The architectural principle that applications and processing power tend to move closer to where large data sets reside.
Compute Gravity is an architectural concept: as datasets grow, moving the data to the compute becomes too slow and expensive, so the compute moves to the data instead.
Moving large volumes of unstructured data from a secure corporate data lake to a cloud provider's AI inference engine is slow and costly, and it adds security risk.
Compute Gravity pushes organizations toward on-premises AI deployments that place inference GPUs next to the secured data storage. That arrangement reduces latency and keeps the data under the organization's own jurisdiction and control.
See also Data Lake, Inference Engine, On-Premises, Data Sovereignty, Large Language Model and Fine-Tuning.