Data mesh vs. centralized data platform: choosing the right model
Data mesh solves a real organizational problem, but it isn't free. Here's how to decide which architecture actually fits your organization.
Data mesh became a popular answer to a real problem: centralized data teams becoming bottlenecks as organizations scale, with every new data product request queued behind dozens of others. Distributing ownership to domain teams can relieve that bottleneck — but it introduces new coordination costs that are easy to underestimate.
A centralized data platform works well when an organization has a small number of domains, a data team with enough capacity to serve them, and a strong need for consistent governance. It becomes a bottleneck when domain count and request volume outpace the team's ability to keep up.
Data mesh works better in organizations with many distinct domains, each with enough engineering maturity to own their own data products responsibly. Without that maturity, distributing ownership often just distributes the inconsistency instead of solving it.
In practice, most enterprises land somewhere in between: a centralized platform team that owns infrastructure, governance, and shared tooling, with domain teams owning the data products built on top of it. The right model depends less on the architecture diagram and more on where your organization's actual bottleneck sits today.