diff --git a/docs/archive-copies.rst b/docs/archive-copies.rst
index 271019e27..04f7e5ec3 100644
--- a/docs/archive-copies.rst
+++ b/docs/archive-copies.rst
@@ -1,11 +1,45 @@
 .. _archive-copies:
 
-Software Heritage archive copies
-================================
+Archive copies
+==============
 
 .. _swh-storage-copies-layout:
 .. figure:: images/swh-archive-copies.svg
    :width: 1024px
    :align: center
 
    Layout of Software Heritage archive copies (click to zoom).
+
+The Software Heritage archive exists in several copies, to minimize the risk of
+losing archived source code artifacts. The layout of existing copies, their
+relationships, as well as their geographical and administrative domains are
+shown in the layout diagram above.
+
+We recall that the archive is conceptually organized as a graph, and
+specifically a Merkle DAG, see :ref:`data-model` for more information.
+
+Ingested source code artifacts land directly on the **primary copy**, which is
+updated live and also used as reference for deduplication purposes. There,
+different parts of the Merkle DAG as stored using different backend
+technologies. The leaves of the graph, i.e., *content objects* (or "blobs"),
+are stored in a key-value object storage, using their SHA1 identifiers as keys
+(see :ref:`persistent-identifiers`). SHA1 collision avoidance is enforced by
+the :mod:`swh.storage` module. The *rest of the graph* is stored in a Postgres
+database (see :ref:`sql-storage`).
+
+At the time of writing, the primary object storage contains about 5 billion
+blobs with a median size of 3 KB---yes, that is *a lot of very small
+files*---for a total compressed size of about 200 TB. The Postgres database
+takes about 8 TB, half of which required by indexes. In terms of graph metrics,
+the Merkle DAG has about 10 B nodes and 100 B edges.
+
+The **secondary copy** is hosted on Microsoft Azure cloud, using its native
+blob storage for the object storage and a large virtual machine to run a
+Postgres instance there. The database is kept up-to-date w.r.t. the primary
+copy using Postgres WAL replication. The object storage is kept up-to-date
+using :mod:`swh.archiver`.
+
+Archive copies (as opposed to archive mirrors) are operated by the Software
+Heritage Team at Inria. The primary archived copy is geographically located at
+Rocquencourt, France; the secondary copy hosted in the Europe West region of
+the Azure cloud.
diff --git a/docs/sql-storage.rst b/docs/sql-storage.rst
index 86a797374..aa3d834a3 100644
--- a/docs/sql-storage.rst
+++ b/docs/sql-storage.rst
@@ -1,14 +1,14 @@
 .. _sql-storage:
 
-Software Heritage SQL storage
-=============================
+SQL storage
+===========
 
 Postgres DB schema
 ------------------
 
 .. _swh-storage-db-schema:
 .. figure:: ../sql/doc/sql/db-schema.svg
    :width: 1024px
    :align: center
 
    Postgres DB schema of high-level Software Heritage storage (click to zoom).