Collate.

Here is a fact about the artificial intelligence business that is both obvious and, somehow, constantly ignored: a model is only as trustworthy as the data you feed it. You can spend a great deal of money on the smartest large language model available and then point it at a data warehouse that no human has fully understood since 2019, and it will answer your questions with total confidence and occasional fiction. The model is not lying, exactly. It just doesn't know which of your seventeen tables called "revenue" is the real one. Neither do you.

This is the problem Collate has decided to make its business. Collate is the company behind OpenMetadata, an open source project for what the industry calls "data discovery, observability, and governance," which is a long way of saying: knowing what data you have, whether it is any good, and who is allowed to touch it. The pitch is that metadata - the data about your data - should not sit inertly in a catalog that people consult twice a year. It should do work. It should document itself, test itself, correct itself, and hand a clean, governed version of the truth to whoever asks, including the AI.

That is a genuinely unglamorous proposition, and Collate seems comfortable with that. Its own tagline is "The AI for Data Platform," and the marketing talks cheerfully about automating "the grunt work." Grunt work is the correct term. Somebody at every company has to write down what a column means, flag when a pipeline breaks, and decide whether the marketing team can see salary data. Historically that somebody was a person, or a committee, or - most often - nobody, which is how you end up with seventeen revenue tables.

The plumbers come back

It helps to know who is making this argument. Collate's co-founders are Suresh Srinivas, the CEO, and Sriharsha Chintalapani. Between them they have spent a couple of decades building the deeply unsexy infrastructure that a lot of the modern data world quietly runs on. Srinivas helped create Apache Hadoop and co-founded Hortonworks; Chintalapani was a committer on Apache Kafka and Apache Storm. Srinivas later served as a chief data architect at Uber, where he helped build DataBook, the company's internal metadata system - which is to say, he has already solved a version of this problem once, at ride-hailing scale, and apparently found it interesting enough to solve again for everyone else.

The through-line matters, because it explains the shape of the company. People who spent their careers in the Apache open source community tend to build companies that start with open source, and Collate is an "open core" business in the textbook sense. OpenMetadata is free. Anyone can download it, run it, and never pay Collate a cent. The community around it has grown to roughly 9,800 members - a number the company describes with a 3,000% growth figure that is impressive and, like all such figures, worth reading as "it started small and got much bigger."

The clever part of that quote is the loop it describes. The old story was: clean your data so the AI can use it. Collate's story is that the AI should help clean the data - the same tool that tidies the room also lives in it. If that works, the grunt work does itself, and the thing that documents your data is the thing that then answers questions about it. If it doesn't work, you have an AI confidently mislabeling your columns at scale, which is a failure mode worth taking seriously. Collate's answer is governance: ground every model in a shared, governed "context layer" so answers cite business meaning rather than guesses.

Where the money actually is

Free software does not, by itself, pay salaries, which brings us to the part with a bill attached. Collate sells a managed, enterprise-grade platform built on top of OpenMetadata - the open source drives adoption, the paid product adds AI automation, security, support, and the enterprise features that a Fortune Global 500 procurement department requires before it signs anything. It's available on the AWS Marketplace. The reported customer list includes Mango, fundcraft, Decisiv, Wix, Carrefour Brazil, PayU Finance, Loggi, inDrive, and the RATP Data Factory in Paris - the sort of spread that suggests the problem is not confined to any one industry, because bad metadata is universal.

The commercial results, as far as they're public, are encouraging in the way early-stage results usually are. Annual recurring revenue grew more than 350% year over year. In July 2025 the company raised a $10 million Series A led by Venrock, with Unusual Ventures and Karman Ventures participating, and Venrock partner Ganesh Srinivasan - a former chief product officer at Confluent, which knows something about turning open source data infrastructure into a business - joined the board. Ten million dollars is a modest round by the standards of the current AI-adjacent fundraising environment, which may be a point in Collate's favor: it suggests a company selling a real product to real customers rather than a slide about the future.

"Last mile" is investor shorthand, but it's an honest description. The first miles of the data business - storage, pipelines, warehouses - are largely solved and largely commoditized. The last mile is the human one: can a person, or now an agent, actually find the right table, trust it, and use it without a three-day email thread. That mile has stubbornly resisted automation because it's less an engineering problem than a bookkeeping one. Collate's wager is that AI has finally made the bookkeeping cheap enough to do properly.

The open-source hedge

One more move is worth noting, because it's the kind of decision that reveals what a company actually believes. In 2025 Collate contributed OpenMetadata's stewardship to the Linux Foundation, the neutral nonprofit that houses Linux itself and much of the world's shared infrastructure. This is a venture-backed company handing its crown jewel to a foundation it doesn't control. The logic is that the moat was never the code - it's the community, the context, and the trust. Enterprises are far more willing to build on a standard that can't be yanked away or acquired out from under them. Giving up control is, counterintuitively, how you keep the customers.

None of this guarantees anything. The data catalog and governance space is crowded - Alation, Atlan, Collibra, and the open source DataHub project are all chasing versions of the same last mile, and "we'll ground your AI in governed context" is a sentence a lot of companies are currently saying. What Collate has that's harder to copy is a founding team that has built this category of software before, a genuinely large open source community feeding the top of the funnel, and a $10,000 grant from the Bloomberg FOSS Contributor Fund, which is a small amount of money and a real signal about who's paying attention.

The bet, in the end, is a bet on the least exciting part of the AI boom. Everyone else is building the brain. Collate is making sure it's fed something true. If the AI era turns out to reward whoever owns the context - the shared, governed meaning underneath the answers - then a modest company in Menlo Park doing the paperwork nobody else wanted to do will have picked a very good spot to stand.