Here is a thing everyone in enterprise software agrees on, right up until it costs them a fortune: to analyze your data, you first have to gather it. You extract it from the systems where it lives, transform it into some agreeable shape, and load it into a warehouse. ETL. It is the plumbing of the modern enterprise, and like most plumbing, nobody thinks about it until it floods the basement.
Zetaris, a data company founded in Melbourne in 2013 by Vinay Samuel and Jason Jaesung Jun, was built on a quietly heretical premise: what if you didn't? What if the data just stayed where it was, and the questions came to it instead? In the trade this is called data virtualization, or federated query, or - if you want the 2025 phrasing - an "open agentic lakehouse." The idea is old enough that it sounded a little dull when Zetaris started saying it. The idea is now, suddenly, the thing every enterprise trying to deploy AI desperately needs.
This is the useful thing about being early. If you are right about a technical idea ten years before the market wants it, you spend a decade looking slightly out of step and then, one day, the market arrives at your doorstep and acts like you invented the future last Tuesday.
The pitch is almost aggressively simple: stop moving your data to the AI. Bring the AI to your data.
What it actually does
The mechanics are more interesting than the slogan. Zetaris presents a single query layer over a sprawl of disparate sources - structured tables, unstructured blobs, streaming feeds - and lets you ask one question that reaches across all of them without first copying anything into a central pile. Underneath, it runs what it calls multi-engine intelligent routing: the platform looks at each job and picks the right execution engine for it - Spark, Trino, Presto, or DuckDB - so a heavy distributed job and a quick local scan don't get forced through the same pipe.
If you have ever managed a data team, you know why this matters. A great deal of money in analytics is wasted running the wrong engine for the workload - spinning up a cluster to answer a question a laptop could handle, or choking a laptop with a question that needed a cluster. Zetaris's claim is that automating engine choice cuts compute meaningfully, and that skipping the copy step cuts cost again. The company puts numbers on it - roughly 40% lower data costs, up to 60% less compute, "10x faster insights." Treat vendor arithmetic with the usual caution. The mechanism behind the numbers, though, is real and unglamorous: you save money by not doing the expensive thing everyone assumed you had to do.
Why now
The AI boom did something strange to the data-infrastructure market. It revealed that the exciting part - the models - is only as good as the boring part - the data they can reach. An AI agent that can't see your governed, current, trustworthy enterprise data is an AI agent making things up in a corporate voice. Getting that data to the agent, safely, without spraying copies across five clouds, turns out to be the hard problem. It is exactly the problem Zetaris has been working on since before anyone said "agentic."
An AI agent is only as smart as the data it can reach. Zetaris made the data reachable without making it movable.
That distinction - reachable but not movable - is the whole business. Data that moves is data that gets duplicated, goes stale, escapes its governance, and shows up in a breach report. Data that stays put, queried in place with role-based security and a unified semantic layer on top, is data you can actually let an AI agent touch. Zetaris frames itself not as a business-intelligence tool but as a "control plane for enterprise AI," which is marketing language for a genuinely load-bearing position: the thing that sits between an organization's data and its AI, deciding what gets seen, by whom, and how.
The Melbourne thing
Zetaris is Australian, headquartered in Melbourne, with a foot planted in Palo Alto and a distributed, remote-friendly team of around 54. This is worth noticing. Deep data infrastructure is not the kind of thing Silicon Valley assumes gets built elsewhere, and Zetaris is a small, unflashy counterexample - engineering-led, founder-driven, built around a specific technical thesis rather than a growth-hacking playbook. Vinay Samuel is not a first-time founder chasing a trend; he spent the 1990s and 2000s around parallel-database pioneers like Netezza, Greenplum, and Vertica before starting the company. Zetaris is, in a sense, the argument he has been making about databases for thirty years, finally packaged for the AI era.
The Equinix moment
In June 2025 Zetaris announced a global collaboration with Equinix, the interconnection giant whose data centers quietly stitch together large parts of the internet. Zetaris would host its Modern Lakehouse for AI inside Equinix's IBX facilities and offer it, free, to Equinix Fabric customers through the marketplace. For a federated-data company, this is a natural fit bordering on poetic: a platform whose entire premise is reaching data wherever it sits, plugging into the physical fabric that connects wherever-it-sits. It also gives a small Melbourne company a very large distribution surface, alongside existing listings on AWS Marketplace and Microsoft Azure.
Is Zetaris going to win the enterprise-data war outright? It is playing against Denodo, Starburst, Dremio, and the gravitational pull of Databricks and Snowflake, all of whom would like to own this exact real estate. But it doesn't need to win everything. It needs to be right that data shouldn't move - a bet it placed a decade ago, and one the AI era keeps making look smarter.