Glossary

Data Gravity

Data gravity describes the tendency for applications, services and computing resources to move closer to large concentrations of data. As datasets grow, moving the data itself can become slower, more expensive and operationally difficult, so organizations often bring compute and analytics to the data instead.

The concept is especially relevant to AI, analytics, cloud infrastructure and large-scale storage. A relatively small dataset may be easy to copy between environments, but a multi-petabyte repository can influence where applications run, which clouds or data centers an organization uses, and how its infrastructure is designed.

How does data gravity work?

Data gravity develops as data accumulates in a particular storage environment. Applications begin consuming that data, users and services depend on those applications, and additional datasets may be created as a result. Over time, these dependencies make the original data repository increasingly difficult to relocate.

Several factors strengthen data gravity:

• Dataset size: Larger datasets require more time and bandwidth to transfer.
• Application dependencies: Applications may depend on specific datasets, APIs, metadata or storage interfaces.
• Data creation rates: AI, analytics, backup and machine-generated workloads can continuously add new data.
• Network constraints: Available bandwidth can limit how quickly large repositories can move.
• Cloud economics: Data transfer and egress charges can make frequent movement between environments expensive.
• Governance requirements: Data residency, sovereignty, retention and security policies may restrict where information can be stored or processed.

Data gravity therefore isn't simply a question of storage capacity. It reflects the growing technical, economic and operational dependencies surrounding a dataset.

Why does data gravity matter?

Data gravity can shape infrastructure decisions long before an organization reaches exabyte scale.

For example, an enterprise may initially store application data in one cloud because deployment is convenient. As the repository grows, analytics applications, backup systems and AI services may all begin consuming the same information. Moving the original dataset to another environment now means accounting for every connected workload as well.

This can affect decisions about cloud migration, data center architecture, storage platforms and application placement. Infrastructure teams need to consider not only where data should reside today, but how easily it can be accessed or moved as requirements change.

Data gravity and AI

AI makes data gravity particularly important because AI systems can consume enormous volumes of unstructured data.

Training datasets, model checkpoints, embeddings, documents, images, video, logs and generated outputs can accumulate rapidly. Many of these datasets also have value beyond a single training run. They may be reused for model retraining, retrieval-augmented generation (RAG), inference, evaluation or future analytics.

Moving every dataset into high-performance GPU infrastructure is rarely practical at large scale. Instead, organizations can design architectures where different compute and storage resources interact according to workload requirements.

Frequently accessed training data might reside close to accelerated compute, while larger repositories remain on scalable object storage. Data can then move between performance tiers according to policy and usage rather than forcing the entire AI corpus onto the most expensive infrastructure.

This approach turns data placement into an architectural decision rather than treating storage as a passive destination.

Data gravity in hybrid and multi-cloud environments

Hybrid and multi-cloud architectures can make data gravity more visible because applications and datasets may span several locations.

An organization might operate data across:

• Private data centers
• Public clouds
• Colocation facilities
• Edge environments
• Sovereign cloud regions
• Multiple geographic locations

The challenge is deciding which data should move and which workloads should instead operate closer to that data.

Large datasets can make repeated transfers impractical, particularly when network capacity or cloud egress charges are significant. Organizations may therefore maintain a durable data repository in one environment while selectively moving subsets of data to applications or compute resources elsewhere.

Open interfaces such as the S3 API can also help applications interact with object data across different infrastructure environments without requiring every workload to use proprietary storage interfaces.

Data gravity vs. data locality

Data gravity and data locality are related but different concepts.

Data gravity describes the tendency for workloads and services to accumulate around a large dataset because moving that dataset becomes increasingly difficult.

Data locality describes the deliberate placement of compute close to the data it needs to process.

In practice, data gravity can create the conditions that make data locality desirable. If transferring several petabytes to a remote compute environment would introduce excessive latency, cost or operational complexity, running the workload closer to the existing data may be more efficient.

Data gravity vs. data mobility

Data gravity explains why large datasets tend to remain where they are. Data mobility describes the ability to move or access those datasets across systems and environments when necessary.

The two concepts can work against each other. As data gravity increases, maintaining data mobility generally becomes more challenging.

Organizations can reduce this friction by using portable interfaces, scalable storage architectures, policy-driven lifecycle management and infrastructure that avoids unnecessary proprietary dependencies. The objective is not necessarily to eliminate data gravity, but to preserve choices despite it.

Can data gravity cause vendor lock-in?

Data gravity can contribute to vendor lock-in, but the two are not identical.

A dataset may be difficult to move simply because it is extremely large. Vendor lock-in adds another layer when proprietary APIs, data formats, services or high transfer costs make migration harder.

This distinction matters. An organization managing 50 PB of data will face a significant migration project regardless of platform. If the same repository also depends heavily on proprietary services, however, changing infrastructure may require substantial application redesign in addition to moving the data.

Using broadly adopted interfaces and maintaining control over data placement can reduce these dependencies.

How can organizations manage data gravity?

Organizations usually cannot eliminate data gravity as datasets grow. They can, however, design infrastructure so that it does not unnecessarily restrict future choices.

Useful approaches include designing around open data-access interfaces, placing compute according to workload requirements, maintaining high-bandwidth connectivity between critical environments and applying lifecycle policies that move data between performance and capacity tiers.

Organizations should also consider migration and portability before datasets become extremely large. Testing how data can be moved, replicated or accessed from another environment can expose dependencies while they are still manageable.

For AI infrastructure specifically, separating the durable data layer from individual compute environments can be valuable. GPU infrastructure can change quickly, while enterprise datasets may need to remain available for years or decades.

How does object storage help manage data gravity?

Object storage is commonly used for large-scale unstructured datasets because it can provide a persistent data layer that is accessible to many applications.

Rather than keeping separate copies of the same dataset for every workload, organizations can consolidate large repositories and allow applications to access the required objects. This is useful for AI, analytics, backup, archive and other workloads that generate or consume large quantities of unstructured information.

Scale-out object storage can also allow capacity to grow without forcing organizations to redesign the storage architecture each time the dataset expands.

For enterprises operating private or hybrid infrastructure, platforms such as Scality RING can provide an S3-compatible object storage layer for large-scale data repositories. Scality's ADI (Autonomous Data Infrastructure) strategy extends this approach around enterprise AI, cyber resilience and sovereign control, with the goal of maintaining access to data across its lifecycle rather than tying it permanently to a particular compute environment.

What is an example of data gravity?

Consider an enterprise that has accumulated 20 PB of video and sensor data over several years.

Initially, the data may support a single analytics application. The organization later begins using the same repository for AI model training, historical analysis and compliance retention.

Moving 20 PB every time a new AI environment is deployed would be inefficient. Instead, the organization can maintain the durable dataset on scalable storage and provide compute environments with access to the portions they require.

As the repository grows, the data effectively becomes an anchor for the surrounding infrastructure. That is data gravity in practice.

Why data gravity matters for enterprise infrastructure

Data gravity becomes more consequential as organizations retain larger datasets and use them for more purposes.

AI is accelerating this trend because the long-term value of enterprise data increasingly extends beyond the application that originally created it. Historical data may become training material, RAG content or input for models that do not yet exist.

Infrastructure strategies therefore need to account for both where data lives and how easily future workloads can reach it. Organizations that preserve portability, scalable access and control over data placement are better positioned to adopt new compute technologies without repeatedly relocating their largest datasets.

Frequently asked questions

What is data gravity in simple terms?

Data gravity is the tendency for applications and computing resources to move closer to large datasets because moving the data itself becomes increasingly difficult as it grows.

Who coined the term data gravity?

Software engineer Dave McCrory introduced the concept of data gravity in 2010, comparing the way applications and services are attracted to large datasets with the gravitational effect of physical mass.

Is data gravity a problem?

Not inherently. Data gravity is a natural consequence of accumulating large amounts of data. It becomes a problem when infrastructure prevents organizations from efficiently accessing, processing or relocating that data.

Does data gravity only apply to cloud computing?

No. Data gravity can occur in public clouds, private data centers, hybrid environments, edge infrastructure and any architecture where large datasets accumulate.

How does data gravity affect AI infrastructure?

AI workloads depend on large datasets for training, RAG, inference and analytics. As those datasets grow, it can become more practical to bring compute to the data or selectively move required data rather than repeatedly transferring the entire repository.

What is the difference between data gravity and data sovereignty?

Data gravity is primarily an architectural and economic phenomenon caused by large concentrations of data. Data sovereignty concerns the legal and regulatory requirements governing data based on the jurisdictions in which it is stored or processed.

Can data gravity be eliminated?

Usually not at large scale. Organizations can reduce its negative effects by maintaining data portability, using open interfaces, designing efficient data movement and avoiding unnecessary dependencies on individual infrastructure providers.