping by the Kloudfuse booth at KubeCon India 2026.

3. "OpenTelemetry is settled. Now tell me what to do with all this data."

This is a meaningful evolution from previous years. The "should we adopt OpenTelemetry?" debate is effectively over. Teams have adopted it, or they're in the process of migrating. The new frontier is: how do you actually extract value from all the telemetry you're now collecting?

Two specific areas came up repeatedly:

Correlation across signals. Engineers don't want to look at logs in one tool, traces in another, and metrics in a third. They want to start from an alert, pivot to the relevant traces, drill into the specific log lines, and see the corresponding infrastructure metrics, all without leaving the platform or losing context. The unified data lake architecture, where everything is stored and queryable together (not just displayed in the same UI through federation), was the "aha" moment for a lot of booth visitors.

Making sense of AI workloads. If you're building applications that use LLMs, you now have a new class of infrastructure to monitor. Token usage, latency per model call, error rates across different providers, cost attribution per request. Teams building AI products were actively looking for observability solutions that can handle these workloads natively alongside their traditional metrics and traces.

Pic 4: The engineers in action. From left: Ayush Singh, Pankaj Thakkar, and Swapnil Kulkarni.

4. "Everyone's talking about AI for operations. But the details matter."

If there was one topic that came up in nearly every conversation at the booth, it was AI. But the nature of the questions has shifted dramatically from a year ago. People aren't asking "will AI change operations?" anymore. They're asking very specific, grounded questions: How do I keep a human in the loop? How do I make sure the AI actually understands my environment? How do I avoid just adding another black box to my stack?

We previewed an upcoming release at the booth that addresses exactly this space, and the reception was strong. Without getting into details ahead of the formal launch, the thing that resonated most with visitors was the emphasis on trust and transparency. Engineers want AI that investigates alongside them, not AI that takes action behind their backs. That design philosophy landed hard, especially with teams who've been burned by overly aggressive automation in the past.

The other piece of the AI conversation was around the MCP Server. The Model Context Protocol is an open standard for connecting AI agents to data sources, and Kloudfuse's MCP Server makes your entire observability data lake accessible to AI agents while keeping everything inside your VPC. Engineers who were already building with LLMs immediately understood the implications: you can query your logs, metrics, and traces through natural language without shipping data to a third party, and you can build your own agents on top of it.

Taken together, these two capabilities point to where observability is heading. It's becoming less of a passive monitoring layer and more of an active data substrate for AI-powered operations. But the teams we spoke to at KubeCon are clear-eyed about one thing: they want that future built on data they own, in environments they control, with humans making the final call. We couldn't agree more.

Pic 5: Ayush Singh (Customer Success Architect) running a live demo for booth visitors.

The Bigger Picture

Three years of KubeCon India, and the trajectory is clear. In 2024, people were curious about unified observability. In 2025, they wanted to see proof. In 2026, they're asking for implementation timelines.

The conversation has matured. Teams aren't evaluating observability tools based on feature checklists anymore. They're evaluating based on three questions: Does it run in my environment? Can it handle the scale I'm headed toward? And does it actually reduce the time my team spends fighting fires?

That shift is what makes these booth conversations valuable. Not as a marketing exercise, but as a genuine signal of where the market is heading. And it's heading toward fewer tools, more ownership, and AI that augments engineering teams rather than replacing their judgment.

Missed Us at the Booth?

If you didn't make it to KubeCon India this year, we'd still love to talk. Whether you're evaluating observability platforms, planning a migration off your current vendor, or curious about what we're building next, reach out and book a session.

Request a Demo | Try the Playground