Databricks · One-Pager · Data Sharing

Delta Sharing & Marketplace

An open protocol to share live data and AI assets across organizations and clouds — no copying, no vendor lock-in — plus an open exchange built on top of it.

GA No data copy · live access Verified June 2026 docs

In one line

Delta Sharing is an open protocol developed by Databricks for secure data sharing with other organizations, regardless of the compute platform they use — and Databricks Marketplace is the open exchange of data products built on it.

Why it matters

It replaces fragile, lock-in-prone sharing (SFTP dumps, cloud-specific bucket grants, per-vendor APIs) with one governed protocol: recipients read live data in place from the provider's storage with no replication, and the recipient can be on any platform — Databricks, Spark, pandas, or Power BI.

Core concepts

Open protocol

A REST protocol where recipients fetch data via short-lived pre-signed URLs; works with any client, decoupled from the provider's compute.

Share / Provider / Recipient

A share is a read-only collection of tables, views, volumes, models or notebooks a provider exposes to a named recipient.

D2D vs open sharing

Databricks-to-Databricks uses a UC sharing identifier (no tokens); open sharing hands non-Databricks recipients a bearer token / OIDC credential.

No data copy

Recipients query live data at the source — no replication or ETL — so there is one copy and one source of truth.

Unity Catalog integration

Shares, providers and recipients are governed UC objects; GRANT controls access, lineage and audit apply (UC not strictly required to share or consume).

Marketplace

An open forum, built on Delta Sharing, where providers list datasets, notebooks, ML models and MCP servers for consumers to discover and request.

How it works

Click a stage

1 · Provider share 2 · Grant recipient 3 · Sharing protocol 4 · Recipient reads

Provider share — the provider creates a share and adds UC tables, views, volumes, models or notebooks to it.

Minimal example

SQL · provider side (Unity Catalog)CREATE SHARE sales_share COMMENT "Q2 sales feed";

-- expose a UC table into the share (read-only, live)
ALTER SHARE sales_share ADD TABLE main.sales.orders;

-- D2D recipient: bind to their Unity Catalog sharing identifier
CREATE RECIPIENT acme USING ID 'aws:us-east-1:0c97...ed1ef';

-- SELECT is the only grant a recipient can hold on a share
GRANT SELECT ON SHARE sales_share TO RECIPIENT acme;

Gotchas & limits

SELECT is the only privilege grantable to a recipient — shares are strictly read-only; you cannot grant write.
Provider needs CREATE SHARE on the metastore and must keep SELECT on any shared table, or sharing silently breaks.
D2D sharing needs both sides on Unity Catalog; open sharing uses bearer tokens / OIDC — rotate and scope them, and prefer short-lived recipient tokens.
Sharing tables that use deletion vectors, column mapping, or some advanced Delta features can carry Preview / version caveats — check recipient compatibility.

Interview soundbites

Say this out loud

"Delta Sharing is an open REST protocol — the recipient reads live data in place from the provider's storage, so there's no copy, no ETL, and no vendor lock-in."
"Databricks-to-Databricks sharing uses a Unity Catalog sharing identifier with no tokens; open sharing hands non-Databricks clients a bearer token so they can read with Spark, pandas, or Power BI."
"Marketplace is the open exchange built on Delta Sharing — providers list datasets, notebooks, AI models and MCP servers; consumers discover and request access from their workspace."

References

Verified against docs.databricks.com · June 2026