Snowpark Container Services

Snowpark Container Services (SPCS) is Snowflake's general-purpose container platform. You package any application as an OCI-compliant Docker image, whether a Next.js frontend, a Python API, a background job, or a GPU inference service, and run it on Snowflake-managed compute inside your Snowflake account, next to the data it uses.

It is built from three pieces: compute pools, clusters of CPU, high-memory, or GPU nodes billed separately from virtual warehouses; an image registry built into the account; and services and jobs, the containers themselves. Services can expose public endpoints secured by Snowflake authentication, call each other over internal networking, and read Snowflake data directly with the account's roles and policies, with no separate connection string or network hop.

That makes SPCS useful for AI and data applications: hosting open-source models on GPUs, running custom inference or retrieval services, serving full-stack apps over governed data, and packaging apps for the Snowflake Marketplace as Snowflake Native Apps. It also runs Streamlit apps when the container runtime is needed.

It only makes sense once Snowflake is already the data platform. Billing is in Snowflake credits for the compute pool nodes, per second after a short minimum, and pools keep billing while idle unless they auto-suspend. It has a smaller ecosystem and less tooling than a dedicated container host such as Cloud Run or Kubernetes.