Share cloud data without copying it blindly

Cloud data sharing asks two different questions: how another party can access a dataset, and what that party is allowed to do with it. Technology can support the first question. Owners, agreements, and controls must answer the second.

This article began as a November 2019 market overview after Azure Data Share reached general availability. It is retained as a dated reference because all three products and their commercial terms may have changed.

Define the exchange before selecting a service

A useful design starts with the relationship between the provider and recipient:

• Is the recipient receiving a copy or reading data in place?
• Is access continuous, scheduled, or a one-time snapshot?
• Who pays for storage, movement, and compute?
• Which fields and time periods are in scope?
• Can access be revoked, and what happens to prior copies?
• How will both parties record use, changes, and incidents?

These decisions shape the architecture and the agreement. “Secure sharing” is too broad unless the page says which risk and control it means.

The services named in the 2019 review

The original comparison named:

• Azure Data Share
• Snowflake Secure Data Sharing
• AWS Data Exchange

Those links point to current vendor documentation, not the 2019 feature set. Review supported sources, regions, identity requirements, pricing, refresh behaviour, and recipient controls before comparing them.

Copying and in-place access solve different problems

A snapshot gives the recipient an independent copy. It can support isolation and repeatable analysis, but it also creates another dataset to secure, retain, refresh, and eventually dispose of.

In-place access can reduce duplication, but it still needs identity, authorization, audit, performance, and availability decisions. Revoking future access does not automatically remove results or exports already created by the recipient.

Make the governance record inspectable

Before release, record the approved purpose, data fields, recipients, duration, controls, and owner. Add a process for changes and incidents. If personal or regulated data is involved, obtain the relevant privacy, legal, and security review.

That record should travel with the data-sharing design. A platform feature can enforce part of the agreement, but it cannot replace it.