Vector search is table stakes - OpenSearch Foundation bets on data sovereignty and enterprise trust instead

Every open source project reaches a moment where community enthusiasm meets enterprise procurement requirements, and the two do not naturally fit. For the OpenSearch Software Foundation, that moment has arrived.

I sat down with Bianca Lewis, Executive Director of the OpenSearch Software Foundation, and Carl Meadows, Chair of the Governing Board and Director of Product Management for OpenSearch at Amazon Web Services (AWS), at OpenSearchCon Europe 2026 in Prague. The conversation followed a strong keynote in which three speakers had argued that the enterprise AI bottleneck has moved from the model to the data layer. In the interview, Lewis and Meadows went deeper – and were very transparent about both the progress and the distance still to travel.

Where community meets procurement

Lewis was candid about what the Long-Term Support (LTS) program is really for. Most of the Global Fortune 500 are already using OpenSearch in some capacity. But enterprise buyers kept running into the same set of needs that open source projects have historically struggled to meet: formal vendor support, security documentation with a fallback if something goes wrong, and a guarantee that the version you build on today will be maintained long enough that you are not forced into constant upgrades. She explained:

These are obvious questions that open source, being such an open and vibrant community, has struggled to solve. And that's why you find bits and pieces of open source projects in enterprise companies. But when you look at some of the main systems, they're often sitting with proprietary vendors.

OpenSearch's answer is to accredit independent vendors – currently BigData Boutique, Eliatra, and Resolve Technology – to provide official, Foundation-certified support for designated LTS releases. Each vendor is bound by a conformance agreement, carries a trademarked badge, and must contribute all fixes upstream. Enterprises can switch between accredited vendors without replatforming, with no migration and no lock-in.

Sovereignty is global, not European

I asked whether data sovereignty has stopped being a European compliance story and become a universal architecture question. Lewis went further – Europe's Cyber Resilience Act (CRA), she argued, is only one piece of a much larger global picture:

Three weeks ago, I was in China. I was talking to one of the leaders of the Vietnamese community in terms of cyber security, and the compliance landscape in Vietnam makes the CRA look like child's play. Not even mentioning the Chinese firewall, and California, and New York... The amount of compliance and regulation legislation being tightened everywhere.

Regulation is converging globally, even though the specific laws carry different names. Lewis sees the LTS accreditation model as built for that reality – a European company can choose a European support vendor, an Asian operation can choose an Asian one, all under the same Foundation umbrella.

Meadows brought it back to operations. Many of his AWS customers also run workloads in their own data centers, he noted:

A lot of my customers will be like, 'yeah, it's great. I get all of that when I'm on AWS, but I don't run everything on AWS. I've got some data centers over here, some data centers over there, and what do I do about them'?

Can your search support a decision?

Dom Couldwell at IBM had used his keynote slot that morning to challenge the fixation on vector search as a silver bullet. I asked Lewis and Meadows how OpenSearch avoids riding the same hype curve.

Lewis did not hedge:

The conversation is not feeding the hype of a vector search engine, which is table stakes, but is rather to look at whether your search is modern enough to base a decision on. And that is the correct language that OpenSearch is going to preach. Not AI. Not vector search. Not hybrid search. Not anything else. But is your search really scalable, with the right data layer, to support that infrastructure, to give you not accurate results only, but accurate results that you can base a decision correctly on.

Meadows was also honest about the difficulty. Potential adopters arrive thinking in vector terms because that is where the market noise is:

The journey is very consistently that somebody starts with just a pure vector solution, and then they're like, 'oh, well, now I need geospatial filtering, and now I need to actually complement this with lexical'. Then they have to replatform.

Vendor neutrality – follow the money

I asked Meadows how he balances chairing the governing board with running the OpenSearch product team at AWS. He was pragmatic about it: AWS initiated OpenSearch when the Elasticsearch license changed, and the broad community that grew around it became central to how AWS makes money. He reflected:

Our business model works really well for us, which is a broad community of folks using this technology all over the place. Eventually, they want to run stuff on AWS, and we make money.

Transitioning to the Linux Foundation gave that commitment legal weight:

It's really even exploded post that, because now it's not just our word. It's real. We can't go back.

Lewis added another layer – AWS sells compute infrastructure, not OpenSearch subscriptions. Owning the application is more of a burden on its business model than it would be for a subscription-based software company. She continued:

Where does AWS actually make their money? They always, from day one, made their money from selling compute infrastructure. So for them to own an application is more of a burden for their business model than it is from any other vendor.

Growing membership tells part of that story – premier members now include AWS, IBM, SAP, and Uber, with CERN joining as an associate member. Lewis's fiduciary duty as Executive Director is to ensure the project innovates independently of any single contributor. Even, as she put it, if a meteorite storm selectively falls on all AWS data centers.

My take

What Lewis and Meadows described is not a search engine pitch. They are building conditions under which enterprises can own their AI data layer without vendor lock-in – and without sacrificing the support, security documentation, and lifecycle guarantees that procurement teams require.

LTS accreditation is the clearest expression of that ambition. Structurally simple, operationally transformative for any enterprise that has wanted to run open source in production but could not get past the procurement conversation. That it took this long for someone to do it properly is, as Lewis observed, the most surprising part. I will be testing the OpenSearch Observability Stack and Agent Hub hands-on in the coming weeks. For now, the message is that the vector hype cycle is a distraction, the data layer is the real work, and open source has just gained a practical new answer to the enterprise trust question.