What is “Good Fabric”? — from Digital MATSUMOTO
Google Cloud has announced a new network fabric called Virgo Network. It is described as an infrastructure that organizes and weaves the flow of communication in an orderly manner so that massive clusters of TPUs can operate as a single computing environment.
The term “fabric” is used across various layers of architecture—but what does it really mean? Let’s think about it through the lens of a data fabric.
What is a “Good Fabric” when viewed through Virgo Network?
Virgo Network, announced by Google Cloud, is attracting attention as a large-scale network fabric for AI. It connects up to 134,000 TPUs within a single fabric, serving as a network foundation to treat a massive AI cluster as one computing environment.
The important point here is that this is not merely about “connecting a lot of devices.” It lies in the fact that it creates order in communication between distributed computing resources, including bandwidth, latency, routing, and isolation in case of failure. A fabric is easier to understand if you think of it not as forcing disparate elements into one place, but as an architecture that enables them to operate as a whole while remaining distributed.
This way of thinking applies directly to data platforms as well. A good data fabric architecture is not about gathering all data into a massive warehouse, but about creating a state where distributed data can be understood, used, and managed while remaining distributed.
Can collected data be properly understood?
In a data fabric, what matters first is not the data itself, but the metadata that supports its correct understanding. Where it resides, who created it, which business process it originated from, what each field means, and which reports or AI models use it. Without this information, even if data exists, it cannot promote correct usage.
Many data platforms start with “just collect everything.” But even if you gather large amounts of data whose meaning is unclear, users will only struggle later to interpret it. In a data fabric, the key is not just connecting where data resides, but also linking its meaning, quality, history, and usage context.
Putting lineage and quality at the center of operations
Data does not end once it is created. When business processes change, the meaning of data also changes. System updates subtly alter field definitions. If input rules break down in practice, data with the same name can become entirely different in substance.
That’s why, in a good data fabric, lineage and quality management are not decorations but the core. Being able to trace where data came from, how it was processed, and where it was used. Being able to understand which downstream uses are affected by anomalies, missing values, or definition changes. If this is weak, data utilization quickly becomes untrustworthy.
The scariest thing in a data platform is not the absence of data, but continuing to use incorrect data as if it were correct.
Integration is not centralization, but making things controllable
The term “data fabric” often brings to mind a clean, centralized platform, but that feels slightly off. In real organizations, data is scattered across on-premises systems, the cloud, SaaS, and departmental systems. Bringing everything into one place may sound ideal, but in practice it often becomes heavy and impractical.
What matters is not physically gathering everything in one place, but being able to handle it logically while it remains distributed. Being able to access necessary data with appropriate permissions, along with proper quality information. The foundation is not to replicate data unnecessarily, but to manage it seamlessly while it remains distributed.
A fabric is not a philosophy of “connecting everything.” It connects what is necessary, reliably, and does not connect what is unnecessary. That judgment must be embedded within the architecture.
Do not bolt on permissions and policies afterward
In a data fabric, adding security and governance at the end often leads to failure. This is because data gets copied, transformed, and brought into different contexts as it is used. Even if you protect the original storage location, control breaks down during usage.
Therefore, rules such as permissions, purpose of use, masking, retention periods, and whether external sharing is allowed should be provided as common modules as much as possible, and implemented where needed within data pipelines. What matters here is being able to detect risky usage and apply necessary controls while maintaining the speed at which teams can actually use the data.
Governance is not a brake, but rather an image of maintaining the road so you can step on the accelerator with confidence.
Creating a weave that does not collapse under change
The value of a good data fabric becomes visible not in normal times, but when change occurs. A new SaaS is introduced. Data from another company increases due to M&A. Core systems are renewed. AI agents begin referencing internal data. Whether the architecture can smoothly adapt to such business and technological changes reveals whether it is truly good.
That’s why a data fabric needs conventions for extension. Of course, there must be technical procedures for how to register metadata, quality, permissions, lineage, and usage logs when adding new data sources—but also a process for who decides on architectural changes. That architect might not even be human; it could be AI.
Summary
With the spread of cloud services, every layer connects distributed resources and quickly scales into something massive. In that context, it is not enough for an architecture to simply be “connected”—it must create order in meaning and usage.
The essence of a good data fabric is not collecting data, but making it understandable, traceable, controllable, and resilient to change. More than a clean diagram, what matters is whether, in actual operations, it forms a weave that does not lose the correct understanding of data. That, I believe, is the most important point.
Digital MATSUMOTO
https://medium.com/@digitalmatsumoto/what-is-good-fabric-85ecdf2e3d2a