Issue No. 042 — Data Virtualization
It is a Tuesday morning, somewhere in the back office of a global bank. A risk team needs an answer that involves three cloud warehouses, one ancient mainframe, a CRM, and a spreadsheet someone in Frankfurt swears is authoritative. The old way takes a quarter. With Denodo's virtual layer in the middle, it takes the length of a coffee.
This is the scene Denodo has been quietly engineering for twenty-five years. The pitch has never been loud. Most of the company's customers cannot say its name on stage because their compliance teams will not let them. And yet Denodo Technologies, headquartered on University Avenue in Palo Alto, sits at the center of the most fashionable conversation in enterprise software: how to make data actually usable, in real time, for both humans and the AI models suddenly demanding it.
01 / The ProblemThe data warehouse ate the office.
By the late 1990s, enterprise data architecture had become a sort of polite hoarding. Every new question demanded a new pipeline. Every pipeline produced a copy. Every copy needed cleaning, governing, securing, and explaining to an auditor who would inevitably ask, "Wait - which copy is the right one?"
Dr. Angel Vina, then a professor at the University of A Coruna in northwest Spain, looked at the trajectory and saw an absurdity. Storage and replication were not converging on a tidy single source of truth - they were diverging, expensively, forever. The traditional approach was sound for the era it was built in. It was not sound for the era arriving.
So in 1999 he started Denodo, not as a venture-funded blitz but as a research spinout. The premise was almost rude in its simplicity: build a layer that pretends all the data is in one place, while leaving every byte where it actually lives. Call it data virtualization. The category did not exist yet. It would, mostly because of them.
02 / The BetAn academic spinout that refused to act like one.
The first six years were spent doing what most startups skip: actually finishing the engineering. By the time Denodo opened its Silicon Valley office on January 1, 2006 - a New Year's Day move-in that Vina still cites as his favorite memory - the platform had been hardened in European telcos and banks, the kind of customers who do not give second chances.
The unusual thing about Denodo is the org chart. The founder is still CEO. The CTO, Alberto Pan, is also still on the books as an associate professor in A Coruna (currently on leave; the door, presumably, remains open). Most of the senior engineering team has been there for over a decade. It is a profitable, founder-led, category-defining enterprise software company - a sentence so rare it deserves its own museum exhibit.
03 / The ProductOne question, four hundred answers, no copies.
The Denodo Platform is, at its core, a logical data management layer. You point it at your sources - cloud warehouses, on-prem databases, SaaS apps, lakes, mainframes, files, APIs - and it builds a single semantic layer on top. Your analyst writes one query. Denodo decomposes it, pushes work down to the right engines, joins the results in flight, applies governance and security, and hands back a tidy answer. The data itself never moves.
In practice this means Denodo plays with everything: Snowflake, Databricks, BigQuery, Redshift, SAP, Oracle, SQL Server, Salesforce, S3, Azure Data Lake, Hadoop, the works. The company likes to point out that a typical deployment connects to 400+ source systems. That is not marketing; it is a procurement diagram nobody wants to look at, which is precisely why a virtual layer is so useful.
Sitting on top of the platform is the Denodo Data Catalog, which now ships with generative AI capabilities: natural-language querying via integrations with ChatGPT and Azure OpenAI, and a feature called DeepQuery aimed at AI-driven insights. The pattern is familiar by now - an old, well-built piece of infrastructure suddenly very relevant because LLMs need clean, governed, real-time access to enterprise data, and shipping them a stale CSV is no longer acceptable.
04 / The ProofNumbers that do their own bragging.
Customers report up to 4x faster time to insight, 345% ROI, and 10x performance gains over traditional data lakehouse approaches. These numbers come from Denodo and its analysts, so apply the usual seasoning, but the underlying point holds: copying less data tends to be cheaper and quicker than copying more data. Strange that this still counts as a revelation.
Revenue, going somewhere
In September 2023, TPG led a $336M Series B. Denodo had taken essentially no growth-stage capital for two decades. Then it took a generation's worth in a single round. The market read that as conviction; the company read it as the right time to push harder on cloud delivery, AI integrations, and partnerships with Snowflake, Databricks, AWS, Azure, and Google Cloud. Everyone signed up.
Midpoint / MilestonesA 25-year scroll, in eight stops.
05 / The MissionMake data act like one thing.
Denodo's stated mission is to turn enterprise data into a strategic, accessible asset - the kind of phrase every data vendor says into a microphone. What is interesting is how literally Denodo means it. The platform was designed, twenty-five years ago, around the assumption that data would always be everywhere. There was no nostalgia for a single source of truth, because the founder never believed there would be one.
That worldview is, awkwardly for the rest of the industry, exactly what AI now requires. Large language models do not care which warehouse holds the answer; they need it fast, governed, current, and explainable. Denodo's logical layer happens to deliver all four. The company did not pivot toward AI. AI pivoted toward Denodo.
06 / Why It Matters TomorrowThe plumbing was the strategy.
The unfashionable truth about enterprise software is that the boring layer always wins. Identity, payments, observability, integration - the categories that sounded like infrastructure plumbing in year one turn out to be the only ones that matter in year twenty. Denodo bet on plumbing in 1999. It is still betting on plumbing. The plumbing now serves AI.
Competitors abound: Informatica, IBM, Oracle, Microsoft Fabric, Starburst, Dremio. Each comes at the problem from a different direction - ETL heritage, lakehouse query engines, cloud-native rebuilds. Denodo's distinguishing claim is that it has been doing the logical layer since before "logical layer" was a term anyone said in meetings. Twenty-five years of edge cases is a moat that does not show up on a feature comparison chart.
Back to that Tuesday morning at the bank. The risk team gets its answer before the coffee gets cold. Nobody on the call thanks Denodo by name; nobody on the call needs to. Somewhere in Palo Alto, a query is decomposed across nine sources, federated, governed, and returned. No data was moved to make this happen. That, twenty-five years on, is still the entire idea.