Data mesh is the data idea everyone's arguing about right now, and the argument tends to be unhelpfully binary: either it's the revolution that finally fixes enterprise data, or it's consultant-ware, old wine in a new bottle. Having sat with it for a while — and having led a platform team big enough to feel the pain it's responding to — I think the honest answer is both, in a specific way that's worth pulling apart. There's one genuinely sharp, genuinely important insight at the heart of data mesh. And there's a growing pile of people trying to sell you technology to solve what is fundamentally an organisational problem. Telling those apart is the whole game.
The problem it's actually responding to
To judge data mesh fairly, you have to understand the real pain it's reacting against, because that pain is genuine and I've lived it. The traditional model is centralised: one central data team owns the platform and, in effect, becomes responsible for understanding, integrating, and serving all the organisation's data, from every domain.
And at scale, that central team becomes a bottleneck — and worse, an ignorant bottleneck through no fault of its own. They're asked to be the experts on data from dozens of domains they don't actually work in — sales data, logistics data, finance data — and they can't be, because that knowledge lives in the domains, not in the central team. So you get a central team drowning in requests, doing a mediocre job of understanding data whose meaning they can only ever grasp second-hand, while the people who do understand it — the domains — are kept at arm's length from producing it. That's a real, painful, common failure, and any honest data leader at scale has felt it.
The genuinely sharp insight
Data mesh's core answer to that is genuinely good, and it's an organisational insight, not a technical one: push the ownership of data to the domains that actually understand it, and have them treat their data as a product they're responsible for serving well to others.
Instead of a central team trying to understand everyone's data, the sales domain owns and serves its data (they understand it best), logistics owns theirs, and so on — each treating their data as a product with real accountability for its quality, its documentation, its usability by others. That's a sharp idea, and it echoes something I keep arriving at from every direction: that the meaning of data lives with the people who produce and understand it, and any model that separates the data from that domain understanding is fighting reality. Data mesh takes that seriously as an operating model, and that's the revolution part — a real rethink of who is responsible for data, which is a genuinely important and largely correct correction.
Where the rebrand (and the selling) creeps in
But here's where I get wary, and where the "revolution or rebrand" question gets its bite. The core of data mesh is an organisational and cultural change — about ownership, accountability, and how teams are structured and responsible. And there is a strong, well-funded temptation to turn it into a technology purchase, because organisational change is hard to sell and software is easy to sell.
So you increasingly see data mesh pitched as a thing you buy — a platform, a set of tools, an architecture you implement. And that's the rebrand, and it's a dangerous one, because:
- The hard part isn't technical, so tools won't deliver it. Getting domains to actually own their data as a product — to accept the accountability, build the capability, change how they're organised and incentivised — is a people-and-org transformation. No tool does that. Buying "data mesh technology" and changing nothing about ownership gets you the old problems with a fashionable label.
- It's genuinely hard, and the hype hides the difficulty. Pushing data ownership out to domains asks a lot of those domains — skills, capacity, discipline they may not have. Done badly, decentralisation just produces fragmentation: a hundred domains each doing data differently and badly, with no central coherence, which can be worse than the flawed central model. The hype rarely mentions how easily this goes wrong.
- Most organisations aren't feeling the pain that justifies it. Data mesh is an answer to the pain of scale — the central-team bottleneck at a large, complex organisation. A smaller organisation adopting "data mesh" because it's the trend is solving a problem it doesn't have, and importing a lot of complexity to do it. The right answer for most is not a mesh.
So — revolution or rebrand?
My honest verdict: the idea is a genuine and important one — a real, largely-correct rethink of who owns data, responding to a real failure of the centralised model at scale. And the packaging is increasingly a rebrand, as an organisational insight gets repackaged into a technology sale that quietly drops the hard, essential, non-technical core.
Which means the same discipline I bring to every hype cycle applies here exactly: take the real insight, refuse the technology-will-fix-it framing, and be honest about whether you even have the problem it solves.
Data mesh's revolution is organisational: put data ownership where the understanding is, and treat data as a product. Its rebrand is the attempt to sell you that as software. The insight is real and hard; the shortcut is fake and easy — and the whole value is in doing the hard real thing rather than buying the easy fake one.
The principle everyone skips: federated governance
There's a part of the original data mesh idea that the hype consistently drops, and it's the part that stops decentralisation from becoming chaos: federated governance. The real insight isn't just "push ownership to the domains" — it's "push ownership to the domains while maintaining enough shared, central standards that the whole thing still coheres."
That balance is the hard, essential middle, and it's exactly what gets lost when data mesh becomes a slogan. Pure centralisation gives you the bottleneck. Pure decentralisation gives you a hundred domains each doing data their own way — no shared definitions, no interoperability, no way to combine anything across domains. That's fragmentation, and it's arguably worse than the bottleneck it replaced. The whole point of the mesh idea, done properly, is the federated bit: domains own their data, and there's a shared set of standards — for how data products are described, how they interoperate, how identity and key definitions stay consistent — that everyone agrees to live within.
Get that balance right and you have autonomy without anarchy. Get it wrong in either direction and you have either the old bottleneck or a new mess. And notice: that federated-governance balance is, once again, a human and organisational achievement — a negotiation about which decisions are local and which are shared — not something a tool hands you. It's the hardest part of the whole idea, and the part the technology-purchase framing simply cannot deliver, which is precisely why the versions of data mesh that quietly skip it fail. If you take the mesh idea seriously, take the federated-governance part most seriously — because it's the part that decides whether decentralisation liberates you or fragments you.
What I'd actually do with it
If you're a data leader looking at data mesh, my advice is to steal the insight and ignore the movement. The genuinely valuable idea — that domains should own their data and treat it as a product, because that's where the understanding lives — you can apply thoughtfully, incrementally, at whatever scale you actually operate, without adopting a capital-letters Data Mesh Architecture or buying anything. Push ownership toward the domains where it helps. Ask producers to be accountable for their data's quality and usability. Keep enough central coherence that decentralisation doesn't become fragmentation.
Do that and you've captured the revolution — the real correction to the central-bottleneck problem — without falling for the rebrand. Buy a "data mesh platform" expecting it to deliver the organisational change for you, and you'll have paid for the label and skipped the substance. As ever, the technology was never the hard part or the point. The hard part is getting humans and teams to own things differently, and no amount of mesh comes in a box.