Data mesh and data fabric caught the attention of data leaders in the early 2020s as a compelling way to retain and deliver information across highly disparate enterprises. Then AI came into the picture in a big way, overshadowing the time and attention devoted to mesh and fabric.
Lately, because data-hungry AI systems demand a seamless flow of information, there has been a resurgence of interest in the flexibility that mesh and fabric provide. To meet this need, mesh and fabric are being layered together, rather than being categorized as differing approaches to data decentralization.
“Ironically, AI has helped bring data mesh and fabric back,” said Tom Thomas, senior data engineering manager at Indeed. “As organizations started building AI products and pilots, they found that AI is only as good as the data behind it. Trusted, discoverable, and well-governed data suddenly has become a priority again.”
What’s different now, according to Thomas, is the tone of the conversation. “A few years ago, people spent a lot of time debating architecture patterns. Today, most are dealing with a more practical question: How do we give teams more freedom to work with data without ending up with dozens of new silos scattered across the business?”
Demand for data mesh and fabric is on the rise, but organizational issues still pose barriers, and use cases vary. “What I see in practice, over and over, is companies becoming what I’d call meshish,” said Animesh Kumar, co-founder, CTO, and CPO for The Modern Data Company. “They adopt the ownership principles and the product mindset, but they keep a strong central platform team rather than fully decentralizing infrastructure.”
Tellingly, enterprises aren’t implementing data mesh or data fabric “exactly as they were originally described,” said Thomas. “Organizations want domain teams to move independently and access data more easily, but they aren’t willing to give up centralized governance, security, lineage, or quality controls. The challenge is finding a balance between autonomy and consistency.”
“Demand is real, but execution is falling behind,” said Michael Cucchi, chief marketing officer at Hydrolix. “Most organizations are building data architectures. Far fewer have the operational maturity to make those products reliable, reusable, or measurable, and that gap is where most initiatives stall.” Additional issues faced include hidden labor, complexity, and cost. “The broad platforms have gaps in their offerings,” he continued.
Use cases tend to be tied to initiatives such as “reliability, process optimization, digital twins, planning and scheduling, and enterprise reporting,” Adam Schindewolf, product marketing manager at Emerson’s Aspen Technology, observed. Data fabrics are likely seen in instances in which “data fragmentation limits decision-making speed or hinders high-priority digitalization initiatives like AI.” Significantly, Kumar noted, fabric is “becoming the default starting point for new builds.”
The core driver is the long-sought goal of achieving a unified view of an organization’s data—without the overhead. “A mesh or fabric does that without piling all the overhead onto the IT team,” said Justin Bolles, CTO at Resultant. “Data mesh and fabric push decision making closer to the people who understand and live that data every day, rather than routing everything through a centralized IT function that may not grasp the intricacies.”
CONVERGING MESH AND FABRIC
While there are distinctions between the two categories—data mesh enables decentralization of data, with distributed accountability and ownership, while data fabric enables data discovery via a technical layer—they can work together successfully.
“For years, the industry treated these as competing philosophies, as if you had to pick a side,” said Kumar. “That’s no longer the conversation. Mesh and fabric are being layered together: Mesh defines the ownership model; fabric provides the plumbing that makes that ownership model technically workable.”
Still, there are important distinctions. Data mesh rests upon “a foundational distributed data platform that makes locally collected data globally available,” explained Jared Pane, senior director of field engineering at Elastic. “And there’s the formalized data mesh concept of purpose-built data products that include domain ownership and governance.”
While data fabric is often conjoined with mesh, it brings capabilities not available through fabric, said Pane. “A data mesh architecture addresses the importance of where and how the data will be used. Unified data access is the whole point: everything potentially relevant, available together, in one call.”
AI, MESH, AND FABRIC
AI has become one of the biggest drivers behind the re-emergence of data mesh and fabric as a key initiative. “A model is only as good as the data it can trust and discover, and most enterprises don’t have governed, discoverable data,” Kumar pointed out. “LLMs [large language models] and AI agents need data that’s trustworthy, semantically understood, and governed, and most enterprises simply don’t have that today. That single requirement is doing more to push fabric adoption than a decade of data warehouse modernization arguments ever did.”
The greatest AI challenge for enterprises “isn’t choosing a model or securing more compute capacity,” said Thomas. “The real bottleneck was their data. If data isn’t easy to find, trusted by users, or managed consistently, even the most advanced AI projects struggle to deliver value.”
If anything, mesh and fabric have become the determinant as to whether AI initiatives succeed or stall in production.
“When agents need to query across domains in real time, problems with ownership, freshness, and access control stop being architectural concerns and start being operational fires,” said Alex Safronov, head of growth at Skyvia.
Along those lines, the emerging trend to watch at this juncture is how data mesh and model context protocol (MCP) are converging and their impact on the efficacy of AI agents, Safronov added. “Domain data products become directly consumable by AI agents with governance actually baked in, not bolted on. Without that in mind, you will find [agents] redesigning sooner than you planned.”